Mind the gap

Chapitre

Value shift

AI in the value chain: the EBITDA upside to negotiate before you sell your business

Écart

Value shift

Lecture

11 min

Août 2026

A company that has embedded little AI in its value chain comes to a business sale with something its balance sheet does not show: untapped margin potential. That potential exists, it can be calculated, and it will be calculated — the only question is by whom. If the buyer is the only one to quantify it, the buyer acquires it without paying for it. If the seller has quantified it, evidenced it and put it on the table, it becomes part of the negotiation. That is where the whole gap lies.

● Station 1 — Visible · today

The signal: the AI upside is still untapped, and buyers are already equipped

Fig. 1 — Actual adoption in SMEs: one link switched on, five addressable cost links left untapped.

First fact: adoption is real, but integration is shallow. According to Bpifrance Le Lab, around one third of French SMEs and mid-sized companies now use AI, and use of generative AI among micro, small and medium-sized businesses jumped from 31% to 55% in a single year. Yet half of adopters stick to free or off-the-shelf tools: drafting a letter, roughing out a memo, looking something up. Occasional use dominates; deep integration into processes — the kind that shifts costs and shows up in a profit and loss account — remains the exception. The consequence for the business transfer market: the vast majority of companies sold in the next five years will come to the table with their AI potential untouched.

Second fact: buyers, for their part, have equipped themselves. FTI Consulting's 2026 Value Creation Index, based on more than 550 private equity decision-makers, finds that AI speeds up the delivery of value creation plans, especially when built into operational initiatives. And the return hurdle has risen: Bain's 2026 global report estimates that around 12% annual EBITDA growth is now needed — against 5% in the previous cycle — to deliver benchmark returns. Operational alpha has become the remaining lever, and AI is its first identified source. A new discipline is appearing in due diligence: the AI impact assessment, which no longer judges only the target's current performance but its exposure to transformation — as both an opportunity and a threat.

Third fact: the language of the negotiation has already changed. Buyers no longer reason only on reported EBITDA and its usual adjustments. They recalculate a target margin under AI assumptions — what this series calls AI-normalised EBITDA (Gap 03). A seller who has not done this calculation will discover it in the other side's letter of intent (LOI), in the form of assumptions they did not choose.

● Station 2 — Invisible · today

The mechanics: the AI-adjusted EBITDA calculation the buyer will make anyway

First mechanism: the value chain and the value proposition do not play the same tune. AI embedded in the chain — production, purchasing, logistics, sales administration, support functions, prospecting — acts on costs: a margin effect, measured in EBITDA points. AI embedded in the value proposition — the product or service itself — acts on mix, price and recurrence: a multiple effect, potentially more powerful but double-edged. Because while AI can enrich the proposition, it can also make it substitutable: a service whose core will tomorrow be produced by a generic model sees its value proposition dissolve, not improve. The buyer will analyse both levels separately — the seller must do the same.

Fig. 2 — The EBITDA bridge: whoever builds it chooses the assumptions. Whoever has not built it has to live with them.

Second mechanism: the calculation the buyer will make anyway. The method is simple and blunt. Take each link, isolate the addressable cost base — hours spent processing documents, data entry, reporting, sales qualification, first-line support — apply a prudent capture rate, and rebuild an EBITDA bridge: reported margin, plus addressable gains link by link, equals AI-normalised margin. The buyer builds this bridge for itself, during due diligence, with its own assumptions. If it is the only one to have built it, it buys the company on reported EBITDA, rolls out the plan after completion and captures the entire gap. The seller's untapped potential becomes the buyer's return — legitimately, since no one else had quantified it.

Third mechanism: the same file supports two opposite readings. A company with little AI is, depending on the analyst, either untapped margin potential (a premium reading) or a structural lag (an obsolescence-discount reading). What tips one reading into the other is not the level of adoption: it is the credibility of catching up. Data quality and ownership, process digitalisation, dependence on legacy systems, the skills pyramid — Bpifrance observes that a company already engaged in digital transformation is five times more likely to use AI than one with little digitalisation. Untapped potential backed by clean data and documented processes earns a premium; the same potential sitting on an ageing information system earns a discount, because the cost and risk of catching up eat into the expected gains.

Fourth mechanism: the J-curve, or why timing is treacherous. Erik Brynjolfsson described the productivity J-curve: general-purpose technologies first require intangible investment — reorganisation, data, training — that depresses measured performance before lifting it. Ajay Agrawal, Joshua Gans and Avi Goldfarb, in Power and Prediction, refine the mechanism: AI lowers the cost of prediction, but value does not come from point solutions — it comes from redesigning decision systems around that now-cheap prediction. Adoption figures confirm it: free, occasional use does not produce EBITDA. And Daron Acemoglu, in Power and Progress, adds the uncomfortable nuance: “so-so” automation, which replaces without rethinking, sometimes degrades service without creating any net gain. The consequence for a seller: launching a full transformation twelve months before the sale means risking presenting accounts at the very bottom of the J-curve — costs already incurred, gains not yet visible. The worst possible timing.

Raw potential is worth nothing in a price: only potential that is quantified, proven on a defined scope and capable of being written into a contract enters the negotiation.

● Station 3 — Announced · tomorrow

What the players are betting on: from paid-for potential to assumed potential

Fig. 3 — Two readings of the same file. The level of adoption does not decide; the credibility of catching up does.

What buyers are preparing: an industrialised playbook. Funds are building quantified value creation plans in which AI is now a standard chapter, with initiatives that are costed, dated and assigned, and tested from one deal to the next. As these playbooks become standardised, undocumented potential will be paid for less and less: why pay for upside the buyer knows it can capture alone, faster than the seller? The price of raw potential tends towards zero; only proven potential is negotiable.

What the market is signalling: a widening gap between deals. Conversely, targets able to demonstrate AI gains already reflected in their accounts — even on a limited scope — will earn a stronger quality-of-earnings reading: a demonstrably improving margin, a de-risked roll-out plan, data ready to use. The valuation gap between the “credible potential” company and the “costly laggard” will widen, for the same business and the same size.

What lies ahead over ten to twenty years: potential turning into a discount. Once AI gains have become the sector norm, margin benchmarks will build them in. A company below the benchmark will no longer be valued as potential but as catch-up to be financed: the benchmark, minus the cost and risk of upgrading. Untapped potential is a selling point only until the norm has shifted; after that, it becomes a liability. This is a fully fledged argument about the timing of a sale: the window in which potential can be monetised is the one in which it is still rare to have proven it.

● Station 4 — In your hands · now

Closing the gap before selling your business: five decisions, twenty-four months before the platform

Fig. 4 — The timetable that avoids the bottom of the J-curve: prove on one scope, extrapolate to the rest.

For an owner considering a sale within two to five years, five decisions belong to today.

1. Map the chain and score each link before the buyer does

Function by function: addressable cost base, technical feasibility, dependencies, time to capture. The value proposition is analysed separately, with the reverse question: what does generic AI make substitutable in what we sell? This mapping is the foundation for everything else.

2. Build the AI-normalised EBITDA bridge yourself

With prudent, sourced and defensible assumptions — conservative capture rates, implementation costs included, a realistic timetable. Arriving at the table with your own numbers means choosing the ground for the discussion; having them imposed means negotiating within the other side's assumptions.

3. Deliver two or three proof points, eighteen to twenty-four months ahead

Not a full transformation — two or three projects with a limited scope, chosen because they read clearly in the accounts: gains that are measured, documented and auditable in the accounts presented. These proof points carry a narrow scope through the J-curve and make the rest of the potential credible by demonstrated extrapolation, without exposing the whole company to the bottom of the curve during the observation period.

4. Clean up the foundations the AI impact due diligence will look at first

Data ownership and quality, contracts with AI providers (reversibility, confidentiality, data location), compliance with the GDPR and the EU AI Act, process documentation. Untapped potential is only credible if what lies beneath it is accessible: these foundations are what tip the reading from “lag discount” to “potential premium”.

5. Negotiate how the upside is shared rather than giving it away

For the part of the potential still to be demonstrated at the time of sale, there are instruments that turn a debate about assumptions into a contractual clause: an earn-out linked to the delivery of AI gains over the first financial years, vendor financing, or a minority rollover by the seller that ties them to the value they prepared. The upside then stops being an unverifiable argument and becomes conditional value, shared according to written rules.

The gap here does not separate AI-equipped companies from laggards. It separates the seller who sells a track record from the seller who sells a track record plus untapped AI potential — mapped, quantified, partly proven and capable of being written into the contract. The first leaves the value gap on the table; the second brings it into the price of the business sale.

Frequently asked questions

Do you need to roll out AI everywhere before selling your business?

No. A full transformation launched too close to the sale exposes the accounts to the bottom of the J-curve: costs incurred, gains not yet visible. The recommended approach is to prove the gains on two or three limited scopes and to quantify the rest of the potential with prudent assumptions.

What is AI-normalised EBITDA?

It is the margin the company would reach once AI gains have been captured on its addressable costs, rebuilt link by link from reported EBITDA. Buyers calculate it during due diligence; the seller has every interest in producing its own figures beforehand.

Is a company without AI valued at a discount when it is sold?

Not necessarily. The same file can be read as untapped margin potential or as a structural lag. What decides between a premium and a discount is the credibility of catching up: data quality, process digitalisation, team skills.

How do you value AI potential that has not yet been realised?

Through conditional instruments rather than in the fixed price: an earn-out linked to the delivery of the gains, vendor financing, or a minority rollover by the seller. These mechanisms turn a debate about assumptions into a contractual clause.

Sources Bpifrance Le Lab, “L'IA dans les PME et ETI françaises” (AI in French SMEs and mid-sized companies) and business climate survey (January 2026) · FTI Consulting, 2026 Private Equity Value Creation Index · Bain & Company, Global Private Equity Report 2026 · E. Brynjolfsson, work on the productivity J-curve · A. Agrawal, J. Gans, A. Goldfarb, Prediction Machines and Power and Prediction · D. Acemoglu and S. Johnson, Power and Progress.

Regulatory notice Collaboration Capital is an independent adviser on business acquisitions, sales and financing, registered with ORIAS (the French register of insurance and banking intermediaries) under no. 25000696 as an IOBSP (intermédiaire en opérations de banque et en services de paiement — a broker in banking transactions and payment services). This article is for general information only and constitutes neither investment advice nor an offer of financial services. The results of any transformation or sale process depend on the specific characteristics of each company and cannot be guaranteed.

AI in the value chain: the EBITDA upside to negotiate before you sell your business

Venture capital leaves the station: tokenised securities, retail savings and porous borders

The business plan will no longer be a document

Dealmakers: the end of M&A league tables, the rise of proof

E-invoicing and Open Finance: when cash flows become programmable