Assessing AI Disruption Risk in a Target: AI-Native vs AI-Vulnerable

1. The Question Every 2026 Deal Now Has to Answer

Before you underwrite a target's growth, answer a second question: will AI make this business stronger over the hold, or quietly break it? Score the target on a line that runs from AI-native at one end to AI-vulnerable at the other. Where it lands changes the thesis, the model, and the price you should pay.

This is a different job from using AI to run diligence. Extraction, document review, and first-pass analysis speed up the diligence process itself, and our complete guide to AI due diligence covers that work. This page is about AI as the subject of diligence: judging whether the target's own moat gets deeper or shallower as AI spreads through its market.

Timing is why it matters now. A target can post clean numbers today and still sit on a moat that AI dissolves over a five-year hold. The buyer who spots that early reprices or walks. The buyer who misses it underwrites growth that will not arrive and pays a multiple the business cannot defend. The scorecard below exists to make that call early, on evidence, and in a form the investment committee can act on. The asymmetry is what makes it worth the effort: getting the call right on one deal saves a markdown, and getting it wrong compounds across the whole hold, because you built the plan, the debt, and the exit on a base that was quietly shrinking.

2. AI-Native, AI-Adopting, AI-Vulnerable

Three positions, with a concrete example of each so the labels stay honest.

AI-native. AI makes the business stronger as it spreads. Think of a company whose proprietary data compounds and gets more useful as models improve, or one whose workflow position lets it ship AI features customers cannot get anywhere else. For these targets, wider AI adoption is a tailwind that deepens the moat.

AI-adopting. A sound business that has to add AI to defend its position. Picture a mid-market software vendor with sticky customers that now has to build AI into the product before a competitor does. The business is fine today, and the thesis holds only if the AI plan actually lands.

AI-vulnerable. AI erodes the moat. A services business that bills for repeatable human work a general model now does for a fraction of the cost, or a product whose main value was doing something AI has made cheap. These targets can look healthy in the trailing numbers while the ground shifts underneath them.

Most targets sit in the AI-adopting middle, which is the interesting one: the deal works if the plan is real and fails if it is a slide. The scorecard exists to tell the three positions apart with evidence, because a founder will describe almost anything as AI-native. The three positions are points on a spectrum rather than boxes, and a target can be AI-native in one product line and AI-vulnerable in another, so score the parts that carry the thesis instead of the company as a slogan.

3. The AI Disruption Scorecard

Score the target across five signals. For each one, the AI-native end is protective and the AI-vulnerable end is the risk. Use a simple score of one to five, or just high, medium, and low, and answer the diligence question in the last column with evidence rather than management's framing.

Add the signals up. Mostly protective, and the target leans AI-native, so the job is to verify the moat is as real as management says. Mixed, and it is AI-adopting, so the AI plan has to be written into the model rather than assumed. Mostly exposed, and it is AI-vulnerable, so the thesis needs a lower price, a different structure, or a pass. Do not average the five blindly, though. One signal can dominate: a business whose entire value is a task AI now does for free reads as AI-vulnerable even with four comfortable scores, so treat any single low mark on pricing power or the labor-cost moat as a flag that overrides a reassuring total.

4. Reading the Five Signals

Pricing power. The first test is whether customers keep paying once a good AI tool is a click away. If the product's value is an outcome AI cannot easily copy, judgment, trust, a regulated sign-off, a network effect, pricing holds. If the product mainly does work a general model now does for pennies, the next renewal is a repricing conversation whether or not management admits it yet. A useful tell: watch what buyers do the moment a capable AI tool ships in the category. If usage and willingness to pay hold, the value was never the task AI automated. If both wobble, the repricing has already started and the trailing numbers just have not caught up.

The labor-cost moat. Many services and outsourcing businesses earn margin by doing repeatable human work cheaply. That margin is exposed to exactly the degree the work is repeatable. Ask what share of the cost of delivery is people doing tasks a model could absorb. A high share is a threat to margins and, for a buyer with a real plan, an opportunity, and which one it becomes depends on who moves first.

Data and workflow advantage. Proprietary data that compounds and gets more useful as models improve is one of the few moats AI strengthens rather than erodes. So ask whether the target owns data or holds a workflow position a new AI entrant could not quickly reach. Commodity data a model already contains protects nothing, no matter how much of it the target has.

Adoption speed. Two companies facing the same AI threat can end up in opposite places based on whether they can ship AI faster than their competitors. Look for the unglamorous evidence: clean data, engineering talent, a culture that runs real experiments. A company that cannot deploy AI internally will not out-run a rival that can, whatever the founder says on the management call.

Entrant threat and switching costs. Last, ask how cheaply a funded AI-native startup could rebuild the core offering, and what would stop customers from leaving if one did. High switching costs, regulation, and owned distribution protect a base even while the core commoditizes. Low switching costs plus a rebuildable product mean the moat is thinner than the trailing financials make it look. Read the five signals together, because they interact: a target strong on data advantage and adoption speed can turn an AI threat into a deeper moat, while one weak on both compounds its exposure. Score the pattern, not the sum, and weight the signals that actually drive this particular business.

5. What It Does to the Model

The score is only useful once it moves numbers. Each position points at a different set of assumptions in the model.

For an AI-vulnerable target, the growth and margin assumptions need a haircut, and the exit multiple should assume the next buyer runs this same test and pays accordingly. Underwriting yesterday's margins on a business whose labor moat is melting is how a good entry price becomes a bad one three years in. Pay special attention to the exit assumption: a five-year hold means the buyer who takes the asset from you runs this same scorecard with five more years of AI progress behind them, so an entry multiple that ignores that is a bet the next buyer will be less careful than you were.

For an AI-adopting target, the model has to carry the cost and the payoff of the AI plan explicitly: the investment to build it, the timeline, and the margin it defends or wins. If that plan is load-bearing for the return, it belongs in the base case, not the upside column where optimistic assumptions go to hide.

For an AI-native target, a premium can be defensible on one condition: the moat is proven with evidence rather than asserted by the founder. The upside case where AI accelerates the business is genuinely real for these targets, and it still has to survive the same five signals as everything else.

6. What It Does to the Price and Structure

Placement should show up in the offer, not just the memo. The scorecard earns its keep when it changes a number the seller sees.

Three moves follow from an AI-vulnerable read. Reprice to the defensible earnings rather than the reported ones. Structure around the uncertainty with an earnout or a larger equity cushion. Or walk, which is a perfectly good outcome that a documented scorecard makes easy to defend to the IC. A pass you can explain beats a loss you cannot.

For an AI-adopting target, the structure question is time. The AI plan takes quarters to land, so the hold-period plan and the operating budget have to fund it from day one, and the value-creation plan should name who owns it. A plan that lives only in the deal memo defends zero points of margin once the deal closes. If you structure around the risk, aim the earnout at the specific thing you are unsure of, the retained pricing power or the delivered AI plan, rather than at headline revenue. A well-aimed earnout prices the exact uncertainty the scorecard surfaced; a lazy one just delays the argument to a worse moment.

7. Running the Assessment in Diligence

Make it a standard workstream, owned by a named person, run on every deal. The commercial diligence lead should own the scorecard, with a technical adviser weighing in on adoption speed and the entrant threat, and it should run early enough in the process to still change the price.

Ask customers the question that matters most: if a credible AI alternative appeared next year, would you switch, and what would stop you? Customer answers beat management answers on disruption risk every time, because the customer has no thesis to protect. Pair those calls with a hard look at the target's own AI roadmap: what has actually shipped, what is a slide, and whether the engineering team can tell the difference. Bring that technical adviser in early rather than as a box-check at the end, because whether a target can really ship AI, and whether its data is as proprietary as claimed, is an engineering read that a commercial team gets wrong more often than it expects. When management shows a roadmap, ask to see the commits, the hires, and the customers already using it, because a roadmap with none of those behind it is a wish.

A large and growing share of PE firms now build an explicit AI assessment into commercial diligence. Industry surveys through 2025 and 2026 point the same way, though the exact share depends on who is asked and how the question is framed, so treat it as a clear direction rather than a precise figure. BCG's research on the widening AI value gap, quoted below, is why that direction is rational: the same divergence that separates AI winners from losers among operating companies is the thing you are trying to price in a target.

8. Where to Start

Start by running the scorecard on your last two deals, blind to how they turned out. If the score would have flagged a problem you later hit, or missed one, you learn how to weight the five signals for the sectors you actually buy. Then build the scorecard into the standard diligence template so no target skips the second question.

If you want the assessment run on a live target by a team that does it every week, our AI Diligence engagement ($15,000 for a screen, $25,000 for a comprehensive review, three-deal pack at $40,000) delivers the AI-native-versus-AI-vulnerable read with people owning every judgment.

And if you want the scorecard baked into how your firm underwrites, an AI Readiness Sprint ($12,500 flat for firms up to 20 people; the $30,000 Comprehensive version for firms of 20 or more) builds it into your diligence process so every deal team applies the same test the same way. The firms that make this a habit do more than dodge the weak targets: they price the strong AI-native ones with more conviction than the buyers still treating AI as someone else's problem, and in a competitive process, conviction is what wins the asset at a price that still works.