Run that through a customer that held the same number of seats in July as it did in June. The vendor books expansion. Net revenue retention goes up. Nobody at the customer used more of anything.

I build software for private equity, and the first file I ask for in a software data room is the renewal ledger, not the retention slide. I read that Microsoft page the same way. It is the cleanest public example I know of a number every software deal underwrites as one thing when it is two.

The thesis

Expansion revenue is two products. Volume led expansion, more seats, more usage, more modules adopted, compounds with the customer’s business. Price led expansion, the same seats at a higher unit price, has a ceiling and a procurement team with a memory. Every diligence pack I have read reports them as one line inside net revenue retention. They deserve different multiples, and in 2026 they are moving in opposite directions.

The mechanism

Net revenue retention is a ratio. What a cohort of customers pays you now, divided by what the same cohort paid a year ago. Churn and contraction pull it down. Expansion pushes it up. The figure says nothing about why the expansion happened.

Split a single renewal and every dollar comes from somewhere specific. Either the customer is paying for more units, or it is paying more for each unit. The first is a claim on the customer’s growth. The second is a claim on the customer’s patience.

Volume compounds because it rides on something outside the vendor: the customer hiring, processing more transactions, rolling the product into another division. Price does not compound that way. A 6% uplift is a one off step, and the next one has to be negotiated again against a buyer who now has the last invoice on file. That memory shows up at renewal as a discount request, a competitive tender, or a quiet seat reduction to hold the total flat.

AI is what makes this urgent. Seat priced software has always been priced on the customer’s headcount, and AI is going after headcount first in the roles that hold the most seats: support agents, frontline staff, junior operations. The volume engine in a seat priced business can now run backwards while the price engine runs forwards. Net revenue retention nets the two into one comfortable number.

Three cases on the public record

Microsoft. E3 up 8.3%, F1 up 33%, both from 1 July 2026. A customer holding seats flat produces 8.3% expansion on E3 alone. The largest percentage rise lands on the cheapest and most numerous seat, the one issued to people on a shop floor or a ward. Any business whose revenue tracks that seat base will show its best renewal cohort in years, and almost none of the improvement will be volume.

Salesforce. On 17 June 2025 Salesforce announced an average 6% list price rise on Enterprise and Unlimited editions of Sales Cloud, Service Cloud, Field Service and some industry clouds, effective 1 August, citing “significant ongoing innovation and customer value”. By early September Marc Benioff had told The Logan Bartlett Show he had cut his own support organisation “from 9,000 heads to about 5,000, because I need less heads.”

Put those two numbers on one account. A customer that makes the same cut on a per agent support product keeps 56% of its seats at 106% of the price. That account now pays 59% of what it paid before. Price can paper over a small seat decline. It cannot cover a 44% one, and the support desk is where AI is landing first.

Workday. Workday is the honest version. CFO Zane Rowe reported gross revenue retention of 97% in November 2025 and again in August 2026, and said in August that net expansion from existing customers drove about 60% of subscription revenue growth. In November an analyst asked how Workday grows if its customers lay people off. Rowe said customer headcount growth was a net positive, “but it’s not a significant amount of our growth.” Carl Eschenbach, then chief executive, said the company was “focused not on just seats, but actually revenue per seat”, and that contracts carry “floors and minimums” on how far a customer can bring headcount down.

Read those together. The expansion is not coming from more employees, so it is coming from modules, agents or price per head, and the floor exists because the vendor expects the head count may fall. That is a listed company telling you on a public call which of the two products it is selling. Most private targets will not tell you unless you ask for the rows.

Where this breaks

Three objections are fair.

First, pricing power is a moat signal. A vendor that raises prices 8% and keeps 97% of its revenue has a product customers cannot easily leave. Agreed. Price led expansion that survives three renewal cycles with gross retention intact is evidence, and I would pay for it. What I will not pay for is one cycle of it extrapolated across a five year hold.

Second, some price rises carry real new value. Microsoft’s announcement pointed to added AI and security features, and Salesforce cited innovation. Where the ledger can show that as a new module, it is volume and should be booked as volume. Where it cannot be separated, I treat it as price, because the buyer will.

Third, volume is not pure either. Seat growth tied to customer hiring is a bet on employment in your customer base, and in a downturn it reverses faster than price does. True, and that is the reason to split them. They fail in different conditions, so anyone holding a book of these businesses needs to know the mix in each.

What I would do on Monday

1. Ask for the renewal ledger at contract line level: units and unit price for every customer at each of the last three renewals. Not the retention slide. The rows.

2. Split every expansion dollar into new units at the old price and the price change on all units. Report a volume NRR and a price NRR beside the blended one.

3. Date every list price event and check gross retention and discount depth in the 12 months after each. That is where procurement memory shows up.

4. Map the seat base by role. Any share sitting in support, frontline or junior operations gets a volume haircut in the model, because those are the seats AI reaches first.

5. Value price led expansion as a step, not a rate. Hold it flat after the next renewal unless the ledger proves it repeats.

The two columns

Microsoft’s licensing page has two columns and no explanation. That is fine for a price list. It is not fine for a retention number, and most retention numbers I am handed are built the same way: a before, an after, and nothing about why it moved.

The why is in the rows. Ask for them.

Diligence on a software asset usually comes down to a renewal ledger nobody has split yet. If one is open on your desk, find an hour with me.

That is the promise inside almost every buyout model signed this year. Bain’s 2026 Global Private Equity Report, published on 23 February, put a number on the new regime: a deal now needs 10 to 12 percent EBITDA growth a year to earn the benchmark 2.5 times return over five years, against about 5 percent through the 2010s. Cheap debt and multiple expansion stopped doing the work, so the operating plan has to. Seventy six percent more earnings in five years, from the same company, run by mostly the same people.

Now halve it.

The thesis

A value creation plan is a bet on how much change a company can absorb, and diligence never measures that. It measures the market, the margins, the contracts and the code. It does not measure the rate at which this organisation has historically turned a decision into a finished change.

So sponsors underwrite a plan that needs three years of change delivered in eighteen months, launch all of it at once to beat the clock, and replace the chief executive when it does not land. The industry calls that an execution problem. I call it an underwriting error. The capacity was never there, and nobody checked.

Halving it

Alvarez & Marsal published its European private equity value creation report on 20 May 2026. Two hundred fund investors and C-level executives across ten countries, interviewed by Statista Q in February. Two findings sit side by side in the release, and I do not think they were meant to be read together.

First, 58 percent of sponsors now deploy resources within the first 100 days, double the 29 percent recorded a year earlier. Second, 65 percent say they have achieved less than half of the value targeted in plans developed over the past two years.

The response to plans that under deliver is to start them sooner. Starting sooner means more initiatives running at the same time. More initiatives running at the same time is the one thing an organisation with finite change capacity cannot carry.

Now run the arithmetic on “less than half”. Take a plan written to the new regime: 12 percent a year, 76 percent over five years. Capture half the value and you land 38 percent. Compounded, that is 6.7 percent a year.

A half captured plan in the 12 percent world delivers something much closer to the 5 percent world. The sponsor paid a 2026 entry price for 2015 performance, and the model will not show it until year three.

Where the capacity actually sits

Change capacity is not headcount. It is the small number of people who run the business today and also own a piece of the change. In most mid market companies I have worked inside, that is the chief executive, the finance lead, one or two operations heads and whoever runs technology. Five or six people.

Every initiative in the plan lands on them. A pricing reset lands on finance and sales. A procurement programme lands on operations and finance. A systems consolidation lands on technology and finance. An AI programme lands on everyone, and on technology twice. Notice which name keeps appearing.

The people underneath them are already full. Gartner’s HR research found the average employee experienced 10 planned enterprise changes in 2022, up from 2 in 2016, while willingness to support change fell from 74 percent to 43 percent. Its April 2025 survey of more than 2,850 employees found 79 percent have low trust in change, and only 32 percent of business leaders report healthy change adoption. A new owner does not arrive at a rested organisation. It arrives at one that has already been changed ten times this year and has stopped listening.

Then the owner column fails, and the sponsor does the one thing it can do quickly. AlixPartners’ 11th annual private equity leadership survey, published 25 March 2026, covered 427 executives. 65 percent of PE firms report chief executive turnover during the hold. 83 percent say unplanned turnover lengthens the hold, and nearly half say it reduces returns. Bain has buyout holding periods at around seven years at exit, against five to six from 2010 to 2021.

Replacing the chief executive is a capacity fix applied after the fact, at the most expensive moment available, and it resets the owner column to zero.

How it compounds

The failure runs in a straight line. The plan is written to the 12 percent requirement. Diligence never asks whether the company can carry it, so every initiative is launched inside the first 100 days. The same five or six people end up owning all of it. Less than half the value lands. The chief executive goes, the hold stretches, and the new leadership inherits a half finished portfolio of change with a shorter clock.

The alternative path is not slower. It is ordered. Test the capacity before close, sequence the initiatives to fit it, and the value lands inside the hold instead of in the next owner’s model.

What I actually look at

On 1 September I argued here that technical diligence audits the codebase and ignores the delivery system. This is the same leak, one level up. The delivery system is where a company’s change capacity becomes visible, because almost every initiative in a modern value creation plan eventually becomes a change to a system that someone has to build, configure or migrate.

So when I read a value creation plan now, I skip the initiative column and read the owner column. I count distinct names. Then I ask the company one question: in the last 24 months, how many cross functional changes did you start, and how many did you finish? The ratio between those two numbers is the most honest estimate of change capacity I know. I have not seen it in a standard diligence pack.

After 30 years of shipping software, the pattern I trust least is a plan where the finance lead’s name appears against more initiatives than there are quarters in the first year.

Where this breaks

The Alvarez & Marsal respondents blame the world first, not their plans. 62 percent name geopolitical volatility as the top obstacle to value creation and 58 percent name tariffs and inflation. Some of that “less than half” is macro, not capacity, and a self-reported survey of 200 European respondents cannot separate the two. I am reading a pair of numbers in a way the authors did not.

The Gartner figures measure attitude, not throughput. Low trust in change does not prove an organisation cannot deliver it. Some deliver anyway.

The clock is real. With seven year holds and a 12 percent requirement, a sponsor who sequences may lose more to time than it saves in capture. My answer is that sequencing is ordering, not waiting, but I concede the trade is tighter than this post makes it sound.

Some sponsors do test for this. Large operating partner benches and management assessments exist for a reason. The limit is that a management assessment grades the individual, not the throughput of the system around them. A strong chief executive with a full owner column is still full.

And the strongest objection: sometimes the chief executive change is the right capacity fix and the plan was fine. Nearly half of AlixPartners’ respondents say unplanned turnover reduces returns. The other half do not.

What I would do on Monday

1. Add one line to diligence: cross functional changes started versus finished over 24 months. Ask for the list, not the number.

2. Read the owner column. If one name owns more than three initiatives, the plan is a wish list.

3. Launch no more initiatives in the first 100 days than the company finished in its best six months. Put the rest in a dated queue.

4. Underwrite at the capture rate, not the target. If the evidence says half, model half, and let the price move.

5. Hire the second line before close. The finance and technology owners you will otherwise hire in year two, after the chief executive has gone.

1.12 to the fifth is 1.76. Half of it is the 2010s.

The gap between those two numbers is not strategy. It is how many things one finance lead can finish in a year. Nobody asked.

I build software for private equity, which means I spend a lot of time inside the systems a value creation plan has to run through. If you are underwriting one this quarter and nobody has counted the owner column, book a call.

Seven more emails. Twenty five more meetings.

Both groups did roughly the same amount of outreach. One group’s messages reached a human being and the other group’s did not. The thing that separated them is not effort and it is not deck quality. It is the warm introduction, and nobody in venture has ever put a price on it.

The thesis

The warm intro is not a courtesy. It is a sorting mechanism, and it sorts on affinity rather than on quality. Founders pay for it in meetings they never get. Investors pay for it in returns, and the second half of that sentence is the part the industry has never accepted.

I have a stake in this. I co-run Capitaly.vc with Houman, which exists because we thought this market clears badly. Read the rest knowing that.

What the sourcing week actually looks like

Paul Gompers, Will Gornall, Steven Kaplan and Ilya Strebulaev surveyed 885 institutional venture capitalists at 681 firms. It remains the largest structured look inside the job.

Their reported week runs 55 hours. Of that, 22 hours go to networking and sourcing and 18 go to working with portfolio companies. Sourcing is the single largest block of the week.

Where the deals come from is more revealing. Over 30% are generated through professional networks. Another 20% are referred by other investors. 8% come from a portfolio company. That is 58% of deal flow arriving through a relationship. Only 10% comes inbound from company management. Almost 30% is proactively self generated.

The funnel underneath: roughly 100 opportunities considered for every investment closed. One in four leads to a management meeting. One third of those reach a partners meeting. About half of those proceed to diligence. Firms offer 1.7 term sheets for each deal they close, and the median firm closes about four investments a year.

Now the number that should be uncomfortable. Asked which of the three things they do contributes most, 49% of these investors said deal selection, 27% said post investment value add, and 23% said deal flow. Forty percent of the working week goes into the activity they themselves rank last of three.

That is only irrational if you believe the network is a sourcing tool. It is not. It is a screening tool wearing a sourcing costume. An introduction carries something a cold deck cannot carry: a named person spending their own reputation on your behalf. It performs diligence before diligence, at zero marginal cost to the investor, using a credential the founder did not earn and cannot buy.

That is why it survives. It is genuinely useful to the person holding the cheque. The cost sits with people who are not in the room.

What it costs the people collecting it

The affinity part of this has been measured, and it was not measured on founders. It was measured on investors introducing each other.

Gompers, Vladimir Mukharlyamov and Yuhai Xuan studied 3,510 individual venture capitalists across 11,895 portfolio companies from 1975 to 2003. They asked a simple question: when two investors share a background, are they more likely to work together, and does the deal do better.

The answer to the first half is yes, strongly. Two VCs who worked at the same firm previously are 64% more likely to co invest. Two who went to the same undergraduate school are 42.5% more likely. Two from the same ethnic minority group are 22.8% more likely.

The answer to the second half is the interesting one. Success, defined narrowly as the portfolio company reaching an IPO, falls. Same previous employer costs 18 percentage points. Same undergraduate school costs 22 points. Same ethnic minority group costs 25 points.

The split inside that paper is the whole argument. Similarity based on ability, such as both partners holding degrees from top universities, improves outcomes by 9 to 11%. Similarity based on affinity destroys them. The network sorts on the second kind and pays for it in the first.

PitchBook surveyed 391 venture investors across the US, Europe and Asia and found 82% naming personal networks as their most valuable sourcing resource, against 44% for inbound and 36% for financial databases. Only 38% used data to source all of their opportunities. That survey ran in Winter 2018 and I would expect the data number to be higher now. I would not expect the 82% to have moved much, because the underlying incentive has not moved at all.

I built D30 for the listed version of this problem. It pulls three statement fact files out of ASX annual reports and runs articulation gates over them, so a company reaches a shortlist because its numbers tie, not because somebody vouched for it. Private markets have no equivalent instrument. They have a phone.

Australia makes the geometry worse

Cut Through Venture reported $5.1bn across 390 deals for Australian startups in 2025, published 3 February 2026. That is a 24% lift on the prior year and the third largest funding year on record. It is also 390 deals. A market that size has one degree of separation inside it, not six.

In the same data, female founders took 24% of deals, down from 28% in 2024. Series A is worse. That is what an affinity sort looks like when it runs at national scale for a decade: not a policy, not a conspiracy, just the accumulated output of a filter that rewards proximity to people who already have capital.

66% of 2025 deals included an international funder. Australian founders are increasingly raising from people who are structurally outside the local network. That is the tell. When the local sort is thin, founders route around it, and the routing costs them 25 meetings.

Where this breaks

Three places, and I want to be precise about them.

The Cost of Friendship measures co investment between investors, not introductions to founders. It also ends in 2003. I am transferring a mechanism, not a coefficient. Anyone who quotes 18 percentage points as the cost of a warm intro to a founder is misusing the paper, including me if I did it.

The intro is a real signal, not only an affinity artefact. A person who introduces you has read enough of your business to risk being wrong in front of someone whose opinion they need next quarter. That is information. A filter that is 60% accurate and free beats no filter when you are reading 100 opportunities to make four investments a year.

And the counterfactual is not open access, it is a different filter. Remove intros and the volume does not vanish, it lands somewhere. Every investor who has opened submissions has closed them again within two years. The honest version of my argument is not that the intro should go. It is that it should be one input among several, and that right now it is the gate.

What I would do on Monday

If you are raising:

1. Count your ratio, not your outreach. Contacts to first meetings is the only number that tells you whether you have an access problem or a business problem. Below 30% and the deck is not the issue.

2. Build the intro asset before you need it. Operators inside the portfolio companies of your target funds are the 8% referral channel, and they answer.

3. Publish the thing only you know. Proactive sourcing is 30% of deal flow. Being findable by a partner running a thesis search is the one channel that does not require permission.

If you are investing:

1. Instrument your own funnel by source. Split closed deals by professional network, other investors, portfolio referral, inbound and self generated, then look at outcomes by source three years later. Most firms have never run this and it takes an afternoon.

2. Cap the affinity channel deliberately. Not on principle. Because the published evidence says the correlation between how a deal reached you and how well it does is negative in exactly the channel you use most.

Close

77 investors and 40 meetings. 70 investors and 15 meetings.

The second group did the work. They just did it into a system that was never reading.

Capitaly connects founders and investors without the warm intro tax. See it here.

Those two numbers describe the same 62 engineers. The capacity can only be spent once.

The thesis

Headcount is not a result. It is an input you happen to control, which is exactly why it gets managed as if it were an outcome. AI frees one pool of capacity, and you convert that pool into a smaller payroll or into more change shipped. Both are defensible. Doing both, in the same year, on the same pool, is a double count, and it is the most common accounting error in enterprise AI right now.

The only figure that exposes it is cost per shipped unit of change. Almost nobody computes it.

What the metric actually is

Take everything it costs to move change into production over a quarter. Fully loaded engineering payroll, contractors, tooling, cloud, and the AI line: seats, tokens, and the API keys somebody expensed. That is the numerator. Then count the units of change that reached production and were still there thirty days later. That is the denominator. Divide.

The numerator is the easy half and most finance teams already have it, scattered. The AI component is now large enough to matter. Gartner reported in June 2026 that nearly a quarter of technology leaders spend between $200 and $500 per developer per month on AI coding tokens alone, and about 6% spend more than $2,000. Jellyfish published a worked example of a 40 developer organisation: $1,560 a month in licences, $2,100 in overages, $480 in extra tools, $1,400 on raw API keys and $300 in personal subscriptions that went through on cards. Total $5,840 a month, or $146 a developer. Inside that same company the platform team ran at $556 per developer and the frontend team at $61. That is a nine times spread, and nobody who has not built a denominator can tell you which of those two numbers is the good one.

The denominator is where the work is, because you have to decide what a unit is and then defend it. A merged pull request is too small and too gameable. A deployment is closer. A change that a customer or an internal user could describe is better.

Look at what happens when the unit is chosen badly. DORA’s own model books its value as time saved, 12.5% per developer, then reports the delivery outcome as deployments rising from 50 to 56 a year against a first year investment of $8.4 million. That is $1.4 million per additional deployment. The model is not wrong. It is measuring the return in one unit and the outcome in another, and the arithmetic only looks absurd because I forced the two units into the same fraction. That is the whole point. When the unit you book value in differs from the unit you measure delivery in, a double count can sit between them for a year and never show up on a slide.

Run the numbers on your own team

DX studied more than 400 organisations and found a median throughput gain of 7.76% from AI assisted development, a mean of 13.1%, and 43.9% at the ninetieth percentile. Take the median and a hundred engineers. Assume a fully loaded cost of $200,000 each, and adjust that to your own market. The AI spend at $350 a developer a month is $420,000 a year. The capacity gain is 7.76 engineers, worth $1.55 million. Close to four to one.

That ratio holds if you take the capacity as output. It also holds if you take it as payroll: cut 7.76 roles, save $1.55 million, ship the same volume, and the cost per shipped change still falls. Both paths work. The failure mode is the third one, where the payroll saving is banked in the January budget and the output increase is promised in the same board pack. Two quarters later the volume is down, because the 62 engineers you modelled were the same 62 you let go, and the ratio is worse than before you started.

Gartner’s finding is the tell. Workforce reduction rates were close to identical between the organisations reporting strong returns and the ones reporting nothing or worse. Helen Poitevin put it plainly: workforce reductions may create budget room, but they do not create return. If cutting correlated with returns you would see it in 350 companies with a billion dollars of revenue each. You do not.

What I count

I run this test on my own ventures before I take it into anyone else’s company.

D30 pulls a three statement fact file out of an ASX annual report and runs eight articulation gates over it. The unit is one completed fact file that passes all eight. I can tell you what a fact file costs to produce, including model spend, and I can tell you how that number has moved. I have never managed the business to analyst hours saved, because hours saved is a number you can claim twice and a fact file is a number you can count.

D23 sells managed Apache Superset. The unit is a dashboard running in production that somebody opened this month. Not a deploy. A deploy that nobody looks at is cost with no denominator.

SearchFIT counts one tracked prompt scanned across the answer engines. When a model price drops, that unit cost drops, and I can see it in a week rather than infer it from a vendor’s slide.

None of those units are obvious. Each took an argument to settle. That is the actual work, and it is why the metric is rare.

Where this breaks

Any denominator you publish becomes a target. Count merged pull requests and you will get more of them, smaller, split across branches, and your ratio will improve while nothing ships. The thirty day survival test slows that down. It does not stop it. If you put this number in front of a board you have to expect the number to be gamed, and the honest position is that this metric is a diagnostic for the people running delivery, not a compensation input.

It also does not travel outside software delivery. If your AI value is sitting in a claims queue or a contact centre, the unit is a resolved case, not a merged change, and forcing a delivery frame onto it will mislead you.

The hardest objection is timing. Cost per shipped change is noisy for two or three quarters, and a CFO under pressure this quarter cannot wait that long. Headcount is measurable today, it is unambiguous, and it moves the P&L in the period. That is not stupidity. That is a real constraint, and it is exactly why the wrong metric keeps winning.

What I would do on Monday

1. Write the unit down in one sentence and get the CTO and the CFO to sign the same sentence. If they cannot agree what a shipped change is, stop there, because that disagreement is the finding.

2. Total the AI line properly. Seats, token overages, API keys on personal cards, and the two tools nobody put through procurement. Split it by team. Expect a spread like the nine times gap in the Jellyfish example.

3. Instrument the denominator with a thirty day survival window. Count changes that reached production and stayed. Reversals do not count.

4. Compute the ratio for the four quarters before AI landed and every quarter since. You need the baseline more than you need the current number.

5. Put a freeze on any headcount decision justified by AI capacity until you have two quarters of the ratio. If the capacity is real it will still be there in six months.

Sixty two engineers is a real number inside a real model published in July. It is either coming off your payroll or going into your output. Decide which. Then check, in two quarters, whether the ratio agrees with you.

Brightlume does this work with enterprise teams. If the gap between the AI pilot and the P&L is the problem you have, talk to me.

A financial report is not a document. It is three statements welded together by about eight pieces of arithmetic that have to close to the dollar, and almost nobody in Australian private capital runs them. A number that has not been tied is not evidence. It is a quote.

What the ties actually are

The gates are not sophisticated. That is the point.

A1 is the balance sheet: total assets equal total liabilities plus total equity. A2 is component summation, run six ways, because a subtotal that does not equal the lines above it means either the extraction is wrong or something is sitting in a bucket nobody named. A3 walks the income statement: EBIT less finance costs plus finance income equals pretax, pretax less tax plus discontinued equals NPAT. A4 is the cash flow: opening cash plus operating plus investing plus financing plus FX equals closing, and closing equals the balance sheet. A5 rolls retained earnings forward: prior year plus profit attributable to members less dividends paid. A7 rolls the share count.

Each one is a subtraction. All of them together are the difference between a model built on numbers and a model built on typing.

The mechanical reason this is not already standard practice in Australia is that our filings are not machine readable. ASIC has offered voluntary digital lodgement since 2010 and, as at 2026, not one company has taken it up. XBRL is in active use across 65 countries. The EU, the UK and the US all mandate digital financial reporting. Australia does not. The Productivity Commission’s interim report recommends mandatory iXBRL using the IFRS Accounting Taxonomy for disclosing entities and a phased end to PDF submission. Until that lands, every ASX annual report reaches an analyst as pixels, gets keyed into a spreadsheet, and the spreadsheet inherits whatever the analyst read.

AASB 18 takes effect for periods beginning 1 January 2027 and rebuilds the face of the income statement. Every model keyed by hand off a PDF will need to be rekeyed.

The errors are not rare and they are not small

ASIC published REP 819 on 31 October 2025 covering the 2024 to 2025 cycle. It reviewed 254 company financial reports, 220 of them ASX listed and 34 large proprietary, plus 60 superannuation entity reports. Twenty two entities went to detailed surveillance. Eighteen entities made or agreed to make changes across 19 areas of concern. Energy World Corporation impaired assets by US$793.8 million. Bell Group Holdings restated to consolidate a subsidiary it had previously accounted for incorrectly. Vulcan Energy Resources restated its employee benefits presentation. Two entities corrected how they presented non-IFRS profit measures.

ASIC’s own enforcement in the same period ran to 21 infringement notices and more than $4 million for FY24 financial reporting breaches, including $792,000 across four Canva Group entities and $594,000 across three Mecca entities at $198,000 each. In its later update ASIC recorded Viva Energy taking $558.8 million of retail site impairment in FY25, of which $25 million, 4.5 per cent, came from a revised impairment testing methodology rather than a change in the assets. Pure Foods Tasmania reversed $4.5 million of deferred tax asset recognition, equal to 31 per cent of its total assets.

The cash flow statement is the softest target of the three. In the United States, where a machine reads every filing, the SEC’s then Chief Accountant Paul Munter noted on 4 December 2023 that cash flows had been the fourth most common accounting issue cited in restatements from 2003 through 2022, and the single most frequently cited issue among large accelerated filers. His point was blunter than the statistic: he rejected the argument that an error which only moves a number between operating, investing and financing is immaterial. Restatements rose 7 per cent in 2024 and Big R restatements hit a nine year high, on Audit Analytics data.

Then there is the lever that sits right on the operating line. Amendments to IAS 7 and IFRS 7 on supplier finance arrangements took effect for periods beginning on or after 1 January 2024, and they exist because the classification choice is worth real money. Call the balance a trade payable and operating cash flow and free cash flow improve. Call it a financial liability and the fees drop below EBITDA. Same cash, two different stories, and which story you get told depends on which metric the seller thinks you underwrite.

Where this breaks

Gates test arithmetic, not judgement, and judgement is where most value gets destroyed. Energy World’s balance sheet almost certainly tied to the dollar in the year before that US$793.8 million impairment landed. Impairment timing, revenue recognition policy, capitalisation of development spend and provision releases all pass every gate I have described. If you run the ties and conclude the numbers are clean, you have confirmed the report is internally consistent and nothing else.

The second problem is signal to noise. On a first pass, most gate failures are my extraction, not the company’s accounting. A residual bucket that swallows 59.7 per cent of current assets is a tagging failure, not fraud. You have to fix the parser before the output means anything, and that work is unglamorous and takes longer than building the gates did.

Third, for a large cap audited by a big four firm, the arithmetic is usually right and the gates earn nothing. They pay for themselves in small and mid caps, in recently listed companies, and above all in the unaudited management accounts that private capital actually spends most of its diligence on. That is the population where I would run them.

What I would do on Monday

1. Take the last three diligence packs you signed off and check A1 and A4 by hand on each. Two subtractions per pack. If any of the six fails, you have your answer about the process.

2. Stop accepting the PDF as the deliverable. Ask for the trial balance and the model that produced the summary. If nobody can produce the workbook, that is the finding.

3. Add one line to the information request list: disclose all supplier finance arrangements, the carrying amounts by balance sheet line, and the range of payment due dates against comparable trade payables. IAS 7 already requires it.

4. Put the gate results in the investment committee paper as a page, with the tolerance you used stated on it. A tolerance you did not write down is a tolerance you will argue about later.

5. Pick one company you already own and rebuild its last three years from the filings rather than from your own model. Compare. That exercise is where the tooling pays for itself or does not.

The line that started it

I wrote A4 BSCASH expecting a plumbing check. It is not a plumbing check. It is the question of whether the document you are underwriting agrees with itself, and until you have asked it, every number downstream is something you read rather than something you know. Thirty one units on a share count is nothing. Not knowing it was there is the problem.

Run the subtraction.

I built D30 to pull three-statement fact files out of ASX annual reports and flag where the numbers stop articulating. If you run diligence on listed or pre-IPO assets, have a look.

For most of last year it was. Which model, what context window, what the token bill looks like at volume. I would spend the first hour of a review on that page, because it was the line with the biggest number next to it, and a big number feels like a decision.

On 30 July 2026 OpenAI repriced GPT-5.6 Luna by 80%. Input went from $1.00 per million tokens to $0.20. Output went from $6.00 to $1.20. Terra came down 20%, from $2.50 and $15.00 to $2.00 and $12.00. Sol, the prestige tier, did not move at all.

Every hour anyone spent last quarter optimising a model choice on price was deleted by one update to a pricing page.

The thesis

Most enterprise AI budget is going into the layer that will be a utility bill inside 18 months. The two things that will still be yours after that happens are your data contracts and your evaluation sets, and almost nobody is funding either.

That is not a call to spend less. It is a call to spend the same money one layer down.

The numbers behind the claim

Menlo Ventures surveyed 495 US enterprise AI decision makers between 7 and 25 November 2025. Total enterprise AI spend for the year came to $37 billion, up from $11.5 billion in 2024 and $1.7 billion in 2023.

Split that $37 billion. Applications took $19 billion. Infrastructure took $18 billion. Inside infrastructure, foundation model APIs took $12.5 billion, model training infrastructure took $4.0 billion, and storage, retrieval and orchestration took $1.5 billion.

Read that ratio again. Enterprises spent 8.3 times more renting model capability than building the layer that feeds it.

Now put that next to the price curve. Andreessen Horowitz measured the cost of a fixed level of model performance falling roughly 10x per year: GPT-3 level quality cost $60 per million tokens in November 2021 and $0.06 three years later, a 1,000x collapse. GPT-4 level quality fell about 62x from its March 2023 launch. The 80% cut on 30 July 2026 is not an event. It is one visible step on a curve that has not broken once in five years.

You cannot build an advantage on a line item that falls 10x a year. Everyone else gets the same cut on the same day.

The switching data says nobody is locked in either

Perplexity published enterprise usage across January to December 2025. The leading model’s share of queries fell from 47.5% to roughly 23% inside twelve months. Early in the year two models held 91.5% of queries. By the end, four models each held more than 10%. Across organisations, 43.6% used more than one model at some point in the year, and 53% of users who pick a model switched between models at least once inside a single working day. The top 50 enterprise accounts averaged 30 models each, against seven for a typical account.

Menlo’s share numbers land in the same place: Anthropic 40%, OpenAI 27%, Google 21%. OpenAI held 50% in 2023.

A market where the leader halves its share in a year, where half of power users change model before lunch, and where the price of the thing drops 80% overnight, is a market with no defensible position in it. That is the definition of a commodity input. It is a fine thing to buy. It is a terrible thing to build a strategy on.

What does not commoditise

Two things. Both boring. Both underfunded.

The first is the data contract: an explicit agreement between the team that produces a table and the team that consumes it, covering schema, ownership, freshness and how a breaking change gets announced. Without it, someone renames a column upstream, your retrieval layer quietly returns worse context, your answers degrade, and your AI team spends three weeks debugging a model that never changed.

The second is the evaluation set: your own labelled examples of what a correct answer looks like for your business, run automatically against every model, prompt and retrieval change.

Here is the part people miss. Evaluation is filed under governance, so it gets treated as a brake. It is the opposite. The eval set is the mechanism that converts a price cut into cash.

When Luna drops 80%, you can only take that money if you can prove the cheaper model still does your job. With a golden set of a few hundred labelled cases you run it in an afternoon and migrate. Without one, you have a meeting, then a pilot, then a quarter of drift, and you keep paying the old price on the old model because nobody can sign off that the new one is safe. The company with evals harvests every price cut in the industry automatically. The company without them watches the curve go past.

The evidence that this layer is genuinely missing

Gartner surveyed 1,203 data management leaders in July 2024 and concluded that through 2026 organisations will abandon 60% of AI projects that are not supported by AI ready data. A separate Gartner survey of 248 data management leaders found 63% either lack appropriate data management practices for AI or are not sure whether they have them.

On the evaluation side, Gartner reported on 30 March 2026 that 15% of generative AI deployments currently include LLM observability, and projected 50% by 2028. Eighty five percent of deployments running today cannot tell you why an answer came out the way it did.

Meanwhile the build and buy ratio flipped hard. In 2024 enterprises built 47% of their AI use cases in house. In 2025 they built 24% and purchased 76%. That is a rational response to falling model costs and improving vendor products, and it has a side effect nobody priced: when you buy the application and rent the model, the only part of the stack you still own outright is your data and your definition of correct. Those are exactly the two lines that got $1.5 billion out of $37 billion.

I run this pattern in my own products. D30 pulls three statement fact files out of ASX annual reports, and the model that reads the PDF is not the asset. The articulation gates are. Eight of them, checking that the balance sheet balances, that the cash flow ties to the movement in cash, that the tagged store reconciles to the statements. If the numbers do not articulate, the fact file fails, whatever model produced it. That gate lets me change the model underneath without changing my confidence in the output. Same logic at D23, where clients pay for Superset to be run correctly rather than for the software. Same at SearchFIT, which has to query answer engines it does not control and would be worthless without a stable definition of what counts as a citation.

Where this breaks

The frontier is not commoditised, and my thesis does not cover it. Sol held at $5.00 and $30.00 on the same day Luna fell 80%. That gap is the market telling you that hard, long horizon reasoning still has scarcity value. If your use case genuinely sits there, model selection decides whether the thing works at all, and no eval set rescues a model that cannot do the task.

The thesis also over-fires for small companies. If you have three AI use cases and no data platform, building formal data contracts is over-engineering. Buy the application, use the vendor’s evaluation tooling, revisit at scale.

And evaluation sets rot. A golden set built on last year’s task distribution will happily approve a model that fails on what your customers ask today. An eval set is a living asset with an owner, not a compliance artefact you build once. If you are not editing it monthly, it is already lying to you.

What I would do on Monday

1. Print the AI budget. Mark every line by whether it survives a model swap. If more than half the spend dies when the model changes, you are funding the commodity.

2. Build one golden set this week. One hundred to three hundred labelled examples for your single highest volume task. Not a platform, not a vendor selection. A spreadsheet and a script are enough to start.

3. Put a contract on the three tables that feed retrieval. Named owner, schema, freshness target, and a rule that breaking changes get announced before they ship.

4. Re-run last quarter’s most expensive workload on the cheapest current model against that golden set. If it passes, take the 80% now. That single exercise usually pays for the eval work in one billing cycle.

5. Change the metric in the monthly pack from tokens consumed to cost per accepted output. Tokens are a price you do not control. Acceptance is a quality you do.

Close

The first thing I check now is the eval set. If there is one, I know the team can move when the price moves, and the model on the architecture diagram is a detail. If there is not, I know the exact conversation we will have in nine months, because the price will have fallen again and they will still be paying the old rate.

The model is rented. The definition of correct is owned.

Brightlume does this work with enterprise teams. If the gap between the AI pilot and the P&L is the problem you have, talk to me.

Those reports run two to four weeks and cover five things: architecture and scalability, the team, security and compliance, infrastructure spend, and roadmap risk. I have been on both sides of them. The operator being audited, and the person an investor calls three months after close when the plan is already behind.

Not one standard scope measures how long it takes the company to turn a decision into a shipped, safe change.

That number is the deal.

The thesis

Technical diligence audits the artefact. The value leaks out of the system.

You are not buying a codebase. You are buying a rate: the rate at which this business converts capital into shipped change without breaking the part that makes money. Nobody measures that rate before the wire clears. Then the value creation plan assumes a rate nobody has ever observed.

The arithmetic changed and the diligence did not

For roughly a decade, a buyout needed about 5% annual EBITDA growth to reach a 2.5x multiple on invested capital. Multiple expansion and cheap debt carried the rest. Bain’s 2026 report puts the requirement now at 10% to 12% annual EBITDA growth for the same outcome, with borrowing costs sitting at 8% to 9% and leverage at 30% to 40% of the capital structure.

Double the operating growth, from the same assets, under more expensive debt.

The rest of the picture makes it worse. There are 32,000 unsold portfolio companies carrying $3.8 trillion of value. Average holding periods at exit have stretched to around 7 years, up from the 5 to 6 that was normal between 2010 and 2021. Distributions to LPs came in at 14% of NAV in 2025, the fourth consecutive year below 15%.

So returns have to come from operations. In a software or software-enabled asset, operations means one thing: how fast and how safely the company can change its own product. That is the engine. Diligence inspects the paint.

The code audit stopped being a signal

Here is what broke the old method.

CloudBees surveyed more than 200 enterprise technology leaders for its 2026 State of Code Abundance report. AI now generates or assists 61% of the average enterprise codebase. In the same survey, 81% of leaders report an increase in production issues tied to AI-generated code, while 92% say they are confident in that code’s production readiness. Organisations could attribute only about one third of their AI spend to a specific business outcome.

Read those four numbers together. The majority of what a code audit now reads was written by a model. It is stylistically uniform, cheap to produce, and syntactically clean. A static scan of it comes back green. And the same leaders sitting on top of it report more incidents, not fewer.

The artefact and the risk have come apart. SonarQube can tell you the code is tidy. It cannot tell you whether this company can safely change it on a Tuesday afternoon with two engineers on leave.

Three things the standard scope misses

One: absorption capacity. The value creation plan lands with a dozen or more initiatives in the first year. ERP consolidation, pricing rebuild, a data platform, two integrations, a security remediation. The relevant question is not whether each is a good idea. It is how many concurrent initiatives this delivery system has ever finished in a single year. That number exists. It is in the last three years of shipped work. Diligence almost never asks for it.

Two: the review and approval path. Writing changes got cheap. Approving them did not. In most mid-market software businesses there are three or four people who can sign off a change to billing, entitlements or customer data. Their names are not in the diligence report. Their retention risk is the single largest technical risk in the deal, and it is usually filed under key person dependency with no measurement attached.

Three: the gap between the process document and the deploy log. Every data room contains a document describing how the company ships software. It is aspirational. The deploy history, the incident records and the pull request timestamps describe what actually happens. The distance between those two artefacts is the most predictive thing in the whole room, and reading it takes an afternoon.

I built D30 to pull three-statement fact files out of ASX annual reports and run articulation gates over them, because I do not trust a summary figure I cannot trace back to the filing that produced it. The instinct is identical here. A process document is a summary figure. The deploy log is the filing.

Where I am wrong

This thesis has real limits and I would rather name them than have a reader find them.

For asset-light businesses where software is a thin wrapper on a services model, the codebase genuinely is the risk, and licence contamination or a single hard-coded customer integration can be the whole finding. For a carve-out, the delivery system you measure belongs to the seller and will not survive separation, so measuring it tells you less than modelling what replaces it. And there are deals where the data simply is not available: a competitive process, a two-week window, no repository access, a seller who will not open the incident history. In that situation the code audit is not the wrong tool. It is the only tool you are allowed to use, and a good one done honestly beats a throughput analysis you invented.

The failure I am pointing at is not doing a code audit. It is doing a code audit and believing you have assessed delivery risk.

What I would do on Monday

1. Add one line to the diligence request list. The last 90 days of production deployments with timestamps, and the last 12 months of incidents with time to restore. Not the process document. The records.

2. Count the approvers. Ask who can approve a change to the revenue-critical paths. If the answer is fewer than four people, that is a retention clause in the SPA, not a footnote.

3. Divide the plan by the observed rate. Take the number of year-one initiatives in the value creation plan and compare it against the largest number of concurrent initiatives the company has actually completed in a year. If the plan is more than double, the plan is fiction and the EBITDA bridge is built on it.

4. Instrument in the first 30 days post-close. Jellyfish, DX and Swarmia all read from Jira and GitHub and give you a baseline inside a fortnight. Set that baseline before the operating partner starts changing things, or you will never know whether anything you did worked.

5. Re-underwrite once, at day 90. With real throughput data in hand, restate the operating case. Investors hate this. Doing it at month 18 instead is worse.

Back to the report

That two to four week technical diligence report will still get written on every one of the next $900 billion of deals. It is not useless. It is scoped for a market where multiple expansion did the heavy lifting and 5% growth was enough.

That market is gone. The number that has to double is produced by a system nobody in the room measured.

Buy the engine, not the paint.

I built D30 to pull three-statement fact files out of ASX annual reports and flag where the numbers stop articulating. If you run diligence on listed or pre-IPO assets, have a look.

This article breaks down the real differences between seed and Series A, moving beyond superficial funding metrics to examine the operational, strategic, and technical transformations that must occur. Whether you're a CTO at a government agency evaluating an AI vendor, an enterprise technology leader building internal platforms, or a founder preparing for Series A, this guide will help you understand what's truly at stake.

The Funding Landscape: Size, Structure, and Investor Profile

Seed Funding: Proving the Concept

Seed funding typically ranges from $500K to $2M, though in AI-driven startups, seed rounds have inflated to $3–5M in competitive markets. According to CRV's analysis of seed funding versus Series A, seed investors are primarily angels, early-stage venture firms, accelerators, and sometimes founder networks (friends and family). These investors are betting on you—the founding team—and the problem you're solving, not necessarily on a proven market or revenue traction.

Seed funding is designed to answer one fundamental question: Does the market want this? You're building a prototype, validating assumptions, and proving that customers will pay (or at least engage meaningfully). The capital is meant to last 12–18 months, giving you runway to iterate, test channels, and gather early customer feedback.

At the seed stage, your technology stack is often scrappy. You might be using off-the-shelf tools, monolithic architectures, or even no-code solutions. The goal is speed and learning, not scalability or enterprise-grade compliance. This is the phase where many founders are still asking: "Should we build this in-house, or should we use existing platforms?"

Series A Funding: Scaling the Proven Model

Series A funding typically ranges from $2M to $15M, with AI-native startups often raising $10–25M in 2024–2025. Series A investors are institutional venture capital firms with deep sector expertise, board seats, and multi-year investment horizons. These are firms like Sequoia, Accel, Andreessen Horowitz, or industry-focused funds.

Series A investors are not betting on the team alone—they're betting on product-market fit (PMF). By the time you raise Series A, you need to demonstrate:

• Customer traction: Paying customers, or at minimum, strong product engagement and clear path to revenue

• Repeatable unit economics: Evidence that your go-to-market motion can scale

• Market validation: Proof that the problem is real and the market is addressable

• Competitive differentiation: Why you'll win, not just why the problem matters

Series A capital is meant to scale what works. You're hiring aggressively (often tripling your team), entering new markets, building enterprise-grade infrastructure, and preparing for the next growth phase. As Hustle Fund explains in their comparison of pre-seed, seed, and Series A, the Series A stage is fundamentally about proving you can execute at scale, not just prove the concept.

Governance, Control, and Decision-Making Authority

Photo by Austin Distel on Unsplash

Seed Stage: Founder Autonomy

At seed stage, you retain significant autonomy. Yes, you have investors, but they're typically hands-off or supportive rather than directive. Board meetings might be quarterly or semi-annual. Decisions about product direction, hiring, technology choices, and go-to-market strategy rest primarily with you and your co-founders.

This autonomy is a double-edged sword. It enables rapid iteration and bold pivots, but it also means you're fully accountable for every decision. There's no institutional safety net telling you to slow down or be more cautious. Seed investors expect you to move fast and break things—within reason.

Series A: Institutional Governance

Series A fundamentally changes governance. Your Series A investor typically gets a board seat. They participate in quarterly board meetings with detailed metrics reviews. They have information rights, anti-dilution provisions, and sometimes protective provisions (the ability to block certain decisions like hiring a new CEO, raising debt, or selling the company).

This shift from founder autonomy to shared governance is profound. You now have a fiduciary obligation to your investors to maximize their returns. This changes decision-making calculus: you can't just pivot on a whim anymore. Major strategic decisions require board alignment. Hiring becomes subject to budget scrutiny. Technology choices must be defensible to institutional investors who understand risk and compliance.

For CTOs specifically, this means your technology strategy now has to be communicated and aligned with board-level stakeholders. Decisions about architecture, infrastructure, and security posture are no longer purely technical—they're business decisions that affect investor confidence and company valuation.

Technology Architecture and Infrastructure

Photo by Christina @ wocintechchat.com M on Unsplash

Seed Stage: Iterate Fast, Scale Later

At seed stage, your technology choices should prioritize speed and learning over scalability. Many successful seed-stage startups use:

• Monolithic architectures or simple microservices (not over-engineered)

• Off-the-shelf platforms for non-differentiating functions (authentication, payments, analytics)

• Managed services over self-hosted infrastructure (AWS Lambda, Stripe, Supabase, etc.)

• Single-region deployment with basic disaster recovery

• Manual processes for deployment, monitoring, and incident response

The reason is simple: you're still learning what customers want. Investing heavily in scalable architecture before you have product-market fit is premature optimization. Your goal is to get something in front of customers quickly and iterate based on feedback.

At seed stage, you might not even have embedded analytics or sophisticated monitoring. You're probably using basic dashboards and direct customer conversations to understand usage patterns. Security is important, but enterprise-grade SOC 2 compliance might not be a priority yet—your customers are early adopters who understand the risk.

Series A: Build for Scale and Enterprise Requirements

Once you've raised Series A and proven product-market fit, your technology strategy must shift. You're now building for:

• Scalability: Your current architecture needs to support 10x growth without major rewrites

• Reliability: 99.5% uptime SLAs become table stakes; customers are now paying significant money and expecting enterprise-grade reliability

• Security and compliance: If you're selling to enterprises or government agencies, SOC 2 Type II certification becomes essential. Understanding what SOC 2 and ISO 27001 audits entail is critical for Series A startups targeting enterprise buyers

• Embedded analytics and observability: You need sophisticated monitoring, logging, and analytics to understand system behavior at scale

• Multi-region deployment: Enterprise customers often require data residency and geographic redundancy

• API-first architecture: You're now integrating with enterprise systems, not just standalone applications

This is the phase where building scalable AI platforms with proper platform engineering becomes critical. If you're an AI-native startup, you need to think about how your models will be served at scale, how you'll handle model versioning and monitoring, and how you'll ensure compliance with enterprise governance requirements.

Many Series A startups also adopt embedded analytics solutions like Apache Superset combined with ClickHouse to provide customers with real-time insights into their data. This requires a different infrastructure approach than seed-stage analytics.

Hiring, Organizational Structure, and Talent Strategy

Photo by Deng Xiang on Unsplash

Seed Stage: Hire Generalists, Move Fast

At seed stage, you typically have a small team—maybe 5–15 people. Your engineering team is lean and generalist. You hire people who can wear multiple hats: full-stack engineers who can build frontend, backend, and DevOps; product people who can also do user research; marketers who can also do sales.

Your hiring criteria emphasize:

• Adaptability: Can they handle ambiguity and change direction quickly?

• Ownership: Do they take initiative without waiting for permission?

• Technical depth in one area: You need some specialization, but generalist capability is more valuable

• Startup experience: Prior experience in high-uncertainty environments is a plus

Compensation is typically lower than Series A, but equity is a larger percentage of total comp. You're asking people to take a risk on an unproven company in exchange for potentially significant upside.

Series A: Build Specialized Teams and Infrastructure

Series A hiring is fundamentally different. You're likely tripling your team from 15 to 45+ people. You now need:

• Specialized engineers: ML engineers, infrastructure engineers, security engineers, data engineers

• Product managers: Dedicated PMs for each product area or customer segment

• Sales and customer success teams: You can't scale revenue with founder-led sales anymore

• Finance and operations: CFO, controller, HR, legal

• Customer success and support: Dedicated teams to ensure customers succeed (and renew)

Your hiring criteria shift toward:

• Specialized expertise: You need people who have solved these problems before at scale

• Leadership capability: As you grow, you need people who can build and lead teams

• Enterprise experience: If you're selling to enterprises, you need people who understand enterprise buying, implementation, and support

• Compliance and security mindset: Especially important if you're targeting regulated industries

Compensation also increases significantly. Series A startups typically offer 50–70% of FAANG salaries (depending on market and role), plus meaningful equity. You're now competing with established companies for talent in ways you weren't at seed stage.

For CTOs specifically, this is the phase where you might bring in a fractional CTO or program leadership to help scale your technology organization if your founding CTO doesn't have experience managing large engineering teams or enterprise sales cycles. Many Series A startups benefit from understanding what a fractional CTO actually does in the first 90 days to accelerate organizational build-out.

Go-to-Market Strategy and Sales Motion

Seed Stage: Product-Led or Founder-Led Sales

At seed stage, your go-to-market motion is typically one of:

• Product-led growth (PLG): Customers self-serve, sign up, and use the product without talking to sales

• Founder-led sales: You (the founder) are doing most of the selling, often to personal networks or warm introductions

• Land-and-expand: Get in with a small initial use case, then expand within the customer

Your target customers are typically early adopters: startups, innovative teams within larger organizations, or forward-thinking smaller companies. You're not yet selling to enterprise procurement teams or government agencies. Your sales cycle is measured in weeks, not months.

Marketing at seed stage is often organic: content, community, word-of-mouth, and maybe some paid acquisition if you have clear unit economics.

Series A: Enterprise Sales and Multi-Channel Go-to-Market

Once you've raised Series A, your go-to-market strategy must evolve to support faster growth. This typically means:

• Building a sales team: You hire a VP of Sales and start building an enterprise sales motion

• Longer sales cycles: Enterprise customers need to evaluate security, compliance, integration, and ROI—this takes 3–6 months

• Sales enablement: You need tools, processes, and training to help your sales team close deals

• Channel partnerships: You might work with resellers, integrators, or consulting firms to reach new markets

• Customer success infrastructure: You need dedicated teams to ensure customers succeed, reduce churn, and expand accounts

This is also the phase where you might start targeting government agencies and regulated industries. These customers have strict requirements around security, compliance, and data residency. Understanding how to structure engagements for government and enterprise buyers becomes critical.

Compliance, Security, and Risk Management

Seed Stage: Security Basics, Minimal Compliance

At seed stage, security is important, but it's not your primary focus. You're likely implementing:

• Basic authentication and authorization: Password management, two-factor authentication

• Encrypted data in transit and at rest: Standard HTTPS and database encryption

• Basic access controls: Role-based access, principle of least privilege

• Incident response procedures: Basic runbooks for common security incidents

Compliance is minimal. You might have a privacy policy and terms of service, but you're not pursuing SOC 2 certification or conducting formal security audits. Your customers are early adopters who accept higher risk in exchange for innovation.

Series A: Enterprise-Grade Security and Compliance

Series A is where security and compliance become non-negotiable. If you're selling to enterprises or government agencies, you need:

• SOC 2 Type II certification: This requires 6+ months of audit evidence, so you need to start immediately

• ISO 27001 certification: Increasingly required for enterprise and government sales

• Data residency and sovereignty: Ability to keep customer data in specific geographic regions

• Audit logging and monitoring: Comprehensive logging of all access and changes

• Incident response and breach notification: Formal procedures and legal compliance

• Vendor risk management: Assessment of your own vendors and third-party dependencies

This is not a nice-to-have; it's a blocker for enterprise deals. Many Series A startups find that 20–30% of their Series A capital goes toward building out security and compliance infrastructure. Understanding the full scope of SOC 2 and ISO 27001 requirements is essential for Series A CTOs.

If you're building AI products, you also need to think about AI governance, model monitoring, and explainability—especially if you're selling to regulated industries or government agencies.

Metrics, KPIs, and Investor Reporting

Seed Stage: Learning and Validation Metrics

At seed stage, your metrics focus on learning and validation:

• Customer acquisition cost (CAC): How much does it cost to acquire a customer?

• Retention and churn: Are customers sticking around? Are they renewing?

• Product engagement: Are customers actually using the product?

• Net Promoter Score (NPS): Are customers happy enough to recommend you?

• Revenue or ARR: If you're monetized, what's your annual recurring revenue?

These metrics help you understand whether you have product-market fit. You're looking for signals that customers want what you're building. Investor reporting might be quarterly, but it's relatively lightweight—a few slides showing progress toward key milestones.

Series A: Investor-Grade Metrics and Financial Reporting

Series A completely changes your metrics and reporting rigor. You now need:

• Monthly recurring revenue (MRR) and annual recurring revenue (ARR): Tracked precisely with cohort analysis

• Customer acquisition cost (CAC) and lifetime value (LTV): Detailed unit economics by customer segment

• Magic number: (Quarter revenue growth ÷ prior quarter sales and marketing spend) to assess go-to-market efficiency

• Churn rate: Both logo churn (% of customers lost) and revenue churn (net revenue retention)

• Burn rate and runway: Monthly cash burn and how many months of runway you have

• Headcount and hiring plan: Detailed org structure and hiring roadmap

You'll also need formal financial statements, board packages, and quarterly investor updates. Many Series A startups hire a CFO or finance person to manage this rigor. Your board will scrutinize these metrics monthly, and they'll inform strategic decisions about hiring, spending, and go-to-market investment.

Technology Debt and Technical Decisions

Seed Stage: Accept Technical Debt for Speed

At seed stage, you should actively embrace technical debt if it gets you to market faster. This means:

• Choosing speed over elegance: Using frameworks and libraries that get you to market quickly, even if they're not perfect

• Minimal documentation: Your team is small enough that knowledge sharing is verbal

• Scripted deployments: Manual processes that work but aren't fully automated

• Single database: Monolithic data storage rather than complex distributed systems

• Minimal testing: Focused testing on critical paths, not comprehensive test coverage

The rationale is simple: at seed stage, the biggest risk is not learning what customers want. Technical debt is acceptable as long as it doesn't prevent you from learning or iterating quickly.

Series A: Begin Paying Down Technical Debt

Once you've raised Series A, you need to start paying down technical debt systematically. This means:

• Investing in testing and CI/CD: Automated testing, continuous integration, and deployment pipelines

• Documentation: Formal documentation of architecture, APIs, and operational procedures

• Monitoring and observability: Comprehensive logging, metrics, and alerting

• Refactoring: Allocating engineering time to improve code quality and reduce complexity

• Infrastructure automation: Infrastructure-as-code, configuration management, and automated deployment

You typically allocate 20–30% of engineering effort to technical debt paydown in Series A. This is the phase where you might also adopt AI-native development practices if you're building AI products, ensuring your development process is optimized for rapid iteration with AI models.

The Role of External Expertise and Advisory

Seed Stage: Occasional Advisory

At seed stage, many founders rely on informal advisory networks: mentors, other founders, and angel investors who provide occasional guidance. You might hire a fractional CFO to help with financial modeling, but you're primarily self-sufficient.

Series A: Structured Advisory and Leadership

Series A is where many startups bring in external expertise systematically. This might include:

• Fractional CFO or controller: To manage financial planning and reporting

• General counsel or outside counsel: To handle legal, compliance, and contracts

• VP of Sales or sales consultant: To build out enterprise sales motion

• CTO or engineering advisor: To help scale the engineering organization and navigate enterprise requirements

Many Series A startups benefit from understanding the role of a fractional CTO in scaling technical organizations. A fractional CTO can help you navigate enterprise sales requirements, build SOC 2 compliance, and scale your engineering team without the cost of a full-time executive. Learning what to look for when hiring a fractional CTO is increasingly important for Series A startups targeting enterprise and government customers.

If you're building AI products, you might also consider working with a venture studio or co-build partner to accelerate product development and navigate enterprise AI requirements. Understanding the difference between building vs buying autonomous agents becomes a critical Series A decision if you're in the AI space.

Customer Profile and Market Positioning

Seed Stage: Early Adopters and Innovators

Your seed-stage customers are typically:

• Startups and scaleups: Companies that are also taking risks and moving fast

• Innovative teams within larger organizations: Forward-thinking departments willing to try new solutions

• Smaller companies: Typically under $100M in revenue, willing to work with immature vendors

• Price-sensitive: They're looking for value, but they're not willing to pay enterprise prices

These customers accept that your product is evolving, your support might be limited, and you might pivot. They're willing to work closely with you to shape the product.

Series A: Enterprise and Government Focus

Once you've raised Series A, your customer profile typically shifts upmarket:

• Mid-market and enterprise: Companies with $100M+ in revenue, formal procurement processes

• Government agencies: Federal, state, and local government, with strict compliance and security requirements

• Regulated industries: Financial services, healthcare, energy, defense—where compliance is non-negotiable

• Price-insensitive (relatively): These customers are willing to pay for reliability, security, and support

These customers have different expectations:

• Formal contracts and legal review: Not simple terms of service

• Security and compliance audits: They'll audit your systems and processes

• SLAs and support: They expect 24/7 support and guaranteed uptime

• Integration and customization: They need to integrate with existing systems

• Multi-year commitments: They want stability and long-term partnerships

This shift in customer profile is profound. It changes everything about how you operate: your sales process, your support model, your product roadmap, and your technology infrastructure.

The Verdict: What Actually Changes

For Founders

If you're a founder navigating the seed-to-Series A transition, here's what you need to understand:

Autonomy decreases, but resources increase. You'll have less freedom to pivot on a whim, but you'll have significantly more capital and talent to execute your vision. This trade-off is generally worth it if you've truly found product-market fit.

Complexity increases dramatically. You're no longer just building a product; you're building a company. You need financial discipline, compliance rigor, and formal processes. Many founders struggle with this transition because it feels bureaucratic, but it's essential for scaling.

Your role changes. If you were doing product, sales, and support at seed stage, Series A requires you to focus. Most successful Series A founders focus on product and strategy, hiring specialists for everything else.

Enterprise and government customers become viable. Series A is when you can realistically target large enterprises and government agencies. These are high-value customers, but they require investment in security, compliance, and support.

For Enterprise CTOs and Technology Leaders

If you're evaluating a Series A startup as a potential vendor or partner:

Series A is a maturity inflection point. A Series A startup has proven product-market fit and has invested in the infrastructure to support enterprise customers. This is fundamentally different from a seed-stage startup.

Look for security and compliance investment. A well-run Series A startup will have SOC 2 on the roadmap (or achieved). They'll have formal security and compliance processes. If they don't, that's a red flag.

Evaluate their technical depth. Series A startups should have specialized engineers (ML, infrastructure, security). If they're still relying on generalists, they're not ready for enterprise scale.

Assess their go-to-market maturity. Do they have a sales team? Customer success? Or are they still doing founder-led sales? Series A startups should have formal sales and CS processes.

Consider bringing in advisory support. If you're evaluating a Series A startup for a critical use case, working with a fractional CTO or technical advisor can help you assess technical risk and ensure the vendor can scale with your needs.

For Government and Regulated Industry Buyers

If you're a government agency or work in a regulated industry:

Series A is the earliest you should consider a startup vendor. Seed-stage startups are too risky for mission-critical systems or regulated use cases. Series A startups have invested in compliance and security, making them viable partners.

Require SOC 2 Type II certification. Don't accept promises of future compliance. If they don't have SOC 2, they're not ready.

Evaluate their ability to meet data residency requirements. Government customers often require data to stay in the US (or specific regions). Series A startups should be able to support this.

Consider multi-region deployment and disaster recovery. Government systems need redundancy and failover capabilities. Evaluate whether the vendor can support this.

Engage a technical advisor early. Working with a fractional CTO or program leadership can help you navigate vendor selection, implementation, and ongoing governance for startup partners.

Conclusion: The Inflection Point

The transition from seed to Series A is not just a funding milestone—it's a fundamental shift in how a startup operates. The capital increases, the team grows, the customers change, and the complexity multiplies. For some founders, this transition is exciting and energizing. For others, it feels like losing the scrappy, autonomous culture that made the seed stage fun.

The reality is both: Series A requires more discipline, more process, and more complexity. But it also enables you to build something truly meaningful at scale. You can hire the best engineers, invest in enterprise-grade infrastructure, and serve customers who have real, mission-critical problems.

For enterprise CTOs and government leaders, understanding this transition helps you evaluate startup vendors and partners more effectively. A Series A startup is fundamentally different from a seed-stage startup. They've invested in compliance, security, and scalability. They've proven product-market fit. They're ready to be a serious partner.

The key is knowing what to look for: security and compliance investment, technical depth, go-to-market maturity, and the ability to scale with your needs. If you're considering a startup for a critical use case, engaging a technical advisor or fractional CTO can help you navigate the evaluation and ensure you're making a sound decision.

Seed versus Series A is not just about funding—it's about maturity, capability, and readiness to scale. Understanding that difference is essential for everyone involved in the startup ecosystem.