AI for Tender Analysis

Tender analysis is the work of taking a pile of returns that all claim to be for the same package and working out what you are actually being offered. It is not adding up columns. It is finding the gap between what the tender documents asked for and what each bidder has priced, then putting a number on that gap so the comparison means something. Get it wrong and you award to the cheapest bid, which turns out to be the one that excluded the most. Everyone finds out in month four. The prompts on this page cover the whole arc: normalising returns onto a common basis, interrogating a suspiciously low bid, scoring quality, and drafting a recommendation you can defend in a room. WHAT AI IS ACTUALLY GOOD AT HERE Reading. That is the honest answer. The bulk of tender analysis is reading four covering letters, four sets of qualifications, four programmes and four method statements, then noticing that bidder C mentioned something the other three did not. It is slow, it is detailed, and it is exactly where attention fails at 6pm on a Thursday. AI does not get bored on page 40. Give it four sets of qualifications and ask for one table listing every exclusion by bidder, flagged where an item appears in some bids but not others, and you get in two minutes what takes an afternoon. That flag is the whole game. When one bidder excludes the builder's work in connection and the other three priced it, one of two things is true: they spotted something, or they are buying the job. Either way you now know to ask. WHERE THE JUDGEMENT STAYS WITH YOU Whether an exclusion matters. AI can find it. It cannot tell you that this client always argues about BWIC, or that the last time you accepted a drainage exclusion it cost £60,000. What a low bid means. A tender 12 percent below the others is either a mistake, a different scope, or a contractor who wants the work badly. Those three have completely different consequences and no amount of prompting will tell you which one you are looking at. You find out by asking them. The award itself. On public sector work the panel scores against published criteria and the record has to show the panel did it. An AI score is an unaccountable number in a document that may need to survive a challenge. WHEN TO USE THESE PROMPTS Use them the day the returns land, while there is still time to go back to bidders with questions. Tender analysis done in the last 48 hours before award is not analysis, it is arithmetic with a deadline. Use them on subcontract packages as much as main contract tenders. The prompts do not care about the value, and a £180,000 groundworks package with a buried exclusion will hurt a job just as much as a main contract error. Use them when there is a gap you cannot explain. If bidder B is £200,000 light and you do not know why, structured interrogation of their breakdown against the others is exactly the job. WHEN NOT TO Do not use them on a single quotation. There is nothing to compare. You want a check against your own estimate, not a comparison matrix. Do not use them where the tender documents were vague. If the ITT did not say who takes the drain-down, four bidders will each have assumed differently and you cannot normalise your way out of it. That is a re-tender or a formal clarification to all bidders, not an analysis problem. Fixing an ambiguous ITT at evaluation stage by adjusting the bids is how challenges start. Do not paste live tender sums into a consumer AI tool without checking. Bids are commercially sensitive, usually held under ITT terms that say so. Check the tool's data handling, your client's rules and your own policy first. On public sector work, assume the process may be scrutinised in detail. And do not use these prompts to justify a decision you have already made. An analysis written backwards from a chosen answer is visible, usually because the quality scores do all the work and none of them are evidenced. FREQUENTLY ASKED QUESTIONS What is the difference between tender analysis and tender evaluation? Roughly: analysis is working out what the bids actually say, evaluation is scoring them and choosing. Analysis is normalisation, arithmetic checks, exclusions and anomalies. Evaluation is applying the published criteria and weightings and reaching a recommendation. You cannot evaluate well without analysing first, which is why so many evaluation reports rank bids that were never comparable in the first place. How much of a price gap should worry me? There is no universal number, but a bid more than about 10 percent below the next one deserves a specific explanation rather than gratitude. Look first at whether the scope matches, then the arithmetic, then the preliminaries and programme. Genuinely keen pricing shows up as thin margin across the board. A mistake usually shows up in one section that is wildly out of line with the others. An exclusion shows up as a section that is missing. Can AI read a tender bill or a priced schedule? It can read one, and it will make mistakes on a big one that you would not make. It is at its best on prose: covering letters, qualifications, method statements, programmes. It is weakest on long numeric documents, where it will confidently mis-transcribe a rate or lose a line. Use it on the words and use your own tools on the numbers. Should I tell bidders I used AI in the analysis? Nobody expects a disclosure for using a tool to help read documents, any more than for using a spreadsheet. What matters is that the decision was made by accountable people against published criteria, and that you did not put confidential pricing somewhere it should not go. If you cannot say honestly that the panel made the decision, that is the problem, not the disclosure. What should I do first when the returns land? Arithmetic check, before anything else. Errors change the ranking and finding one after you have normalised means doing the normalisation again. Then qualifications and exclusions, then normalisation, then scoring. In that order. Most of the pain in tender analysis comes from doing those four things in the wrong sequence.