When Estimating Lives in One Person's Head: Capturing It Before They Leave

How to get estimating knowledge out of one person's head in an MEP firm: start from the old pricing spreadsheet and the senior estimator, then use AI.

Written by the Miriad team with AI assistance. Figures link to their sources; the example is illustrative. Updated 2026-10-21.

TL;DR: In most MEP firms, estimating knowledge sits in two places: an old pricing spreadsheet and the senior estimator who built it. The spreadsheet holds the numbers; the person holds the reasons. Capture both. Interview the estimator against real past bids, let AI turn the recordings into written rules with a source for each, and have the estimator correct every line. Do it before notice is given, not after.

Every building services firm of a certain age has the file. It is called something like "PRICING MASTER v7 FINAL", it has tabs nobody else opens, and the labour units in it were last argued about years ago. Next to it sits the person who understands it. When that person is on holiday, bids slow down. When they retire or leave, the firm finds out how much of its pricing was never written down. This article is part of our guide on how an MEP contractor can use AI to win more bids.

Where does estimating knowledge actually sit in an MEP firm?

It sits in two places: the pricing spreadsheet and the senior estimator's memory. The spreadsheet holds labour units, material markups and a few standard allowances. The estimator holds the rest: which GCs underspecify, which plant rooms always need extra access time, which exclusions save a job. Capturing only the spreadsheet saves the numbers and loses the judgement.

There is a third, smaller source: past bids. Every proposal, quote comparison and loss email records a decision that was made for a reason. They are scattered across folders and inboxes, but they are the evidence that turns an estimator's habit into a rule someone else can follow.

Why it is urgent is visible in the hiring data. In AGC's 2025 workforce survey of firms with $50.1 million to $500 million in revenue, 75% of firms with estimating openings reported difficulty filling them (AGC, 2025). Among candidates construction firms do find, 57% of firms cite missing skills (NCCER with AGC, 2025). A new estimator will need your knowledge written down.

What is in the old pricing spreadsheet, and what is missing?

The spreadsheet usually contains labour units per item, material markups, crew rates and a few allowances such as access or commissioning. What it lacks is the reason behind each number, the date it was last checked and the conditions where it does not apply. Without those, a new estimator applies the numbers to the wrong jobs.

Go through the spreadsheet tab by tab and mark each value in one of three ways: a fact with a source (a supplier price list, a payroll rate), an assumption (a labour unit the estimator believes is right), or unknown (nobody remembers why it is there). That simple marking is often the most useful hour in the whole exercise, because it shows how much of the firm's pricing is assumption.

This is the data problem contractors name when they talk about AI. In research by Dodge Construction Network and CMiC, only 26% of contractors said their data was in good shape, the biggest barrier to AI in the report (Roads & Bridges, 2025). An unmarked spreadsheet is exactly that kind of data.

How do you interview the senior estimator?

Interview against real past bids, not in the abstract. Pick a few recent bids, won and lost, open the proposal and the quote comparison, and ask the estimator to explain each decision: why that labour allowance, why that exclusion, why that markup. Record the conversation. Specific jobs bring out rules that a general question never will.

Questions that work well:

  1. On this job, which labour units did you change from the spreadsheet, and why?
  2. Which exclusions do you always include for this GC or this job type?
  3. What made you add an allowance here: access, working hours, existing systems?
  4. Which supplier would you not use for this item, and why?
  5. Where in the spec do you look first for commissioning and submittal requirements?
  6. What did we get wrong on the last job like this, once it was built?
  7. Which bids would you never price again?

Keep each session short and tied to one or two bids. Over a few weeks, the recordings cover most of what the estimator does without thinking.

How can AI turn interviews and spreadsheets into something usable?

AI can transcribe the recordings, pull out every stated rule, link each rule to the bid it came from, and organise the rules by system and stage. The output is a draft rulebook. The estimator then corrects it line by line, because AI will sometimes turn a remark into a rule or miss a condition.

A working instruction for the transcript:

From the interview transcript below, list every estimating rule the
estimator states. For each rule give: the rule in one sentence, the
system (heating, cooling, ventilation, plumbing, controls), the bid it
came from, any condition where it does not apply, and the exact quote
from the transcript. Mark each rule as FACT, ASSUMPTION or UNCLEAR.
Do not add rules that are not in the transcript.

The source column is what makes the rulebook trustworthy. A rule with a quote and a bid reference can be checked and challenged. A rule without one is just another unexplained number, like the ones in the old spreadsheet.

Most contractors will be doing this without a manual: a 2025 Houzz survey found 92% of contractors using AI had no formal AI training (Houzz, 2025). Write the instruction down and use the same one for every session.

Example: capturing a senior estimator's ventilation rules

Example (illustrative, not a real case): A 20-person mechanical contractor's senior estimator plans to retire within two years. The owner books a weekly half-hour session. In the first, they open two recent ventilation bids, one won and one lost, and she explains her choices: why she increased labour on ductwork in occupied buildings, why she always excludes builder's work openings in existing masonry, and why one GC's bid packages always need an allowance for missing fire damper details.

An AI assistant turns the transcript into a list of eleven rules, each with the bid reference and her exact words. She deletes two that were offhand remarks, rewrites one condition, and marks three as assumptions she wants the junior estimator to test on the next jobs. The rules go into a shared folder next to the pricing spreadsheet, with a column in the spreadsheet pointing to the rule behind each changed labour unit.

After several sessions, the junior estimator prices a ventilation bid using the rulebook and the senior estimator reviews it. Her corrections become new rules. That review loop is the real transfer.

How do you keep the knowledge current?

Make updating part of the bid cycle. After every award or loss, add one line to the rulebook if a rule was confirmed, broken or new. After every completed job, record what the actual labour was against the allowance. Review the rulebook with the senior estimator at fixed intervals, and date every change.

Store the rulebook with the comparison tables and proposals, so that rules, prices and outcomes sit together; the table format is in turning subcontractor quotes into one comparable table. If you want to see what the missing knowledge costs on each bid, the cost of a lost bid has a worksheet.

This is the kind of work our Business Brain Starter Kit is being built for: a workbook and a chain of prompts that turn what a company knows into plain files where every statement is marked as fact, assumption or unknown. Get notified when it launches. The full bid cycle is in the MEP contractor guide to AI and bids.

FAQ

When should we start capturing an estimator's knowledge?

Now, while the estimator is still enjoying the job. Capturing knowledge during a notice period rarely works: there is too little time, and the person's attention is elsewhere. Short weekly sessions tied to real bids, spread over months, produce a far better rulebook than a rushed handover. Treat it as normal work, not as preparation for someone leaving.

Will the senior estimator resist being interviewed?

Some will, especially if it feels like being replaced. Frame it honestly: the firm needs to reduce the load on one person and train the next estimator, and the rulebook carries their name as the source. Let them correct and approve every rule. Most experienced estimators are glad when someone finally asks why they price things the way they do.

Can AI learn our pricing from old bids automatically?

Not reliably. AI can extract figures and patterns from old proposals and spreadsheets, but it cannot tell which figures were deliberate and which were mistakes, or why a labour unit changed on a particular job. Use AI to organise what the estimator explains and to link rules to sources. The estimator's corrections are what make the result usable.

Where should the estimating rulebook be stored?

In one shared place next to the pricing spreadsheet and past bids, in plain files that anyone in the estimating team can open without special software. Every rule should carry a date, a source and a status such as fact or assumption. Avoid storing it only inside an AI tool, because the knowledge should belong to the firm and stay readable.

Sources

Related articles

Products from Miriad

Business brainBusiness Brain Starter Kit

Stop explaining your business from scratch. Write it down once, with a source next to every line.

49 USD

Get notified

See all prompt packs