# Tenders: analyse supplier bids in hours, not days

> How an AI agent reads supplier bids, pre-fills your scoring grid and flags points to clarify, without taking the decision away from the buyer.

Amine Jabri · BYC21 & Co · 2026-11-16 · https://www.byc21.com/en/insights/ai-tender-analysis/

A tender often means several bids of dozens of pages each, in different formats, to be compared criterion by criterion. The buyer spends days **looking for information** before being able to **assess it**. That is precisely the part an AI agent can take on.

## What takes time in bid analysis

- **Reading and finding**: each supplier presents its answers its own way, in a different order.
- **Filling in the grid**: transferring prices, lead times, service commitments and technical answers into a comparable table.
- **Spotting gaps**: a requirement not addressed, a reservation slipped into an appendix, a pricing condition that changes everything.
- **Preparing clarification questions** to send to the bidders.

This work is essential, but it does not require the buyer’s expertise. That expertise comes in for the trade-offs and the negotiation.

## What the agent actually does

Based on your specifications and your evaluation grid:

1. **It reads each bid**, whatever the format, and finds the elements that answer each criterion.
2. **It pre-fills the scoring grid**, with **the source passage** in the supplier’s bid for each score, so the buyer can check it in one click.
3. **It flags points of attention**: uncovered requirements, reservations, inconsistencies between the stated price and the conditions, vague commitments.
4. **It suggests clarification questions** for each bidder.
5. **It produces a comparative summary**, which serves as the basis for the selection committee.

## What stays human

The agent prepares, **the buyer decides**:

- the weighting of the criteria and how they are interpreted;
- the final score, which the buyer approves or corrects;
- exchanges with suppliers and the negotiation;
- the choice of the winning bidder, and its justification.

Every proposed score links back to its source. Nothing is a "black box": if the agent got something wrong, the buyer sees it and corrects it.

## What do we need?

- **a recent or ongoing tender**, with its specifications, its grid and the bids received;
- ideally, **the analysis your teams carried out at the time**, to compare the agent’s result with a reference;
- **a buyer available for a few hours** to test and give feedback.

Supplier bids are confidential: they stay in your environment, and processing follows your security rules.

## How to measure the gain

| Metric | What it tells you |
|---|---|
| **Analysis time per bid** | Does the buyer save time? This is the reference metric. |
| Share of scores approved without correction | Is the pre-filling reliable? |
| Points of attention detected | Does the agent catch what a buyer would have spotted? |
| Time from receiving bids to decision | Does the purchasing cycle get shorter? |

Comparing with an analysis your teams have already done is the most telling method: we know exactly what the agent found, missed or added.

## The risks, and how we control them

- **Information misread or made up**: every element is linked to its source, and the buyer approves.
- **Bias in scoring**: the agent applies your grid; weighting and decisions stay human and justified.
- **Confidentiality of bids**: processing in your environment, limited access, no reuse of the data.
- **Rules specific to public procurement**: the tool supports the analysis; it replaces neither the committee nor the rules of the procedure.

## How to get started

- **The Workshop (2 hours, no commitment)** to choose the test tender and check that procurement is the right first topic.
- **The Proof Sprint (5 days)** for an agent that analyses the bids of a real tender. On Monday, we set the criterion; on Wednesday, the buyer tests; on Friday, you compare the result with your reference analysis and decide.

If the gain is demonstrated, we integrate the agent into your purchasing process. If not, we stop, with the measurement and our recommendations.
