AI in the value creation plan: how to build AI into the first 100 days and carry it to exit

What a value creation plan is, and where AI fits in

The pressure on this plan has grown. In Bain’s illustrative deal model, a typical buyout today needs 10–12% EBITDA growth a year to return 2.5 times the equity over five years, against about 5% in 2015 (Bain, Global Private Equity Report 2026). With flat multiples, more of that growth has to come from operations.

AI does not need its own chapter in the plan. It is a way of delivering the levers the plan already has, often faster and with less hiring. A mistake we often see is the opposite: an “AI initiative” next to the plan, with its own budget, its own dashboard and no line in the EBITDA bridge.

The first 100 days: a sequence that works

The first 100 days set the rhythm of a holding period. AI fits into them if it starts small and concrete. This is the sequence we use; it is a guide, not a fixed timetable.

Days 1–30: scan and value map

Go through the company’s processes with one set of questions: where is work manual, slow or knowledge-heavy, where is margin lost, which data and systems exist, who would own a change. The result is a value map: ten to twenty candidate cases, each with a first estimate and a feasibility rating. If the fund holds several companies, use the same lens everywhere, so that cases can be compared and reused.

Days 31–60: calculate, decide, govern

Calculate the two or three best cases properly: baseline, effect range, realisation share, running costs, owner. Management decides which go first. In parallel, settle the rules: which tools are permitted, which data must not go into which system, who approves, and, where there is one, the works council. Start the first build as soon as the first case is decided.

Days 61–100: first tool live, baseline, rollout plan

A controlled MVP typically takes two to four weeks of build time: the AI does the groundwork, a person approves, the ERP or CRM stays the system of record. By day 100 the first tool runs on real cases, its KPI baseline is measured, and the plan names the next cases with dates. For comparison, Partners Group reports 14 weeks from concept to production for its first AI use case at Foundation Risk Partners (Partners Group 2026).

How to count AI in the EBITDA bridge

AI cases are easy to overstate. Time saved rarely leaves the cost base in full: Uber cut invoice handling time by 70% but process cost by 25–30% (Uber 2025); Maximus cut processing time by 85% and its AP team by 20% (Rossum). Four rules keep the plan honest:

• Cost cases count time saved times a realisation share, the part that actually leaves the cost base.
• Volume cases count extra revenue at contribution margin, never at revenue, and must not overlap with price cases.
• Running costs of the tools (licences, cloud, upkeep) are subtracted; one-off build costs are shown separately.
• Baseline before go-live: measure the KPI before the tool runs, or the effect cannot be shown later.

We put every case into a calculation model with a cockpit: every number derived and backed up with its source and quality grade, every assumption visible, every scenario at the push of a button. Management decides on facts, and the model stays with the company after the project, so the plan can be updated without us.

An illustrative example, not a forecast: a company with €100m revenue and a 12% EBITDA margin (€12m). A price case of one point on all revenue adds €1.0m if volume holds. An order intake case saves 40% of handling time in a process that costs 1.5% of revenue; at a 40% realisation share that is 16% of €1.5m, or €240k. After 10% running costs, the two cases add about €1.1m, or 9% of EBITDA. At an exit multiple of ten, that is about €11m of enterprise value from two calculated cases. Whether the inputs hold is exactly what the calculation in days 31–60 has to establish.

Governance: what belongs in the plan from day one

Rules decided late slow everything down. Three topics belong in the plan from the start:

• EU AI Act. A company that uses AI under its own authority is a “deployer”. Deployers take measures to support their staff’s AI literacy (Art. 4, reworded in July 2026) and observe the prohibited practices (Art. 5). Using high-risk systems, for example AI for selecting job applicants, brings deployer obligations under Art. 26 from 2 December 2027 (AI Act; Digital Omnibus).
• Data protection. Where personal data goes to an AI provider, a processing contract is needed (GDPR Art. 28(3)), and the plan should say which data may go into which tool.
• Works council (Germany). The works council must be informed about planned use of AI (§ 90(1) no. 3 BetrVG) and has co-determination rights over technical systems that can monitor conduct or performance (§ 87(1) no. 6 BetrVG) (BetrVG). Involving it early saves months.

This is not legal advice; it lists the questions the plan should answer, not the answers.

Choosing cases that fit AI

Not every process gains from AI. In a field experiment with 758 consultants, AI users did markedly better on tasks inside the AI’s capability frontier and were 19 percentage points less often correct on a task outside it (Dell’Acqua et al. 2026). For the plan this means: prefer frequent, bounded tasks whose result a person can check before it takes effect, and keep judgement calls with people. AI use cases in portfolio companies lists the cases with the best documented results.

The AI part of the equity story at exit

A buyer will ask what AI has done for the company and whether it will keep working without the seller. From our point of view, three things make the answer credible:

1. Live tools with a KPI history. Twelve months of reported effect beat any roadmap.
2. Ownership. Code, data and know-how sit with the company, not with a consultant or a platform it cannot leave. The core systems stay in the lead.
3. People who run it. Managers who know how the tools work, where they fail and how to steer them, so the effect does not depend on one person.

We work with private equity funds and their management teams on exactly this: value map, calculated cases, tools in production within weeks, leaders who carry them, impact tracked against the value creation plan. The two founders, former partners at McKinsey & Company and Roland Berger, lead every mandate personally. More on pe.leadbeyond.de; the wider context is in AI in private equity.