On this page
What goes wrong when everyone uses their own AI?
- Every junior has their own prompts and settings. Two IMs from the same firm read like the work of two firms.
- Nobody knows which text or table came from which model.
- What one person worked out stays in their chat history.
- Figures get “calculated” in a chat window, and nobody sees the calculation.
What does one fixed workflow look like?
A workflow is a fixed sequence of steps for recurring work. For an IM or teaser it looks like this:
- The seller’s documents go into a fixed folder, in a fixed form.
- Each document has a fixed template, in the firm’s structure and tone.
- The model drafts each chapter from the source documents.
- Tables and ratios come out of a tested calculation step.
- The team reads, corrects and approves before anything goes out.
Everyone works with the same steps. A new junior has nothing to invent, and last year’s IM can be traced back.
Why are the calculations a separate step?
A language model does not calculate by itself. It writes code that calculates, or it estimates. The more a company deviates in its accounting policies, the higher the risk of an error.
So we build calculations as fixed steps:
- the same formula for adjustments and ratios, every time;
- tested before the team relies on it;
- every figure in the IM traces back to its source.
Who checks the output?
The team. Every workflow ends with a check by someone at the firm. The model delivers the first draft; the judgement and the sign-off stay with you.
Who owns what gets built?
You do. The workflow runs in your own environment and moves with you if you switch to another AI model provider. You are not tied to us or to one model. We put that in the engagement confirmation.
What else works as a fixed workflow?
An example from my own work: a two-page company profile, from nothing but a company name.
- website, owners and structure from Company.info;
- competitors from Gain.pro;
- a news check, so you do not miss a sale that has already been announced.
The analysis is prepared, the judgement stays with the team. What was half a day to a day of work becomes checking and polishing.
Do you need to hire an AI specialist for this?
For a boutique, usually not. The gain is in a few well-configured work processes the whole team uses every day. That needs no extra role.
Where do you start?
Map the work from mandate to closing. For each step, look at what takes time, what comes back every time and where errors creep in. Start with one or two steps that recur often, such as the teaser or the first draft of the IM.
How an engagement runs from there is on How we work. Are you even allowed to put deal documents into such a tool? See Deal documents in ChatGPT.