In short
AI use cases in a portfolio company succeed or fail with the people who are supposed to use them. Find the cases together with the teams that do the work, not in a management workshop, and test each one for value, feasibility and acceptance. Then take people along along the whole chain from what they believe about AI to how they act: leaders who use the tools first, training on their own cases, tools built with the users, clear rules and an honest answer to what happens with the time saved. BCG puts about 70% of the value of an AI transformation in workflows, culture and collaboration, and only about 10% in the AI application itself.
Key takeaways
- Most of the value of AI sits with people and processes: BCG attributes about 10% to the AI application, 20% to data and technology and 70% to workflows, culture, governance and human-AI collaboration.
- Adoption stalls below management level: in BCG’s 2025 survey, regular use among frontline employees is stuck at 51%.
- Leaders make the difference: with strong leadership support, the share of employees who feel positive about AI rises from 15% to 55%. Only about a quarter of frontline employees get that support.
- The best use cases come from the people who do the work. Test each candidate for value, feasibility and acceptance before anyone builds.
- We look at adoption along the chain we use in LEADBeyond 360°: Belief, Capability, Trigger, Behaviour, Outcome. A tool nobody believes in is not used under pressure.
- Say honestly what happens with the time saved. Without a plan for it, saved hours neither reduce cost nor create growth.
Why AI value is decided by people
Investment teams usually ask which AI use cases a portfolio company should pursue. In our experience, the more important question is who will use them, and whether they want to. BCG puts it in a rule of thumb: about 10% of the value of an AI transformation comes from the AI application itself, 20% from data and technology and 70% from workflow redesign, culture, governance and human-AI collaboration (BCG).
Survey data shows where it breaks. In BCG’s 2025 survey of more than 10,600 leaders, managers and employees in 11 countries, regular use among frontline employees has stalled at 51%. Only one third say they were properly trained (BCG, AI at Work 2025). A use case that the order desk, the service team or accounts payable does not use will not show up in EBITDA, however good the model.
Definition
AI change management means preparing the people in a company to work with AI so that use cases are actually used and their effect reaches the numbers: from choosing the cases with the teams, through training and rules, to new routines and visible results.
Step 1: find use cases with the people who do the work
Use cases found in a management workshop tend to be the ones that sound good. Use cases found at the desk are the ones that save time. We walk through the processes with the teams that run them and ask four questions: Where do you type the same thing twice? Where do you search for information? Where do you wait for others? What would you never let a machine decide?
The last question matters most for acceptance. It shows where people want to keep judgement, and the design follows it: the AI does the groundwork, a person decides where there is no clear answer. Each candidate is then rated on three axes:
| Axis | Question | Evidence |
|---|---|---|
| Value | What is it worth in euros of EBITDA or cash, as a range? | Process cost, frequency, realisation share |
| Feasibility | Do data, systems and rules allow it? Can a standard feature do it already? | Sample documents, licence check, system access |
| Acceptance | Does the team that does the work want it, and who owns the result? | Named owner, users involved in the test cases |
A case that scores high on value and feasibility but low on acceptance is not a quick win. It is a leadership task first. How the three axes feed the plan is shown in AI readiness of a portfolio company and AI use cases in portfolio companies.
Step 2: take people along, from belief to behaviour
In LEADBeyond 360° we measure leadership along a chain: Belief → Capability → Trigger → Behavior → Outcome, from what a leader believes and feels able to do, through what throws them off under pressure, to what others observe and what results follow. The same chain explains why AI tools are adopted or quietly dropped.
| Link | Question for the team | What helps |
|---|---|---|
| Belief | Does this tool help me, or does it replace me? | An honest answer on what happens with the time saved; leaders who say it first |
| Capability | Can I work with it, and can I check what it produces? | Training on the team’s own cases; a tool that shows its sources |
| Trigger | What do I do under pressure, at month-end or with an angry customer? | A clear human route; the tool built for the busy day, not the demo |
| Behavior | Do I use it every day, or only when someone is watching? | Usage visible to the team; old workarounds closed |
| Outcome | Does anyone see the difference? | One KPI the team already knows, reported back to them |
The trigger link is easy to miss. Under pressure, people fall back on the routine they trust. A tool that works in a calm demo but slows down the month-end close will be switched off in the first busy week.
Five levers that make adoption stick
- Leaders go first. With strong leadership support, the share of employees who feel positive about generative AI rises from 15% to 55%; only about a quarter of frontline employees say they get that support (BCG 2025). In a portfolio company that means the managing director and the functional heads use the tools in their own work and talk about it.
- Train on real cases, not on features. Regular use is sharply higher among employees with at least five hours of training (BCG 2025). In a survey of 25,000 Danish workers, employer initiatives such as encouragement, in-house tools and training nearly doubled take-up for the typical worker, from 47% to 83% (Humlum & Vestergaard 2025). We train with the team’s own documents and processes.
- Build with the users, not for them. Test cases come from the team’s real documents with expected results, and the team signs off before go-live. People trust what they helped to test.
- Set clear rules early. More than half of employees (54%) say they would use AI tools even if not authorised (BCG press release 2025). Clear permitted tools are safer than bans. In the EU, companies using AI must take measures to support their staff’s AI literacy (Art. 4 AI Act, reworded by the Digital Omnibus); in Germany, the works council must be informed about planned AI use and has co-determination rights over systems that can monitor performance (§§ 90, 87 BetrVG). Not legal advice.
- Make the gain visible and fair. Report one KPI the team knows back to the team, and say what happens with freed time: more customers served, less overtime, no backfill for leavers. Silence on this question is heard as a threat.
Why this decides the EBITDA effect
Adoption is not a soft topic. Saved time only becomes value if someone decides what happens with it: 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% through natural attrition (Rossum). The gains are also not evenly spread: in a field study with 5,172 customer service agents, AI raised resolved issues per hour by 15% on average, with the largest gains for less experienced agents (Brynjolfsson et al. 2025). Who is trained first matters.
And quality has to be watched. Klarna began hiring human service agents again after its CEO said quality had suffered (Entrepreneur). Teams who use AI on tasks outside what it does well get worse results: consultants were 19 percentage points less often correct on such a task (Dell’Acqua et al. 2026). Training people to recognise that boundary is part of change management, not an extra.
Across a portfolio: leaders learn from leaders
A fund has an advantage a single company lacks: several management teams facing the same questions. We use it. Leaders from different portfolio companies learn in cohorts and share what works, so a pattern from one company travels to the next. Before a programme starts, an AI maturity and leadership diagnostic shows where each team stands; LEADBeyond 360° makes visible how a leadership team leads under pressure. Formats fit a deal calendar: one- to two-day bootcamps on the team’s own cases, sprint series with a working pilot between sessions, and 90-minute masterclasses for boards and investment teams.
Frequently asked questions
What is AI change management?
Preparing the people in a company to work with AI so that use cases are actually used and their effect reaches the numbers: choosing cases with the teams, training, clear rules, new routines and visible results.
How do you identify AI use cases in a company?
With the teams that do the work: walk through their processes and ask where they type things twice, search, wait or would never let a machine decide. Rate each candidate on value, feasibility and acceptance, and start where all three are high.
Why do employees not use AI tools?
Often because they doubt the tool helps them, feel unable to check its results, fall back on old routines under pressure or fear for their jobs. BCG finds regular use among frontline employees stalled at 51% and much higher positive sentiment where leaders actively support AI.
What role do leaders play in AI adoption?
A decisive one. In BCG’s 2025 survey, the share of employees who feel positive about generative AI rises from 15% to 55% with strong leadership support. Leaders who use the tools themselves and answer honestly what happens with saved time make adoption stick.
How much training do employees need for AI?
BCG finds regular use sharply higher among employees with at least five hours of training. More important than hours is the content: training on the team’s own documents and processes, including where AI should not be trusted.
Does the works council have to be involved when introducing AI?
In Germany, the works council must be informed about planned AI use (§ 90 BetrVG) and has co-determination rights over technical systems that can monitor conduct or performance (§ 87(1) no. 6 BetrVG). Involving it early saves months. Not legal advice.
Sources
- BCG. AI transformation (capability page). Retrieved 26 October 2026., Boston Consulting Group. Cited for BCG’s split of where the value of an AI transformation comes from: about 10% from the AI application itself, 20% from data and technology, 70% from workflow redesign, culture, governance and human-AI collaboration. A consultancy rule of thumb, not a measurement.
- BCG. (2025). AI at work: Momentum builds, but gaps remain., Boston Consulting Group (2025). Survey of more than 10,600 leaders, managers and frontline employees in 11 countries and regions (June 2025): regular use among frontline employees stalled at 51%; the share who feel positive about GenAI rises from 15% to 55% with strong leadership support, which only about a quarter of frontline employees receive; regular use is sharply higher with at least five hours of training; only one third say they were properly trained.
- BCG. (2025, 26 June). Companies must go beyond AI adoption to realize its full potential. Press release., Boston Consulting Group (2025). Press release on the same survey; cited for 54% of respondents who say they would use AI tools even if not authorized.
- Humlum, A., & Vestergaard, E. (2025). Large language models, small labor market effects (BFI Working Paper 2025-56). Becker Friedman Institute, University of Chicago., Becker Friedman Institute for Economics (2025). Survey of 25,000 Danish workers: chatbot users report saving 2.8% of their work hours; firm-led initiatives (encouragement, in-house tools, training) nearly double take-up for the typical worker, from 47% to 83%.
- Brynjolfsson, E., Li, D., & Raymond, L. (2025). Generative AI at work. The Quarterly Journal of Economics, 140(2), 889–942., Oxford University Press (2025). Field study with 5,172 customer service agents: on average 15% more issues resolved per hour, with the largest gains for less experienced agents.
- Dell’Acqua, F., McFowland III, E., Mollick, E., Lifshitz, H., Kellogg, K. C., Rajendran, S., Krayer, L., Candelon, F., & Lakhani, K. R. (2026). Navigating the jagged technological frontier: Field experimental evidence of the effects of artificial intelligence on knowledge worker productivity and quality. Organization Science, 37(2), 403–423., INFORMS (2026). Pre-registered field experiment with 758 consultants: inside the AI’s frontier 12.2% more subtasks, 25.1% faster, 32% higher graded quality; on a task outside it, 19 percentage points less often correct.
- Ryseff, J., De Bruhl, B. F., & Newberry, S. J. (2024). The root causes of failure for artificial intelligence projects and how they can succeed (Report No. RR-A2680-1). RAND Corporation., RAND Corporation (2024). Interview study with 65 data scientists and engineers; cited only qualitatively for the five main causes of failure. Its failure-rate estimate from secondary sources is deliberately not used.
- Uber Engineering. (2025). Advancing invoice document processing using GenAI., Uber (2025). Self-reported: invoice handling time cut by 70%, process cost by 25–30%.
- Rossum. Maximus customer story., Rossum. Vendor case: invoice processing time cut by 85%; AP team from 10 to 8 people through natural attrition.
- Entrepreneur. (2025). Klarna CEO reverses course by hiring more humans, not AI., Entrepreneur (2025). Report that Klarna began hiring human agents again after its CEO said quality had suffered.
- Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act). Official Journal of the European Union, L series, 12 July 2024., Publications Office of the European Union (EUR-Lex) (2024). Official text. Cited for the role of “deployer” (Art. 3(4)), AI literacy (Art. 4), prohibited practices (Art. 5) and the obligations of deployers of high-risk systems (Art. 26 with Annex III). Not legal advice.
- Regulation (EU) 2026/1744 of the European Parliament and of the Council of 8 July 2026 amending Regulation (EU) 2024/1689 and others as regards the simplification of the implementation of harmonised rules on artificial intelligence (Digital Omnibus on AI). Official Journal of the European Union, L series, 24 July 2026., Publications Office of the European Union (EUR-Lex) (2026). Official text. Cited for the rewording of Art. 4 (measures to support AI literacy) and the new date of the high-risk obligations for Annex III systems (2 December 2027). Not legal advice.
- Betriebsverfassungsgesetz (BetrVG) in der Fassung der Bekanntmachung vom 25. September 2001 (BGBl. I S. 2518), Bundesministerium der Justiz und Bundesamt für Justiz (gesetze-im-internet.de) (2001). In German. Cited for informing the works council about planned use of AI (§ 90(1) no. 3) and co-determination on technical devices that can monitor conduct or performance (§ 87(1) no. 6). Not legal advice.