AI-assisted company valuation, without losing traceability
The valuation methodology the team had mastered for years, turned into a prototype that drafts the report and leaves the signature to the expert.
The challenge
Demand for company valuations is growing, and much of it arrives at a delicate moment: ahead of a restructuring, in a negotiation over a partner joining or leaving, in a deed for the sale of shares, or in court proceedings. These are reports that have to hold up.
The team knew the methods: financial ratios, balance-sheet valuation, multiples, discounted cash flow, and approaches like Berkus or First Chicago for growing companies. The problem was not the knowledge, it was the time. Every report was built from scratch, by hand, by the same few people.
And that same team was carrying many open files at once, with interruptions that cannot be planned for and progress tracked on tools that had been outgrown. With that bottleneck, the only way to serve more demand was to hire more people with the same experience.
Approach
We started with the diagnostic, and with a technology audit carried out on their own premises: which tools existed, which were barely used, where the data lived, how backups were made, and how much mental load was going into work a system could carry.
Then came the hard part: reviewing old valuation reports to extract the internal logic. Which data is always requested, where it comes from, which methods apply in each case, what structure the report has, and which part of it is calculation and which part is judgement.
With that logic made explicit we built a prototype: a private AI agent that takes the annual accounts, applies the methods and returns a structured draft report. The expert reviews it, corrects it and signs it. At no point does the machine decide.
Before building anything there was a decision to make: low-cost tool, premium platform, or infrastructure for third parties. That is not a technical question but a business one, and it changes the entire product.
- On-site technology audit: current state, desired state, and what falls away in between
- Review of old reports to make the calculation and judgement logic explicit
- Structured ideation across the innovation funnel, using the SCAMPER technique
- An upfront positioning decision: tool, platform or infrastructure
- A prototype built on a private AI agent that processes the accounts and drafts the report
- Two report levels: a basic one for lead capture and a full one with traceable calculations
- A dynamic component: change key variables and watch the valuation move
- Model comparison by task, keeping the cost of each query under control
Key outcomes
1 week
from decision to a working prototype, tested against a client's real annual accounts.
6
valuation methods built into the model: ratios, balance sheet, multiples, discounted cash flow, Berkus and First Chicago.
2
report levels defined: a basic one for web enquiries and a full one showing the workings that expert-witness and court use demand.
0
lines of bespoke development before validating: the prototype was assembled by configuring a private agent, so demand is tested before committing months of programming.
It is worth saying what this is and what it is not: a prototype validated against real data plus a roadmap, not a launched service with revenue behind it. What has been proven is that the report logic can be automated without losing traceability, and that it can be done in days at minimal cost rather than committing months of development before knowing whether the demand is there. The audit also left a list of less glamorous but equally necessary homework: sort out where the data lives, make real use of the tools already being paid for, and systematise case tracking.
What the machine does not do
A valuation report can end up in a courtroom or in a notarial deed. That is why traceability was not a technical whim: if the reasoning cannot be followed number by number, the report is worthless. The prototype writes the draft and takes away the repetitive work; the qualitative analysis, the reading of the sector and the signature still belong to a person. Artificial intelligence here does not replace the expert: it removes the hours that added no judgement.
Got a process that hurts?
Tell us about it in half an hour. No forty-page deck. If we are not the right people, we will say so and point you towards someone who is.