The Five-Person Venture Is Harder to Staff, Not Easier
If You Only Have a Minute
- Venture teams really have shrunk. The average Series A company raising in 2025 employed 14 full-time people, down from 22 in 2022. The capital cycle explains part of that and AI explains part of it, and the data does not separate them cleanly.
- Our own team playbook is out of date, and we wrote it. Team 01 prescribed two founders plus six functional roles. That lands near a dozen people. We would not write it that way today.
- Smaller is not easier. The scarce capability in a five-person venture is not building. It is being able to tell that plausible-looking output is wrong, in this domain, before it reaches a customer.
- This is a worse problem for corporates than for startups. 77% of companies staff business building from people they already have, and peer-reviewed research finds corporate venture units staffed mainly with internal hires were more likely to be shut down.
- The strange part is that corporates already know this. Three quarters of corporate venture capital units recruit their investors externally. The same parent company that would never staff its investing arm with whoever was free will staff its venture build exactly that way.
Three years ago we published a piece describing the team you need to start a venture. It prescribed two founders, then product, engineering, sales and marketing, operations, finance and HR. Counted honestly, that is ten to twelve people before a venture has a customer.
We would not write it that way today.
The numbers have moved underneath it. Carta, which administers equity for more than 45,000 US startups, reports that the average Series A company raising in 2025 employed 14 full-time people, down from 22 in 2022. The same dataset shows startups adding 31,000 net headcount since the start of 2023, against 865,000 across the four years before it. Something structural happened to the size of a young company.
Carta’s Peter Walker reads the sequence in stages: interest rates turned the market in 2022, capital discipline held hiring down through 2023 and 2024, and the flat line since mid-2024 he attributes to AI, asking whether startups are automating roles they would once have hired for and answering “I think yes.” We would put slightly less weight on that last step than he does. The freeze began before AI coding tools were deployed at any scale, and two forces arriving within eighteen months of each other are difficult to separate cleanly. But the direction is not in dispute.
For corporate venturing it raises a question we have not addressed in three years of writing about governance, capital and speed. If the venture that used to need twelve people now needs five, who are those five, and can a corporate actually get them?
The answer to the second half is the uncomfortable one.
A note on the number before we go further. There is no dataset anywhere that reports the launch headcount of corporate ventures. “Five” here is a way of describing a shape, not a benchmark we can cite, and we would rather say so than imply a precision that does not exist.
The Team We Told You to Hire
Team 01 was not wrong when we published it. It described a world where building took quarters, where a working product required a squad of engineers sustained over months, and where the supporting functions had to exist early because someone had to run payroll and close the books while the engineers built.
Most of that has held. One part has not.
The functions we told you to staff first are the ones where the work has compressed most: analytical work, documentation, first-draft design, boilerplate code, market sizing, competitive landscaping. The roles did not disappear. The boundaries between them dissolved, and the number of people required to cover them fell.
Meanwhile the parts of Team 01 that have aged best are the parts about mindset rather than headcount: founder mentality, coachability, leading through doing. Those were the paragraphs that read as soft when we wrote them. They are now the operative ones.
The two-founder model has also held, and arguably strengthened. A commercial founder and a product founder with complementary skills was a good structure when it broadened the talent pool and split the workload. It is a better structure now, because in a team of five there is no management layer to catch a founder’s blind spot. The check has to sit inside the founding pair or it does not exist.
What we would cut is the assumption underneath the rest of the list: that each function needs a person on day one. Operations and finance in particular were specified as headcount because there was no other way to get the work done. There is now.
Corporates Staff Ventures With Whoever Is Free
Corporate venturing has a problem here that startups do not. A startup founder assembling five people hires from the whole market and is judged on the result. A corporate assembling five people for a new venture reaches, overwhelmingly, into the building it already occupies.
BCG surveyed 1,051 senior leaders across 17 countries about how they build new businesses. 77% said they rely on their internal staff for business building. About 65% said they already have the internal talent they need and merely needed to reassign them. Only 54% said they actively recruit entrepreneurially skilled people. BCG sells business building services and framed the survey to show a gap it fills, and we should say so plainly. But these are corporates describing their own practice rather than a consultant’s characterisation of it.
That is the secondment default. The venture gets the people who were available, which is a different selection criterion from the people who were best, and in a large organisation availability correlates with things no one wants to say out loud.
Be specific about who gets reached for. In a large corporate the available and entrepreneurial-seeming people tend to sit in strategy, and they tend to hold MBAs. Ilya Strebulaev’s team at Stanford’s Venture Capital Initiative has assembled a database of 4,021 unicorns and 6,729 founders, with education records hand-collected for 4,487 of them, and reports that 15.8% of US unicorn founders hold an MBA, against 15.0% of founders outside the US. Whatever else the degree does, it is not the modal credential of someone who builds a billion-dollar company. Roughly five in six did not hold one.
That should not be overread. Unicorn founders are a survivor sample and the data says nothing about who failed, education coverage runs to only 42% for non-US founders and 22% for Chinese founders, and this is a researcher publishing his own dataset in the trade press rather than a peer-reviewed result. But the direction is a useful corrective for anyone assembling a venture team off a corporate bench, because that bench is selected for precisely the credential the data suggests is not the one doing the work.
There is peer-reviewed evidence that this is the wrong default, and it is the most useful finding we have come across in a while.
Vibha Gaba at INSEAD and Gina Dokko at UC Davis studied 70 US corporate venture capital units across sixteen years, tracking which survived and which were abandoned. They hypothesised that units staffed with high levels of internal hires would be less likely to be abandoned, on the reasoning that internal staffing embeds a unit in its parent and protects it. The data said the opposite. Units with above-median numbers of internally hired managers were more likely to be abandoned, and the hypothesis was not supported. Units with at least one manager carrying independent venture capital experience were less likely to be abandoned.
We should be careful about what this proves. It studies investing units rather than venture building teams, the data runs from 1992 to 2008, and the sample is US technology firms. It says internal staffing correlates with the unit being abandoned. It does not say seconded people build worse products. But the mechanism the authors describe, that units staffed from within absorb the parent’s assumptions and lose the outside knowledge that justified their existence, is not specific to investing, and anyone who has watched a corporate venture slowly turn back into a department will recognise it.
Corporates already know how to do this properly. They just do it somewhere else in the building. Global Corporate Venturing’s benchmarking of nearly 400 corporate venture capital units found that 75% recruit from institutional investors or other corporates, and 63% report average relevant investing experience above five years. When a corporate sets up an arm to write cheques into startups, it goes and hires people who have written cheques into startups.
The same parent company will then staff a venture build with three people from strategy who were between projects.
We are not aware of a good justification for the asymmetry. The most likely explanation is simply that investing looks like a specialist skill and building looks like general management, which is a category error a great many corporate ventures have paid for.
What Actually Changes the Shape of a Team
The intuitive case for a smaller team is that AI does the work five people used to do, so you need fewer people. The evidence does not support that as stated, and the way it fails is instructive.
The most careful measurement available is from the US Census Bureau, whose nationally representative study of AI diffusion, drawing on data collected between November 2025 and January 2026, found that the breadth of AI integration across business functions and the level of operational investment are both positively associated with employment decreases, while worker-level AI use shows no significant link to headcount reduction once those two are accounted for. The same study found AI-related employment decreases are rare, occurring in only 2% of firms.
Those two findings belong together. Handing people tools changes almost nothing. Restructuring how work moves through an organisation, and investing behind that restructuring, is what changes the number of people. And it is still uncommon.
McKinsey’s 2026 State of AI, surveying 1,719 respondents across 97 countries, found the same split in the returns. Among its AI high performers, defined as the 6% of respondents who attribute at least 5% of EBIT to AI and describe the value as significant, nearly three quarters report fundamentally redesigning workflows because of their AI use, against roughly one quarter of other respondents.
For anyone planning a venture team, that is where the argument lands. A five-person venture is a workflow redesign. It is not a twelve-person plan with seven names removed, and a corporate that treats it as a headcount exercise will get the headcount and none of the result.
The framing that has become popular in engineering circles, and which we think is directionally right without being evidence, comes from Google’s DORA programme: “AI’s primary role in software development is that of an amplifier. It magnifies the strengths of high-performing organizations and the dysfunctions of struggling ones.” That is a conclusion DORA asserts rather than tests, in a self-reported cross-sectional survey published by a company that sells AI developer tools. Their 2025 data does show AI adoption associated with higher delivery throughput, reversing their own 2024 finding, and still associated with higher delivery instability. Take the amplifier line as a practitioner’s summary that matches what most people observe, not as a measurement.
The Scarce Skill Is Knowing When the Output Is Wrong
If tooling now covers much of the building, the obvious conclusion is that a small team can be staffed with less experience. The evidence points the other way, and this is the part of the argument we would most want a reader to take away.
The best-identified study in the field is Brynjolfsson, Li and Raymond’s, published in the Quarterly Journal of Economics, which covered 5,172 customer support agents and roughly three million conversations. Access to AI assistance raised issues resolved per hour by 15% on average and by 36% for the lowest-skilled quintile. The most experienced and highest-skilled workers saw small gains in speed and small declines in quality.
That result is usually reported as AI levelling a workforce, and half of it is. On well-specified work with a known target, AI genuinely closes much of the distance between a novice and a veteran. A new analyst sizing a market, an engineer writing a specified function: the tooling lifts them.
The other half is less comfortable and gets quoted less. The experts did not merely fail to gain. Their output quality declined slightly. Whatever AI is doing, it is not simply adding capability to everyone.
At the edge of what the tools can do, the picture changes again. A preregistered randomised trial of 758 consultants, published in Organization Science, found that on tasks inside AI’s current capability participants completed 12.2% more work 25.1% more quickly at significantly improved quality, while on a task outside that capability they were 19% less likely to produce correct solutions than colleagues working without AI. The study was run with BCG on BCG’s own consultants, a conflict that preregistration and peer review reduce without removing.
The boundary between those two regimes is not marked. That is the study’s central point: the frontier is jagged, a task that looks like the last one can sit on the other side of it, and nothing warns you when it does.
And the work of catching that does not disappear, it moves. Research by BetterUp Labs with Stanford’s Social Media Lab found 40% of surveyed US desk workers had received AI-generated work of inadequate quality in the previous month, costing around two hours of rework each time. BetterUp sells coaching and the cost extrapolations are self-reported, so treat the two hours as indicative. The structural point holds: generation became cheap, review did not.
Taken together, those results point to a staffing rule close to the opposite of the intuitive one. The binding capability is not the ability to produce work. It is the ability to look at confident, well-formatted, plausible output and know that it is wrong in this particular market, for this particular customer, this time. This is narrower than general judgement. It is domain execution experience, the specific kind that comes from having done the work before, and it is exactly the input a small team has least of.
In a twelve-person venture, a wrong call met a product lead, a finance review and an investment committee. In a five-person venture it meets the market.
The Corporate Pay Problem
Suppose you accept all of this and set out to hire five people with that kind of depth into a corporate venture. Two obstacles are waiting, and both are structural rather than a matter of effort.
The first is compensation. McKinsey surveyed 1,176 senior managers and C-suite executives between May and July 2024 on corporate venture building, and found that one quarter of expert builders have the flexibility to compensate talent more highly than the core business can, compared with just 13% of novice builders. Read that the right way round. Even among corporates that have built repeatedly and know what they are doing, three quarters cannot pay a venture team differently from the parent’s grade structure.
The second is upside. Global Corporate Venturing’s research on corporate venture building programmes found that only 22% give equity to their founders. No sample size is disclosed and it is a trade publication’s own survey rather than a census, so treat it as indicative. But it matches what we see. Most corporate ventures ask people to take startup risk on corporate terms, with a salary band set by a job architecture designed for a different kind of work and no share of what they create.
Get Wright Partners’s stories in your inbox
Join Medium for free to get updates from this writer.
Team 01 made the cash-versus-equity trade-off a central point: founders and early teams routinely accept below-market cash for meaningful equity, and that trade is what lets a young company hire above its weight. A corporate venture that offers neither a premium nor equity is not making a trade-off. It is asking for a favour.
This was survivable at twelve people, where a few strong hires could carry a mixed team. At five it is not. The compensation question and the team size question are the same question, and most corporates are still answering them in separate meetings.
The practitioner language on this is blunter than ours. Jasdeep Sawhney, who runs Jaguar Land Rover’s Inmotion Venture Studios, describes internal people brought into a venture as “not business builders, they’re subject matter experts”, which is why they recruit venture founders externally. Both are valuable. They are not interchangeable.
There is a real tension here, and it complicates our own argument. The domain depth we just said was the scarce input is often sitting inside the corporate. The seconded subject-matter expert knows the customer, the regulation and the failure modes. What they usually lack is the experience of building something from nothing, and the two together are rare enough that most corporates will have to hire for one and second the other, deliberately, rather than assume one person carries both.
What This Looks Like in Southeast Asia
The regional picture makes the argument sharper rather than softer, because Southeast Asian corporates are not behind on adoption. They are ahead, and it is not converting.
McKinsey, working with Singapore’s Economic Development Board and Tech in Asia, surveyed 330 senior executives across six Southeast Asian economies. 46% of the companies surveyed had moved beyond piloting to scaling or fully scaled AI, against a 35% global average, with Singapore at 56% and Indonesia at 51%. The sample covers companies already using AI and skews to the digitally advanced, and both McKinsey and EDB have an interest in the region looking capable, so read the level with caution.
The next number in that study is the one that matters: 60% reported AI contributing less than 5% of EBIT, and 18% reported no financial impact at all.
Asked to name the biggest challenge in scaling AI, one in five executives named a lack of internal expertise and talent, the single largest answer. Not capital, and not a reluctant board.
That is the region’s own executives saying the constraint is people. It is the same finding as our June piece, arrived at from the opposite end: the budget exists and the capability to use it does not.
For context on the level rather than the direction, Bain puts the share of companies globally that have meaningfully scaled generative AI at under 20%, against McKinsey’s 35% global figure, and argues that transformation is harder in Southeast Asia because of market fragmentation. Bain discloses no sample or collection dates. The two are not reconcilable as stated, which tells you something in itself: there is no stable definition of what “scaled” means, and any single figure in this space should be read as one publisher’s construct.
The Indonesian evidence gets closest to the mechanism. Research commissioned by AWS from Strand Partners, surveying 1,000 business leaders, found that among Indonesian firms already using AI, 21% of large enterprises deliver a new AI-driven product or service, against 34% of startups building entirely new AI-driven products. The two measures are worded slightly differently and AWS sells the fix, so this is indicative rather than decisive. But both groups are drawn from firms already using AI, with access to the same models at the same prices. The gap is not in the technology. It is in what surrounds it, which is the argument this series has been making since June.
The Asian Development Bank, interviewing across the Indonesian startup ecosystem, recorded the consequence: “unicorns (and large traditional companies) seek to hire the same talent as startups, and they have a hard time competing for engineers, programmers, designers, and others”. It is a qualitative finding from a small sample and should not carry a number.
One finding in Strebulaev’s data will get quoted in this region, and it needs careful handling. Ranked by founders of non-US unicorns, INSEAD comes first, ahead of Harvard and Stanford. The margin is 14 founders to 13. That is a one-founder gap in hand-collected records covering fewer than half of the relevant founders, and the correct reading is that the three are indistinguishable rather than that one leads. The more durable finding sitting next to it is that non-US founders are more than four times as likely to hold a business master’s that is not an MBA, and that among business-educated founders 87% of Americans hold MBAs against 59% internationally. The credential that marks someone as a builder in Jakarta or Singapore is not the credential that marks them in Palo Alto. A corporate screening its own bench for the American shape of it will filter out people it needs.
One thing has moved in the corporates’ favour, and few have acted on it. Singapore’s Ministry of Manpower reported information and communications sector employment falling by 5,100 across 2025, with 1,930 retrenchments in the sector, and professional, manager and executive retrenchment incidence at 10.1 per thousand against a 2015 to 2019 average of 8.0. Senior operators with real platform experience are more available now than at any point since the boom. The objection that a corporate cannot hire this calibre of person was more true three years ago than it is today.
The Five Roles
With the caveat already given, that five is a shape rather than a measured benchmark, here is where we would now put the headcount.
1. A commercial founder. Unchanged from Team 01 and, if anything, more important. This is the person who owns the customer relationship, the pricing conversation and the distribution deal. None of it compresses. All of it is what the corporate parent is uniquely positioned to supply and uniquely bad at handing over.
2. A product founder. Also unchanged. In a team of five this person is doing what a product manager, a designer and a technical architect did three years ago, which is now possible and was not.
3. Two builders, at least one of whom has built in this domain before. This is where the compression is real and where Team 01’s “at least one squad” no longer holds. It is also the role most often misjudged now, because the tooling makes inexperience invisible until it is expensive. The evidence says AI lifts capable people fastest on well-specified work and degrades output at the edges without signalling it. You need at least one person who has seen this category of product fail before and recognises the shape of it.
4. One person who owns the operating reality. Not an operations manager in the Team 01 sense. The person who owns whatever in this specific venture cannot be automated and will kill it if neglected: regulatory approval, supply relationships, merchant onboarding, clinical compliance. This role is venture-specific by definition and is the one most often left until it is urgent.
Finance, HR, marketing operations and the rest of Team 01’s supporting cast do not need day-one headcount. They need a service arrangement with the parent and a founder who knows when to escalate. This is the one place where being a corporate venture rather than a startup is an unambiguous advantage, and most corporates use it to insert reporting requirements instead of removing work.
What is missing from that list is a layer. There is no one whose job is to review, approve or coordinate. In a five-person venture the governance is external, which is what our August piece was about, and the internal structure has to be flat enough that a decision costs a conversation.
Where This Stops Being True
Any Southeast Asian reader will already be thinking of the counter-example. Grab reached its first full year of net profit in 2025, with revenue of US$3.37 billion, up 20% year on year, and profit for the year of US$200 million, a US$358 million improvement. It did not get there by running a five-person team, and the region’s most closely watched path to profitability has not been a story about radical headcount compression.
That is not a contradiction of this argument, but the distinction has to be made rather than assumed. This is a claim about the founding team of a new venture, in the period before product-market fit, where every person is a large fraction of total capability. It is not a claim about steady-state headcount in a scaled platform operating millions of transactions. Confusing the two is how a good argument about starting becomes a bad argument about running.
The second limit is more practical. Five people works when the venture’s hardest problem is finding a customer and shipping something they will pay for. It stops working the moment the hardest problem becomes throughput: onboarding at volume, operating a fleet, clearing regulatory approval in four markets at once. Those are headcount problems and no tooling has made them otherwise. The five-person shape is a starting configuration, not a target operating model, and a corporate that mistakes it for the latter will strangle a venture that was working.
What This Means for Corporates
The three pieces before this one described an environment. AI made building cheap, capital discipline became the price of entry, and decision latency became the binding constraint.
This one is about the thing all of that acts on. If the venture is now five people, the corporate’s advantage has moved. It is no longer capital, which is abundant for anyone with a credible plan. It is no longer capacity, which the tooling supplies. It is the ability to put five genuinely excellent people against a problem where the corporate holds distribution, customers and data that no startup can rent.
That is a real and defensible advantage. Almost nothing about a standard corporate operating model is set up to deliver it. The grade structure prevents the pay and the absence of equity prevents the trade, while the secondment habit fills the seats before anyone asks who should be in them. And the research says that when corporates staff this way, the unit is more likely to be shut down.
The fix is not complicated, though it is uncomfortable. Hire from outside for the building experience the inside does not have, which will usually be most of it, and second from inside for the domain depth that only the parent holds. Pay them outside the grade structure, or accept that you are recruiting from a smaller pool than your competitors. Give them equity in the venture, because a five-person team with no upside is a project team with a different name. And apply the discipline you already apply to your investing arm, which hires specialists and would never dream of doing otherwise.
Corporates that manage this build ventures at a cost that would have been implausible in 2022, against startups that cannot match their distribution. Corporates that do not will run the same five-person experiment with the five people who happened to be free, watch it fail, and conclude that venture building does not work.
Corporate venture building still works. It has simply stopped forgiving a team assembled out of convenience.
Author
Ziv Ragowsky, Founding Partner at Wright Partners
Sources
17. Singapore Ministry of Manpower, “Labour Market Report Fourth Quarter 2025”
18. Grab Holdings, fourth quarter and full year 2025 earnings release, 12 February 2026
