AI agent for management consultants
Recommendation Evidence Chain Agent
Make every recommendation in the deck traceable to data the team can show.
What it does
A client asks, where does this number come from, and the team scrambles. Recommendations in a deck often rest on a chain: source data, analysis, finding, recommendation. Somewhere the chain breaks. This agent reads the deck and lists each recommendation. For each it traces the supporting numbers to the analysis files and then to the source data. It flags weak links, such as a number in the deck that does not appear in the model, an old data extract or a conclusion that needs an assumption nobody wrote down. It asks the team for more analysis and checks again. The lead approves before the deck goes to the client. Edge case: a number rounded differently in two slides is flagged for consistency.
How it works
Follow the arrows from top to bottom. The orange dashed arrow is the loop: when a check fails, the agent goes back and tries again.
Read the steps as a list
- Draft deck is ready for review
- Extract each recommendation and its supporting numbers
- Find the matching figure in the analysis files and source data
- Does each number match the analysis and the source?If not: List the mismatches and ask the analyst to correct the deck or the model. Back to step 3.
- Check the date and scope of each source
- List assumptions the recommendation depends on and whether they are written down
- Draft a request for more analysis for each weak link
- Are all weak links closed after the team's update?If not: Rerun the trace on changed items and list any that remain. Back to step 3.
- Lead approves the deck for the clientThe agent waits here for your OK.
- Evidence chain table
How it decides
Follows each claim back to a source file; a claim with no source, an old source or a mismatch of more than 1% with the model is flagged as a weak link.
- Mismatch above 1% between deck and model is a weak link
- Source older than the agreed data date is flagged
- Assumption not written down is flagged
- A recommendation with no chain is marked unsupported
Make it yours
Every agent is a starting point. You choose these settings for your own situation.
- Tolerance for number mismatches (default 1%)
- Data date required
- Which decks to check
- Assumption list format
- Who fixes weak links
What keeps you in control
It always asks you first
- Release of the deck
- Any removal of a recommendation
- Any new analysis commissioned
Hard limits
- Never change the deck or models itself
- Never mark a claim as supported without a source
It stops when
- Done: all recommendations traced
- Stop: data confidential, not accessible
- Stop: lead withdraws recommendation
Set it up
We guide you through the set-up, step by step
Members get the full set-up guide for this agent. No technical skills needed: you copy, paste and upload.
- One set of instructions to paste into your AI, with the clicks for ChatGPT, Claude, Microsoft 365 Copilot, Gemini and Grok
- The agent then walks you through connecting your own data, one source at a time
- A downloadable copy with the flow chart, the rules and the full guide
