AI in theQuoting Office

1. Where the time goes on a tender pack

On a full tender pack, most of the days go into reading the documents rather than pricing them. A pack for a process-plant package might run to a couple of hundred pages, with a specification, a set of drawings, a schedule of quantities, preliminaries, a pre-qualification questionnaire and a form of tender, and the same requirement appearing in three of them in slightly different words. Before anyone can price, someone has to read all of it, find the questions, the standards and the dates, and build a list of what is being asked for.

Once the list of things to price exists, the pricing itself is usually a matter of hours. The questionnaire answers are written fresh each time, even when the firm has answered the same question forty times before, because the previous answers are in forty different folders.

Current AI tools do that part well. Given a long document, they pull out what it says and put it in a list. They cannot tell you whether the list is complete, they read drawings badly, and a price must never come from one of them. The method in this playbook has the tool read and list, and an estimator check every item against the page it came from, price it, and decide whether to bid.

2. The rule set

Agree these before anyone uses a tool on a live pack. Put them on one page and pin it above the estimators' desks.

The tools may	The tools may not
Read the pack and produce a structured summary	Produce a price, a rate, a margin or a programme duration
Extract every question in the questionnaire into a list	Read a drawing, or state a dimension that comes from a drawing
List the standards, approvals and certificates referenced	Decide whether to bid
Extract dates, deadlines and submission requirements	Send anything to a client
Draft answers to the questionnaire from your own approved answer library	Invent a reference, a certificate number, a customer name or a figure not in the library
Rewrite an estimator's draft for clarity, at the same length or shorter	Be given confidential pricing data, customer pricing, or drawings marked as controlled
Flag inconsistencies between documents in the pack	Be the only reader of any document

Three more rules apply whatever the task. The tool gives a page or section reference for every number, date and requirement it reports, and an estimator opens that page and ticks it before the figure goes near a spreadsheet. The tool sees only what it needs: the specification and the questionnaire go up, the pricing schedule with your rates does not, and drawings marked confidential stay out unless the contract allows it and the tool's business terms keep the data private. The senior estimator owns the prompts, the answer library and these rules, and changes go through that person.

Choosing a tool

Any of the current assistants from the major providers will do the reading tasks in this playbook. Whichever you choose, use a business subscription rather than a free personal account, get written confirmation from the provider that your uploads are not used to train their models, and set the account up so that each estimator logs in as themselves. Keep it to one tool for the office so that the prompts and the checks are the same for everyone.

From section 3, the pack summary prompt: "For every item give the document name and page number. If an item is not stated anywhere in the pack, write 'not stated'." The wording of each of the fifteen prompts has been adjusted so that the tool reports references and admits what it cannot find.

Sources for the free sections: Axi Growth, observations from estimating and quoting work with engineering firms, 2025 to 2026; provider terms for business accounts on the major AI assistants. General guidance, not legal or contractual advice.