Every target now has an AI narrative. Almost none of them have been diligenced. We score the AI a company actually runs — how it operates, what it sells, and whether its market survives — across eight weighted domains, and convert every finding into a deal term: a number, a verdict, a decision.
Five diligence scopes are already owned — Commercial, Financial, Operational, Technology, Legal. AI is the white space, and it moves the return more than any of them. This is the instrument: the gap sits at the center, and the sweep runs it gate by gate, screen to exit.
A company can run brilliantly on AI, sell a product whose AI claims collapse under scrutiny, and sit in a market about to be disrupted. Three surfaces, each re-pricing the deal differently — all underpinned by one: their financial impact.
The AI inside the business — a margin and efficiency lever, or a fragile dependency with no fallback.
The AI in the offering — defensible, model-backed growth, or claims outrunning the model.
The AI reshaping the category over the hold — positioned to ride it, or disintermediated before exit.
The through-line. Operational efficiency, product growth, and market position all resolve here — in EBITDA, cash, and enterprise value. This is the surface where AI capability and AI risk become money.
Named AI failure modes — the ones a generalist workstream misses because it isn't looking for them. Each maps to one of the fourteen red-flag triggers.
not in the product
no moat over a model
one provider, no fallback
can't prove ownership
ungoverned, invisible
EU AI Act unmapped
wrong answer is costly
capability that can walk
One flow, three moves. Eight weighted domains build the composite (S = Σ wᵢ·sᵢ), with 14 red-flag triggers that can cap it; the banded scale places 2.4 against the 3.6 sector baseline; and provenance proves it — every number drilling to its source.
The value engine translates AI capability into six drivers of enterprise value, so a finding lands as a number — and that number lands in the deal as price, structure, or a walk. An Impairment finding is the overpayment you don't make.
The same finding that costs you at entry can pay you over the hold. An AI-specific fix does two things: it remediates the value at risk, and it creates new value on top — margin, revenue, operating leverage. We price both, and hand you the value-creation play.
We assess the AI dimension of all five diligence workstreams across all four surfaces — nothing sampled, nothing skipped. Where you already run a workstream, we plug into it; the shading marks where each surface carries the most weight.
| AI surface ↓ Workstream → | Commercial | Financial | Operational | Technology | Legal |
|---|---|---|---|---|---|
| Operational AI | |||||
| Product AI | |||||
| Market AI | |||||
| Financial impact the base |
Delivered as three AI-native applications on the Diligence360 engine. Filter to your thesis, drill from a score into the evidence, and move from the whole fund down to a single model.
The deep interactive read — surfaces, value, risk, evidence. Ask Copilot anything.
Open the app →The one-glance verdict for a fast pre-LOI go / no-go.
Open the app →Every holding on one panel, benchmarked on a common coordinate system.
Open the app →Diligence usually means shipping a target's data out to whoever is analyzing it. We invert that: the engine crosses into the target's environment, works in an isolated clean room, and only governed findings ever come back — sized to the confirmatory window.
The engine and the questions. Models and analysis are sent to the data — sized to the confirmatory window.
Findings, scores, evidence references. Governed results only — each one traceable back to its source.
Raw data, PII, model training. The target's data is never copied out and never used to train any model.
Trace a single finding from a surface to a signed term — and to $6.2M of value at risk. Then set that one number against the fee: the engagement is sized to your window and priced against the overpayment it prevents, so it typically returns on a single finding.
Three parties, one deal, one evidence base — each entering at a different gate with a different question.
Buy the depth the gate calls for. Screen a target in days, run the full confirmatory read when the deal is real, then keep the whole fund scored over the hold. Every tier includes named SME hours — the engine does the sweep, people do the judgment.
No finding reaches your investment committee without a human SME standing behind it. The hours below are included in the fee — not billed on top.
Start with a scoping call. We'll score the AI behind your target — how it operates, what it sells, what it's worth — and hand your deal team the price, the terms, or the walk.
Book a scoping call →The target has an AI story. Your model has a number. Nobody in the process has checked whether they match. We score the AI a target actually runs — across eight weighted domains — and convert every finding into price, structure, or a walk, inside your confirmatory window.
Commercial, financial, operational, technology, legal — all scoped, all owned. The AI a target runs is scoped by no one. We read it across three surfaces, all resting on a financial foundation that turns every finding into a number.
Automation with no fallback is opex you can't hold and a 100-day fire you inherit on day one.
AI revenue the usage data won't support is a multiple you're paying on air.
The moat has to survive your hold period — not just their pitch.
This is the surface that converts a finding into price, structure, or a walk.
We quantify each finding against the six drivers of enterprise value, then walk your entry price down to one you can defend in the investment committee. An impairment finding is the overpayment you don't make.
A go/no-go read before you commit the diligence budget.
The IC-defensible read, sized to your window.
Bring us the target. We'll tell you what the AI is really worth — and hand your deal team the price, the terms, or the walk.
Book a scoping call →Every serious buyer is about to diligence your AI — and an unanswered AI question becomes a price chip in their hands. We run the buy-side read on you, before they do, so the answer is already in the data room when they ask.
A buy-side AI read has three surfaces and a financial foundation. Every one of them is a place your price can move — so every one of them is a place to be ready.
They will ask for the fallback. Have it built, not promised — promises get escrowed.
Your AI revenue claim will meet the usage data. Make sure it survives the meeting.
If you don't quantify the moat, the buyer will quantify its absence.
Every surface you leave unanswered becomes a price chip in their hand.
The buyer's diligence and your preparation are the same exercise, run by different people, at different costs to you. One ends in a discount. The other ends in a defended price.
The same finding costs you very differently depending on who surfaces it. Found by them, it's leverage. Found by you, it's a footnote — or a fix you've already made.
Run the buyer's screen on yourself, while there's still time to fix what it finds.
The full, evidence-linked read — in the data room on day one.
Run the buy-side read on yourself first. We'll tell you what a buyer will find, what it would cost you, and what to fix while you still can.
Get ahead of the buyer →Increasingly, the answer is a model — one the borrower may not own, may not control, and may not be able to replace. We score the AI dependency behind the cash flows, at underwriting and every quarter of the loan.
By the time a covenant breaches, the cause is three quarters upstream and nobody underwrote it. The dependency is scoreable at origination; the breach is only observable at the end.
The same weighted read the equity gets — read for durability of cash flow rather than upside. Three surfaces, resting on the financial foundation that connects them to your covenants.
A model in the critical path with no fallback is an opex spike and a missed payment.
The cash flows you're lending against may be attributed to AI that doesn't do what it claims.
The moat must outlast the tenor of your loan, not the borrower's growth story.
This is the surface that connects an AI failure to DSCR and covenant headroom.
The AI exposure behind the borrower's cash flows, before the paper is signed.
Re-scored every quarter — so the exposure shows up before the covenant does.
Bring us the credit. We'll score the AI dependency behind the cash flows — at origination, and every quarter after.
Book a scoping call →Diligence ends at close. The AI risk doesn't. We score the AI across every holding, re-score it every quarter, and tell your operating partners where the hidden risks sit and where the synergies actually are — before the exit process finds out for you.
A deal team sees one company. A fund-level read sees the same risk appearing in five of them, the value clustering in three, and the fix for one company already built and running inside another. That is the whole argument for scoring the portfolio rather than the deal.
Every holding on one composite, ranked worst first, against the sector baseline — with the movement since entry.
The same exposure, appearing in holding after holding. Correlated risk is invisible from inside any single company — and it is the only kind that can hit the whole fund at once.
If your primary model vendor reprices inference by 30%, or deprecates the model your holdings are built on:
No deal team can answer this question. A fund-level read answers it every quarter.
The upside is not spread evenly. Knowing where it clusters is the difference between a value-creation plan and a to-do list.
The fix a portfolio company needs is often already built, tested, and running in a company you also own. We map the assets against the needs.
Score against the exit calendar. A holding below the baseline with a process starting in eighteen months is the most actionable fact in the fund.
The bus factor: how many people could leave before the AI stops working. In most portfolio companies the honest answer is one — and nobody has written it down.
The offensive play. Two holdings whose combined data creates an asset that neither one could build alone — and that no competitor can replicate, because no competitor owns both companies.
What makes the synergy map compound: the same fix gets dramatically cheaper each time it moves to the next holding. Build once, deploy four times.
The models change under them. The vendor reprices. A competitor rebuilds the category. A company that scored 3.0 at close can be a 2.1 two years later without a single board paper mentioning it — and the first person to notice will be the buyer in your exit process.
The AI position erodes while the P&L still looks fine. Re-scoring surfaces it while there's still hold left to fix it.
Four holdings on the same model vendor is a fund-level exposure that no single deal team could ever see.
The fix that created 2.4 margin points in one company is a playbook for the next three.
Each AI fix does two jobs: it remediates the value at risk, and it creates new value on top — margin, revenue, operating leverage. Priced per holding, it becomes a portfolio-wide value-creation plan rather than a list of IT projects.
Every holding scored and re-scored, with the value-creation plan tracked against the original thesis.
The full eight-domain read — for a new acquisition, or a holding you've never scored.
We'll read every holding in the fund, tell you which one is going backwards, and price what fixing it is worth — while you still have hold period left to act.
Score the portfolio →We read every target twice. Once against the failure modes that actually turn up in its industry — payer rules, inference margins, model governance, plant fallbacks. And once against the business-model patterns that cut straight across all seven sectors — because AI risk follows business models, not industry lines. The specialist knows the symptom. We know the symptom, the pattern, and where the fix has already been built.
Each read starts from the failure modes that actually turn up in that industry — not a generic AI checklist with your logo on it.
PHI in a training set is an indemnity, not a footnote
Top failure modeCoding automation drifting against payer rules
Open the sector read → SectorInference cost per seat turns an 80% margin into 60%
Top failure mode“AI-native” that is a wrapper on someone else's model
Open the sector read → SectorAn ungoverned pricing model is an exam finding
Top failure modeNo adverse-action explanation path
Open the sector read → SectorA model in the critical path with no fallback is downtime
Top failure modeForecasting AI setting inventory with no override
Open the sector read → SectorAI breaks the link between headcount and revenue
Top failure modeFees priced on hours AI completes in minutes
Open the sector read → SectorHow much of this multiple is AI narrative?
Top failure modeRevenue concentrated in two hyperscalers
Open the sector read → SectorA fraud model drifting 40bps is an EBITDA event
Top failure modeReserves modeled on pre-AI loss patterns
Open the sector read →Seven business-model patterns account for most of the AI value destruction we find. Almost none of them belong to a single industry — which is precisely why a single-industry adviser keeps meeting them for the first time.
Take the most common pattern we find — a model sitting in the critical path with no fallback. It appears in all seven sectors, and in every one of them the operators call it something else. That's why it goes unscoped.
Because the patterns repeat, the fixes travel. The failover harness we built for a distribution business is the one a payments platform needs — and the second build is always a fraction of the first. That is the compounding advantage of a partner who works across industries, and it is the one thing a sector specialist structurally cannot offer.
The first time a pattern appears in your sector, it's a crisis. It may be the fortieth time we've seen it — just wearing a different uniform.
Remediation playbooks move between industries. You pay for the second deployment, not the first.
For funds holding assets across sectors, the eval harness in one company is the missing control in another. We map it.
Bring us the target. We'll read it with the failure modes of its own industry — and with the patterns we've already found in the other six.
Book a scoping call →Ambient scribes, coding automation, prior-auth bots, triage tools. In healthcare services the AI sits inside the revenue cycle and the clinical record — which means an AI finding is never just an IT finding. It's revenue integrity, regulatory tail, and PHI exposure.
The method doesn't change. What changes is where the risk concentrates — and in healthcare services, it concentrates hard.
Coding and prior-auth automation sits directly in the revenue cycle. A drifting model is a denial rate, not a bug report.
Clinical-facing AI can be a regulated device. If it's making a claim about care, someone has to own that claim.
As AI cuts cost-to-serve, payers reprice. The margin you're underwriting may be temporary by design.
In healthcare the tail is the deal: PHI in a training set is an indemnity, not a footnote.
These are the findings we look for first — ranked by how often they turn up, and what they hit when they do.
A go/no-go read before the diligence budget commits.
The IC-defensible read, sized to your window.
Bring us the target. We'll score the AI behind it — with the healthcare services failure modes checked first — and hand your deal team the price, the terms, or the walk.
Book a scoping call →In software the AI question is a margin question and a moat question. An “AI-native” product built on someone else's model has someone else's cost curve and someone else's roadmap — and the multiple you're paying assumes neither.
The method doesn't change. What changes is where the risk concentrates — and in software / saas, it concentrates hard.
Internal AI is rarely the risk here — but the inference bill is, and it lands in COGS.
AI-attributed ARR is the whole thesis. Does the usage data support the revenue claim, cohort by cohort?
The most dangerous sentence in software: “the model provider could ship this next quarter.”
A SaaS multiple assumes SaaS margins. Inference cost per seat can quietly turn 80% into 60%.
These are the findings we look for first — ranked by how often they turn up, and what they hit when they do.
A go/no-go read before the diligence budget commits.
The IC-defensible read, sized to your window.
Bring us the target. We'll score the AI behind it — with the software / saas failure modes checked first — and hand your deal team the price, the terms, or the walk.
Book a scoping call →Underwriting, pricing, claims, fraud, credit decisioning — in financial services the AI is the operating model, and it runs inside a supervisory regime. An ungoverned model isn't a technical debt item. It's an exam finding, a remediation order, and a fine.
The method doesn't change. What changes is where the risk concentrates — and in insurance & financial services, it concentrates hard.
Claims and servicing automation touches the customer outcome the regulator examines.
A pricing model is a regulated artifact. If it can't be explained, it can't be defended — in an exam or in court.
The supervisory perimeter is moving faster than the models. Today's compliant model is tomorrow's finding.
Here the AI finding becomes a number the hard way: remediation cost, capital, and penalty.
These are the findings we look for first — ranked by how often they turn up, and what they hit when they do.
A go/no-go read before the diligence budget commits.
The IC-defensible read, sized to your window.
Bring us the target. We'll score the AI behind it — with the insurance & financial services failure modes checked first — and hand your deal team the price, the terms, or the walk.
Book a scoping call →Forecasting, pricing, routing, predictive maintenance, vision inspection. In industrials the AI is wired into physical operations — so the failure mode isn't a bad dashboard, it's downtime, scrap, stranded inventory, and a safety file.
The method doesn't change. What changes is where the risk concentrates — and in industrials & distribution, it concentrates hard.
This is the surface that matters here. A model in the critical path with no fallback is downtime, and downtime is cash.
“Smart” products carry a promise. If the model underperforms, the warranty is real and so is the recall.
If a competitor's AI takes 3 points out of their cost curve, your pricing power is gone before the hold ends.
Forecasting AI sets inventory. A bad model is working capital — the most expensive kind of quiet.
These are the findings we look for first — ranked by how often they turn up, and what they hit when they do.
A go/no-go read before the diligence budget commits.
The IC-defensible read, sized to your window.
Bring us the target. We'll score the AI behind it — with the industrials & distribution failure modes checked first — and hand your deal team the price, the terms, or the walk.
Book a scoping call →Staffing, accounting, agencies, engineering, BPO. Here AI doesn't threaten a system — it threatens the delivery model itself. The question isn't whether they use AI. It's whether the work they sell survives it, and whether the headcount leverage your model depends on still holds.
The method doesn't change. What changes is where the risk concentrates — and in business & professional services, it concentrates hard.
Delivery is people. If AI does a third of the work, the leverage model changes and the margin moves — up if the firm owns it, down if a competitor does.
The deliverable is the product. If the client can now generate it themselves, you're buying a commodity at professional-services prices.
Price compression arrives from the client side: procurement is already benchmarking the fee against what AI costs them to do it in-house.
This whole asset class runs on revenue per head. AI breaks the link between headcount and revenue — in both directions, and only one of them is in your model.
These are the findings we look for first — ranked by how often they turn up, and what they hit when they do.
A go/no-go read before the diligence budget commits.
The IC-defensible read, sized to your window.
Bring us the target. We'll score the AI behind it — with the business & professional services failure modes checked first — and hand your deal team the price, the terms, or the walk.
Book a scoping call →Cooling, power, thermal management, electrical, data-center fit-out. These businesses now command the richest valuations in the market — on the assumption the AI capex keeps coming. So the diligence question inverts: not is their AI real? but is the demand you're underwriting durable — and what is this worth if the capex pauses?
The method doesn't change. What changes is where the risk concentrates — and in ai infrastructure & specialty industrials, it concentrates hard.
The plant still has to run — but in this sector operational AI is the smallest risk on the page. The risk is the demand curve.
The question that decides the multiple: is this genuinely specified into AI builds, or a commodity repriced for sitting next to one?
Customer concentration is the whole game. Three procurement teams deciding to pause isn't a market correction — it's your revenue.
Ask the uncomfortable question out loud: how much of this enterprise value is AI narrative, and what is the multiple without it?
These are the findings we look for first — ranked by how often they turn up, and what they hit when they do.
A go/no-go read before the diligence budget commits.
The IC-defensible read, sized to your window.
Bring us the target. We'll score the AI behind it — with the ai infrastructure & specialty industrials failure modes checked first — and hand your deal team the price, the terms, or the walk.
Book a scoping call →Fraud scoring, credit decisioning, KYC and AML, dispute automation. In payments the AI sits directly on the money — every model decision is an authorization, a loss, or a regulatory finding. And the take rate you're underwriting rests on models no one in the deal team has ever opened.
The method doesn't change. What changes is where the risk concentrates — and in payments & fintech, it concentrates hard.
Fraud and AML models are the line between a loss provision and a consent order. Both have a number, and both land on you.
If risk decisioning is what they sell, the model is the IP — and it may turn out to be a vendor's model with a wrapper on it.
Agentic commerce is changing who initiates a payment. If an agent chooses the rails, checkout placement stops being an asset.
Loss rate, take rate, reserves. A fraud model drifting 40bps is an EBITDA event, not an engineering ticket — and the reserve was set before it drifted.
These are the findings we look for first — ranked by how often they turn up, and what they hit when they do.
A go/no-go read before the diligence budget commits.
The IC-defensible read, sized to your window.
Bring us the target. We'll score the AI behind it — with the payments & fintech failure modes checked first — and hand your deal team the price, the terms, or the walk.
Book a scoping call →