What AI can do in a tender, and what a person should keep doing

What AI already does well: reading, structuring and writing

Reading the specifications in full

A set of specifications and its annexes runs to hundreds of pages, with the requirements scattered between the administrative specifications (PCAP), the technical specifications (PPT) and the annexes. An AI system built for the job can turn all of that into structured requirements: the subject of the contract, the lots, the budget, the deadlines, the award criteria and their formulas, the solvency required, the guarantees, the penalties, the administrative documentation, the page limits and the potential grounds for exclusion. That is not a summary. It is a list of what has to be met, what scores, and what has to be delivered.

Drafting against the award criteria

With the award criteria in front of it and the company's real documentation to hand, AI can propose the table of contents, weight each section by the points on offer, and draft the methodology, the work plan, the team, the schedules, the resources, the quality arrangements, the risks and the improvements. What sets this apart from generic text is where it starts: every section answers a criterion and rests on something the company can actually evidence. How this changes the technical proposal is the subject of technical proposals with AI.

Reviewing before submission

Requirements left unanswered, contradictions between the technical proposal, the financial offer and the administrative documentation, incomplete annexes, page limits exceeded, claims with nothing behind them, criteria not fully covered, grounds for exclusion. That review is mechanical, exhaustive and dull, and it is exactly the kind of work where a tired person on the last day of the deadline slips and a system does not.

What it can do once it knows your company

This is the difference between pasting a set of specifications into a general-purpose AI model and using a system built for public procurement. A general-purpose model can help you read or draft a piece of text on the day, but it does not know what your company sells, which projects it has delivered, what experience it can evidence, what team and certifications it has, which contracts it wants and how much risk it is prepared to carry. Without that context, the search returns noise, the technical proposal invents capabilities, and the decision to bid is taken in the dark.

• Finding and ranking opportunities against the company's own criteria, not against keywords.
• Assessing the fit of each tender: whether the company can bid, what is missing, what risks it carries and how much work it would take.
• Working through the financial offer: the scoring formulas, price scenarios, sensitivity, the risk of an abnormally low bid, all within the minimum margin the company sets.
• Learning from every result: the score obtained, the competitors' scores, the winning bid, the price gaps, what worked and what to change.

What a person should keep doing

Some decisions commit the company, and no tool, however good, should take them on its own. Bidding for a tender or passing on it. Putting a question to the contracting body. Setting the final price. Submitting the proposal. A well-designed system leaves each of those gates configurable: it can pass through them on its own where the company has decided it should, or stop and wait for a person to approve. What is not acceptable is a system that walks through them without anyone having decided.

There are also two limits that have nothing to do with technology. AI can explain the general concepts of public procurement and analyse the information in a file, but it is no substitute for legal advice where a binding legal interpretation is needed. And no tool, no consultancy and no company can guarantee that a public administration will award a particular contract: that decision belongs to the contracting body and turns on the offers submitted. Be wary of anyone who promises otherwise.

How all of this fits into a real process

The order matters. First the company's context; then the search and the fit assessment; then the reading of the specifications, the strategy and the drafting; then the review; and, once the offer is in, the follow-up and the learning. Each stage feeds the next, and the last feeds the first of the following tender. That is what lets the system improve with every file instead of treating each one as a job on its own. It is how Carabela works, and you can follow it stage by stage in the complete system or read about it in AI for tenders.