Every angel investor has had this moment. You agree a pre-money valuation. You sign the paperwork. And you know, somewhere behind the eyes, that the number wasn't calculated.
It was negotiated. The framework gave you cover. The framework was the negotiation.
You probably wouldn’t say so over dinner. You might say it after a second glass of wine, to another angel. There’s a slight relief in saying it out loud, because the other person nods. They’ve had the same moment. Every method in the room knows it’s a method, not a calculation. The methods don’t pretend otherwise. The people using them sometimes do.
No confession in that. It’s the job. Pre-revenue companies have no performance data to value on multiples, so something has to fill the gap. For thirty years, smart people have built increasingly sophisticated somethings. A paragraph on each is worth the time.
1996
Dave Berkus, a California angel, kept being asked to value companies with no revenue, no customers, and sometimes no finished product. He proposed assigning monetary value to the things you can assess. A working prototype is worth something. A credible team is worth something. A clear go-to-market plan is worth something. Add the components, cap the total, and you have a pre-money valuation.
The Berkus Method gave structure to instinct. It’s still in widespread use today, nearly three decades later.
Berkus himself was honest about what he was doing. The method was designed for pre-revenue companies because post-revenue companies had better tools available. He was pricing the absence of evidence, and he knew it.
2001
Bill Payne’s Scorecard Method compared startups against regional averages, with explicit weightings (team 30%, market 25%, product 15%, and so on). It introduced useful discipline. Systematic comparison rather than freeform judgement.
The benchmark, though, reflects what other investors paid in previous deals, which were themselves structured guesses. Averaging prior guesses to calibrate the current one.
Through the 2000s and 2010s
Risk Factor Summation lists every material risk and scores each from -2 to +2. It produces a spreadsheet. It feels like analysis. A checklist of opinions wearing a numerical jacket.
Comparable Transactions asks what similar companies raised at. Two SaaS companies at seed can present identical surface characteristics and follow wildly different trajectories. The data tells you what investors paid before. It says nothing about whether they were right.
Monte Carlo simulations apply probability weightings to revenue scenarios. The maths is genuinely complex. The outputs carry decimal places. The inputs are projections produced by founders whose incentives run entirely in one direction. Astronomy dressed as physics.
Each method added structure. None of them added evidence. The methods got more elaborate; the data didn’t get any closer.
2025: same telescope, sharper lens
A new generation of AI-augmented diligence platforms now promises to solve early-stage valuation through machine learning. They benchmark startups against millions of others, score founding teams against historical patterns, parse pitch decks with natural-language processing, size markets by scraping every available trace of online activity. One VC firm has reported a sixty per cent reduction in due diligence time.
This is genuinely useful. The speed gains are real. The bias-reduction case is at least plausible. An algorithm doesn’t care about a founder’s accent.
But look at what the AI is processing. LinkedIn profiles, GitHub history, web traffic, pitch-deck language, hiring signals, founder demographics. Proxies for performance. Processed faster, cheaper, more consistently than human assessment alone. Still proxies.
The faster the telescope, the more obvious the fog. AI has done remarkable work on the workaround. It hasn’t touched the data problem the workaround was built around.
The instrument that admits it
There’s a second response to the valuation problem that deserves credit, because it doesn’t pretend.
Convertible notes, SAFEs and Advance Subscription Agreements all defer the valuation decision. The investor’s money converts to equity later, at a discount to whatever price the next priced round establishes. A cap protects the investor if the next round goes to the moon.
The instrument is doing something honest. It’s saying: nobody can price this company today, so let’s not pretend. Let’s wait until someone with more information sets the number.
Watch what happens when angels negotiate a cap, though. Everyone in the room knows the cap is itself a guess. The guess gets dressed in numbers, anchored to recent comparables, defended with conviction. The instrument was invented to dodge the valuation theatre. It produced a smaller, more compressed version of the same theatre, performed under different stage lights. The cap is the valuation in costume.
That’s clean as far as it goes. It’s also deferral. The valuation problem doesn’t disappear; it lands six months later on a Series A investor’s desk, where the same problem exists with slightly more data points.
There’s nothing wrong with deferral when measurement is impossible. The question worth asking is whether measurement is actually impossible, or whether the industry has settled into the assumption too comfortably.
The thing you already do without noticing
Here’s the moment worth sitting with. Every angel syndicate already knows execution matters more than presentation. They prove it every time they make a follow-on decision.
When you’re deciding whether to put more money into a portfolio company, the pitch deck is irrelevant. You look at revenue trajectory, customer retention, gross margin, unit economics, hiring momentum. You look at how the founder has actually executed. The conversation has a completely different texture from a first-meeting deck-and-coffee. You’re calmer. The numbers are doing work the founder used to have to do with charisma.
The texture is different because the information is different. Not because you’ve become a better investor in the six months since the first cheque. The data exists now, and you trust data over performance theatre.
Sit with that for a moment. The same investor, looking at the same company, with the same founder, in the same market, has more conviction in month seven than in month one. The conviction tracks the information, not the relationship. And the information that flips conviction (the revenue trajectory, the retention curve, the unit economics moving in the right direction) is information you couldn’t have demanded at month one because the data didn’t exist yet.
The structural question is whether the data needs to be absent at first cheque. For thirty years, the answer was yes. The methods Berkus, Payne and the rest invented were the entire response to that yes. Convertibles route around the same yes by deferring the question.
The yes is becoming a no. That’s the change worth tracking.
What it costs to get the number wrong
There’s a conversation that happens in every angel network, six months after a deal closes. Someone admits, quietly, that they probably paid too much. The company is performing fine; the valuation just doesn’t make sense in retrospect. Nobody knew at the time. The framework gave everyone cover. The number is locked in.
Pay too high at first cheque and your exit multiple drops in proportion. Pay fifty per cent over the right number and your eventual return shrinks by roughly a third, before any other variable kicks in. Worse, the price you paid becomes the reference point that haunts every subsequent round. The founder may be unable to raise at a flat price even when the business is delivering, because the previous mark was indefensible. Companies get killed by entry overvaluation more often than the post-mortems admit.
Pay too low and the cost lands on the founder, then back on you. The angel wins on entry multiple and loses on founder motivation. UK SEIS deals have a documented pattern of founder disengagement when early dilution leaves them owning too little of their own company by the time they reach Series A. The headline multiple looks good in the spreadsheet. The exit that would justify it never arrives because the person you were betting on stopped showing up the way they used to.
The methods of the 1996-2025 arc don’t help you avoid either failure mode because the number they produce isn’t falsifiable. There’s no point in the company’s future where you find out the Berkus calculation was off by 50%, or the Scorecard regional average was wrong, or the Monte Carlo probability weighting was misspecified. The number arrives, the deal closes, and the feedback loop never closes. You discover the answer at exit, which is the one point in the whole arc where the discovery is useless.
A year of verified revenue narrows the plausible range. You can still be wrong. You can’t be wrong by a factor of three, in either direction, and not know it. The asymmetric downside that the structured guess has been quietly carrying for thirty years gets smaller.
What 52 weeks of revenue looks like, from the Prize Provider's chair
An Investment Prize Provider in The Startup Race signs a Commitment Letter for £5,000 (more, within reason, if they want a larger position) toward a £100,000 SEIS-eligible prize that will fund the winner of a fifty-two-week revenue competition between fifty founders. The capital stays in the investor’s own account. No money moves until April 2027, after the race ends and the winner is identified. The Commitment Letter is the mechanism; the watching is the substance.
What you do, in the year between the signature and the cheque, is watch. The leaderboard updates every week. Revenue is pulled from each founder’s bank account via open banking. The cohort can be sorted by trajectory, by absolute revenue, by week-on-week growth, by retention. The week a founder loses a major customer is visible; so are the next four weeks of how she responds. The month another founder’s churn spikes shows up alongside the product changes that follow. The founders who gain ground when conditions tighten separate from the ones who don’t.
By week 52, the winner is the founder who delivered the trajectory you’ve been watching. The investment decision was made, in effect, in slow motion across a year. The founder didn’t pitch you. The methodology pitched her.
When you arrive at the valuation conversation, the texture is different. You’re not pricing intangibles or comparing against regional averages. You’re pricing a year of verified performance. The Berkus components (team, market, product, go-to-market) are visible in the evidence, not asserted in the deck. The team is the team that delivered the trajectory. The market is the market that produced the customers. The product is the product the customers kept paying for.
This isn’t a Series A valuation framework. The absolute numbers are smaller. The customer base is small. The market may not be tested at scale. The methodology can’t predict outliers and won’t suit deep-tech where revenue arrives late by design.
But for the substantial portion of the SEIS-eligible market that can generate verified revenue at small scale (most of it, in practice), the methods Berkus invented in 1996 are no longer the only option.
The evidence that this matters
Wiltbank’s Kauffman Foundation work on three thousand angel investments found that angels who spent more than twenty hours of due diligence achieved deal-level returns 5.9 times higher than those who spent less. The same research found a 65% failure rate on under-diligenced deals, against far fewer write-offs on the well-diligenced cohort. Colin Mason’s thirty years of UK research has reached compatible conclusions.
Information at the point of investment is among the most predictive variables in angel portfolio outcomes. Twenty hours of structured diligence and fifty-two weeks of verified revenue are different shapes of information. They share a direction of travel: a richer picture at the moment the cheque is signed.
There’s a quartile gap of roughly £340,000 between top-quartile and bottom-quartile angel portfolios on the same starting capital over a ten-year horizon. Most of that gap is variance reduction (fewer zeros, not higher peaks). The portfolios that produce top-quartile returns are built on better information at the moment of commitment.
What we're betting
Thirty years ago Berkus built a workaround for missing data. Thirty years later AI tools are working the same workaround at speed. The trajectory of the industry has been toward better proxies. The trajectory was the right one when no alternative existed.
There’s an alternative now. Open banking verifies revenue at source. Accounting integrations expose unit economics in real time. The infrastructure for measuring actual performance at small scale is in place and improving. Most of what gets called early-stage investment is happening in companies where this infrastructure works.
The Startup Race is built on the assumption that for those companies, fifty-two weeks of verified revenue is a better instrument than thirty years of refined guesswork.
We’ve tested the assumption in stages. The £1,000 race (Edinburgh, 2020, ten weeks) was the proof that a public revenue league table changes founder behaviour at all; benchmarking was manual, the cohort small, and the winner grew revenue 12X over the duration. The £10,000 race (2022, six months) was the test of whether weekly revenue verification could be automated at scale via open banking; it could, the winner grew 44X, the runner-up 17X. The £100,000 race (2026/27, fifty-two weeks, fifty founders) is the test of whether a substantially larger prize sustains 20X revenue growth across a full year, and whether the resulting cohort produces an investable winner against the angel selection criteria we’ve described above.
Each stage has tested one variable the last didn’t. We may still be wrong. The data will tell us.
The version of the future I find most plausible looks like this. Berkus’s method, Payne’s scorecard, and the AI tools that automate them continue to do their job at the genuinely-unpriceable end of the market. Deep tech, hard science, the founders whose value really does exist primarily as conviction. Convertibles continue to do their job alongside them.
The middle moves. The substantial part of the SEIS-eligible market where customers exist and revenue can be measured stops using methods designed for the absence of evidence, because the evidence is no longer absent. The valuation conversation in that middle becomes one where the framework gave you cover for a number that was, this time, actually calculated.
That’s the moment worth heading toward. The cover wasn’t shameful (it was the best available solution to a real problem) but it stopped being necessary somewhere around the time open banking went mainstream, and the industry hasn’t quite caught up.
Berkus solved a real problem with the data available to him. The data has changed. Most of the methods haven’t.
James Shoemark is CEO of The Startup Race. Previous essay in this series: Revenue is the Brightest Star. But Investors Navigate by Constellations. The Race is open for applications for the 2026 cohort at thestartuprace.com.