AI readiness of a portfolio company: what to check before and after the deal

What AI readiness means for a portfolio company

Generic AI maturity models score a company from one to five and say little about value. An investment team needs something else: where AI could move EBITDA in this particular company, what it would take, and what could stop it. Most companies are early on that path: in a BCG survey of 1,250 senior executives, 60% report minimal gains from AI so far and only 5% achieve value at scale (BCG 2025).

Six dimensions of an AI readiness assessment

Value: start with the money, not the technology

Map the processes where work is manual, slow or knowledge-heavy, and put a rough cost on each: order handling, customer service, finance, proposals. Then ask where margin leaks: discounts outside policy, slow quotes, errors that lead to credit notes. These are the candidates. A good candidate occurs often, is clearly bounded and can be checked by a person before its result takes effect.

Data and systems: what is there, and what is allowed

Most useful AI cases in mid-sized companies run on data the company already has: past offers, orders, invoices, tickets, policies. The questions are whether these exist in a form a tool can read, whether the company has the rights to use them, and whether the core systems can take results back. In our work the ERP or CRM stays the system of record; the AI does the groundwork around it.

People and leadership: who will own it

Every case needs a manager who owns the result. If no one in the management team wants that role, the case will stall regardless of the technology. Also check where know-how sits: a process that lives in the head of one person is both a risk and an opportunity for a knowledge tool.

Governance and legal: the questions that slow everything later

• AI Act role: a company that uses AI under its own authority is a deployer; it must take measures to support its staff’s AI literacy (Art. 4) and observe the prohibited practices (Art. 5). High-risk uses, such as AI in applicant selection, bring deployer obligations under Art. 26 from 2 December 2027 (AI Act; Digital Omnibus).
• Data protection: contracts with processors for every AI tool that receives personal data (GDPR Art. 28(3)).
• Works council (Germany): information about planned AI use (§ 90(1) no. 3 BetrVG) and co-determination where a system can monitor conduct or performance (§ 87(1) no. 6 BetrVG) (BetrVG).
• Vendor contracts: who owns what is built, what happens to data, and whether the company can leave.

For a structure, the NIST AI Risk Management Framework offers four functions, Govern, Map, Measure and Manage. It is voluntary and not a checklist, but it helps to make sure nothing is forgotten. None of this is legal advice.

Exposure: does AI strengthen or threaten the business model?

This question can change the investment case itself. Does the company sell something AI can already do cheaply? Do its customers gain new options? Or does AI make its own offer better, faster or cheaper to deliver? A company with high exposure needs a strategic answer, not only efficiency cases. How to assess market options like an investment decision is described in Go-to-market strategy.

Before the deal: questions for due diligence

Before signing, there is little time and limited access. Focus on the questions that change the price or the plan:

1. Exposure: Which parts of revenue depend on tasks AI can already perform? Which customers could do it themselves?
2. What already runs: Which AI tools are in use, by whom, with which data? Is there an AI policy?
3. Rights and contracts: Does the company have the right to use its customer and operational data for AI? Which AI vendors hold its data, and on what terms?
4. Key people and know-how: Is critical knowledge documented, or does it sit with a few people?
5. Upside: Where are the two or three largest value pools for AI, roughly sized, to inform the value creation plan?

Answers to the last question should be ranges, not promises. Published results are lever by lever and mostly self-reported; AI use cases in portfolio companies shows what has been documented and with what caveats.

After closing: from assessment to value map

After closing, the same six dimensions become the basis of the plan. Every candidate case is rated on two axes: value (euros of EBITDA or cash, as a range) and feasibility (data, systems, owner, rules). The first case is where both are high. That is the logic of a portfolio AI scan: one lens for every company, a value map per management team and a heatmap for the investment team, so that cases can be compared and a tool that works in one company can be reused in the next.

Two findings from research shape the ratings. RAND’s interviews with 65 data scientists and engineers name five root causes of AI project failure: a misunderstood problem, missing data, technology before purpose, missing infrastructure and tasks too hard for AI (RAND 2024). And AI helps markedly on tasks inside its capability frontier while making people 19 percentage points less often correct on a task outside it (Dell’Acqua et al. 2026). A feasibility rating that checks the problem, the data and the fit of the task catches most of this before any money is spent.

We run portfolio AI scans for private equity funds as an impact partner, led by two former partners at McKinsey & Company and Roland Berger, and then build the first tools on the company’s own systems. More on pe.leadbeyond.de.