Every company is scored against the same written criteria. This is the full list, how the score is worked out, and what it does and doesn’t decide.
The criteria
Each criterion is written down with what strong, adequate and weak look like, and the impressive-sounding claims that shouldn’t count, like a slide full of enterprise logos or a board of advisors who’ve never taken a call.
Founder & Team
Whether these are the right people for this problem, and whether they can build it.
Domain depth
How they found the problem, whether they know how their buyers actually buy, and what they know about the industry that only comes from working in it.
Founder-idea advantage
What edge this founder has on this exact problem, and whether it’s hard to copy or something anyone could have.
Execution evidence
How fast they ship, from idea to product to first customer, and how closely customer feedback drives the product.
Technical capability
Whether the technical work is done in-house, what the team has built before, who they’ve convinced to join, and whether they run the company on AI too.
Perseverance and grit
Whether they’ve pushed through hard things before, including moments when quitting would have made sense.
Coachability
Whether they know what they don’t know, name their biggest risks unprompted, and can point to decisions they changed on advice.
Market & Problem
Whether the problem is real, the market is big enough, and the timing is right.
Problem validation
Evidence the pain is real: customers who can describe it, trials that convert to paid, inbound demand and low churn.
Market size
A bottom-up count of real customers at realistic contract sizes. For AI companies, the labour spend they can replace, not only existing software spend.
Timing and adoption
Where buyers sit on the adoption curve, and whether sales conversations sound like “not ready” or “we’ve been looking for this”.
Market insight
Where the founder thinks the market is heading, why now, and whether that view comes from experience or from reports.
Category creation
The change it’s taking advantage of, like a new technology, regulation or behaviour, what it believes about the market that most people get wrong, and whether it changes how the category works or competes inside it.
Competition
What stops someone else doing the same thing, now and at scale.
Defensibility
What the product knows after 1,000 customers that a new entrant couldn’t copy, and how deeply it’s built into work the customer can’t afford to get wrong.
Product differentiation
Why customers say it’s better, not just cheaper, and whether the innovation is in the data and workflow or only in the model call.
Distribution
Whether the company can reach customers in a way that doesn’t depend on its AI.
Market dynamics
Reliance on any one model provider, how hard it is for customers to leave, and how many tools could copy the core idea within a year.
Moats at scale
The barriers it can build as it grows: regulation, distribution lock-in, platform effects, trust, and unit costs that fall with volume.
Traction & Validation
The evidence that customers want it and will pay for it.
Traction
Revenue, paying customers, gross profit, pipeline, customer concentration, and how much it cost to get there.
Product-market fit
Retention, usage, and whether customers sound enthusiastic or just polite.
Buyer value
The measurable results customers report, how fast they arrive, and whether the return on investment holds up when the customer’s finance team checks it.
Revenue expansion
Whether existing customers spend more over time, and whether pricing grows with their usage.
Go-to-market
A specific ideal customer, the first 10 to 20 named targets, acquisition cost against lifetime value, and which channels repeat.
Revenue model
How the company charges, what each new customer costs to serve (AI inference included), and how predictable the revenue is.
AI Depth
Whether AI is the product or a feature on top of someone else’s model.
Data advantage
Proprietary data that gets better with use and can’t be replicated.
Model dependency
How much the company relies on one foundation model, and whether it survives a model change.
Position against foundation models
Whether it builds where foundation models won’t go, or competes with what they’re getting better at.
Workflow integration
How deeply the product sits in the customer’s work, and what removing it would take.
Platform ambition
Whether it stands alone or rides on another platform, and how a first product grows into a platform within 18 months.
Return profile
Whether this deal, at this price, can return the fund.
Fund return potential
The three to five things that must be true for this company to return the fund, and whether the entry price makes sense for its stage and traction.
Ownership and dilution
What our stake looks like after the Series A, the Series B and at exit.
Exit paths
Comparable acquisitions and listings in the space and region, with best, base and downside cases.
How the score is worked out
Evidence comes first
Doc Intelligence reads the deck and site, then founder, market and competitor research run from outside sources. Every claim is tagged verified, claimed or unverified, with its source.
The gate notes warning signs
The pre-screening gate checks the deal against our mandate and notes warning signs, like a solo founder, no revenue yet, a crowded market or a valuation ahead of the stage. A warning sign doesn’t end a deal. It becomes a question for the founder.
Synthesis A rates every criterion
Each criterion is rated strong, adequate or weak against what we’ve written down, with the evidence for the rating, and grouped by category. The five most similar past deals come along as comparables.
Synthesis B and the committee test it
A second pass checks every claim in the draft against its source. Then the AI committee argues the deal from four opposed positions and records where they agree and where they don’t.
Code works out the score
The ratings become a score from 1.0 to 3.0. A set amount comes off for each warning sign the evidence didn’t clear up, and the result maps to proceed, pass or more info. The model doesn’t do the arithmetic.
Gaps are named
What couldn’t be verified is listed in the memo instead of filled in. A model forced to answer is how hallucinations happen, so it’s told to return a gap.
People decide
A partner decides whether the deal proceeds to a discovery call, and confirms every recommended pass. The investment committee decides whether we invest.
What the score decides, and what it doesn’t
It decides
Where the company sits against comparable past deals.
The recommendation a partner reads: proceed, pass or more info.
The questions for the next meeting, and what to check in due diligence.
It doesn’t decide
Whether a deal proceeds to a discovery call. A partner decides.
Whether a deal is declined. A partner confirms every pass.
Whether we invest. The investment committee does.
The terms, or what a founder is told. A person writes every reason.
Where the criteria come from
We ran an accelerator for years, first with the Melbourne Accelerator Program and then as our own program. For every cohort we had a long application from everyone who applied, monthly reporting from everyone we took, and our own people working alongside them. We know what those companies looked like going in and what happened to them afterwards. The criteria are built on that.
Shahirah Gardner, one of our Venture Partners, wrote up hundreds of past deals by hand the way we wanted the AI to read them, then marked every line where the AI’s version differed. That hand-marked set is what every change is tested against.
The criteria and the screening rules live in two Notion databases that partners edit. They’re copied into Supabase, where the AI reads them on every run, and a rule only reaches the AI once it’s marked Live.