Founder Track Record · Case Study 02

The Copilot trial failed. Here is what worked instead.

An £80bn AUM private equity business wanted to know whether AI was a real shift or simply hype.

“The tool was not the problem. The delivery model let it down.”
£80bnapproximate AUM
<50initial trial users
12week first trial
3models evaluated

Attempt one

A low-friction trial with no definition of success.

Microsoft Copilot was the obvious first move. It required minimal new legal review, fitted the existing technology estate, and could be switched on for a trial group without committing more than 500 users.

Fewer than 50 participants were selected for their roles, technical confidence and influence. Twelve weeks later, the trial had failed. Users described Copilot as a nice tool, then returned to their old ways of working.

The failure came from three specific mistakes. No success metrics were agreed before launch. Users received licences without meaningful training. The initiative was led by technology rather than the business, so nothing sustained engagement once the initial excitement faded.

The lesson

Giving people access to AI is not transformation. Without a business problem, education and accountable ownership, even a strong tool becomes an expensive novelty.

Attempt two

Rebuild the programme around value, not licences.

The second attempt began with a structured evaluation, clear business ownership and measures agreed before rollout.

  1. 01

    Evaluate for fit

    Compared OpenAI, Anthropic and Google against integration, data and trust requirements rather than market hype.

  2. 02

    Document the decision

    Recorded why OpenAI and ChatGPT were selected, and why alternatives were ruled out, so the rationale could withstand future scrutiny.

  3. 03

    Prioritise use cases

    Applied RICE scoring and a dependency matrix, then visualised the portfolio on a live dashboard sized by value and sequencing importance.

  4. 04

    Put the business in charge

    Created an Operational Review Board of selected C-suite leaders to approve use cases and connect each one to corporate objectives such as margin improvement.

  5. 05

    Build peer support

    Introduced an internal Champions Network, recurring support, drop-in sessions and Town Hall demonstrations led by real superusers.

  6. 06

    Measure what matters

    Defined adoption and engagement separately, with named and dated measures covering active seats, daily use, custom assets and repetitive time removed.

The results

From passive licences to operational use cases.

The second programme produced materially stronger engagement because people were shown what AI could do, supported while learning, and connected to approved business problems.

Investment memos

A queryable knowledge base from siloed documents

Structured extraction across saved investment memos enabled backtesting of older investment assumptions and was estimated to remove more than two days of processing per memo.

Deal analysis

Structured data from 100+ page GP reports

High-accuracy extraction of investment-analysis data substantially reduced the manual effort previously required to process long reports.

Senior executives remained engaged throughout and actively brought forward new ideas for consideration.

Why this matters

AI success depends more on behaviour and change than technology choice.

Corbelle starts before technology is purchased, with the single most important problem the organisation needs to solve.

Start a conversation