Find bids through invitations from contractors you know, bid boards or plan rooms, and public procurement portals, then screen each one before you estimate. Score it on fit, margin, competition and client, and use AI to summarise the tender documents into a one-page brief. Price only the bids that clear your threshold, because a full estimate costs hours.
Most bids you could price are not worth pricing. A full estimate eats days of estimator time, and a contractor that prices everything wins some jobs it should have skipped: the ones where the client pays late, the retainage is long and fifteen firms are racing to the bottom. A screen takes about twenty minutes and decides whether the estimate happens at all. AI does the reading; you do the deciding.
Where construction bids come from, and what each source asks of you
Finding bids is the easy half of the problem. There are six common routes, and a firm usually needs three of them working.
| Source | How bids reach you | What to check |
|---|---|---|
| Contractors who know you | Email, a phone call or an invitation on a bid platform | Reply quickly, and log every invitation, including the ones you decline |
| Bid boards and plan rooms | Searchable project lists with documents and an invitation inbox | Free profile or paid plan, the radius and trade filters, whether invitations from elsewhere can be added |
| Public procurement portals | Notices published by public bodies, often with alerts by category | Registration steps, prequalification, bond and insurance rules |
| Bid-matching tools | One search across many portals, with scoring | Which portals are covered, the monthly price, whether the free tier is usable |
| Architects, owner's representatives and project managers | Direct invitations once they trust you | Time spent on relationships and not paperwork |
| Past clients | Repeat work and re-bids | A calendar reminder for each client's next likely project |
Three product names come up often, and their own pages say the following (read on 1 November 2026). PlanHub says a free profile lets you receive direct invitations from general contractors and bid on projects within a local radius, and does not state that tier's limits. Its paid Premier plan is priced yearly by radius, $1,999 for 50 miles, $2,349 for 100 and $3,299 for 200, and the page lists unlimited bids, filters for location, project type, construction type and timing, and viewing or downloading plans. A "Project Evaluation" tool, described as PlanHub AI, is labelled new, but the pricing page does not say which plan includes it.
Autodesk's BuildingConnected page describes Bid Board Pro for subcontractors: a plan room that adds new opportunities to your bid board, bid forwarding that pulls key details from invitations received elsewhere, and tracking of invitations and due dates. It states no price. constructionbids.ai lists a free tier that lets you browse bids, with five bid views a day, then paid plans at $59, $79 and $99 a month. Fit scores and a go/no-go card start at the lowest paid plan, AI risk analysis of the specification sits only on the top plan, and there is a seven-day trial of that plan with a card required.
None of that is a recommendation, and I have no link with any of them. Start with free tiers and trials, and judge them by whether they bring bids you would genuinely price. Whatever tool you use, its scoring is generic. It cannot know your capacity this month, your history with a client or your margin target, which is why you need a screen of your own. The general bid/no-bid tutorial covers the same idea for any small firm, with a scorecard and a capacity check; this one builds on it for construction work.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowA score with weights you can defend
A screen works when it is fast, written down and the same for every bid. Five criteria cover most of what separates a good bid from a bad one. Each is scored from 1 to 5, multiplied by its weight, and the weights add to 20, so the top score is 100 points.
| Criterion (weight) | A score of 5 looks like | A score of 1 looks like |
|---|---|---|
| Fit (5) | Your trade, your usual size, your usual type of client | Outside your licence or experience, or far larger than anything you have run |
| Margin (5) | You can price at or above your target margin on fair terms | The budget, retainage and delay penalties leave no room for profit |
| Client (4) | A repeat client that pays on time | A history of late payment, disputes or changed scope without pay |
| Competition (3) | Few invited bidders, or you are the incumbent | An open bid with many bidders where lowest price wins |
| Capacity and risk (3) | Estimator and crew free, simple terms, bond and insurance in hand | A bond, programme or insurance limit you cannot meet |
Before any scoring, apply a short list of knockouts. These are automatic passes whatever the score: a bond, licence or insurance requirement you cannot meet; a due date that leaves too little time to price properly; scope outside your licence; a client known for not paying within terms; and contract terms you would not accept even after taking advice. A bid that fails a knockout is not scored at all, which is the main way a screen saves time.
Then use three bands. A score of 75 to 100 means price it. A score of 55 to 74 means send a written clarification first, and price only if the answer lifts the weakest score. Below 55, pass. The thresholds are a starting guess; the bid log described further down is how you replace the guess with your own data. If you want to see how margin is judged after the fact, the construction job costing tutorial shows how to find which finished jobs actually made money, which tells you what a good margin score means for your firm.
Three tenders, scored with numbers
A drywall and ceilings subcontractor has three tenders on the desk on the same morning. All details are illustrative.
- Tender A: an invitation from a general contractor the firm has worked for six times, for partitions and ceilings in a two-storey clinic fit-out. Payment in 30 days, 5% retainage, six-week programme.
- Tender B: a public school refurbishment, open to any registered bidder. The portal shows 14 firms have downloaded the documents. Lowest compliant price wins, retainage is 10% until completion, and there are daily penalties for late finish.
- Tender C: a design-build package where the drawings are marked "indicative", the scope says "all works necessary", and payment depends on the owner paying the main contractor.
| Criterion (weight) | A: score, points | B: score, points | C: score, points |
|---|---|---|---|
| Fit (5) | 5, 25 | 4, 20 | 3, 15 |
| Margin (5) | 4, 20 | 2, 10 | 3, 15 |
| Client (4) | 5, 20 | 4, 16 | 2, 8 |
| Competition (3) | 4, 12 | 1, 3 | 3, 9 |
| Capacity and risk (3) | 4, 12 | 3, 9 | 2, 6 |
| Total | 89 | 58 | 53 |
| Decision | Price it | Clarify first | Pass |
Tender A is easy: a known client, a known trade and few rivals. Tender B lands at 58, the clarify band. The weak points are margin and competition, and the useful questions go to the client in writing: does the retainage really run to completion, and are the delay penalties capped? If the answers soften the terms, the margin score might rise to 3 and the capacity and risk score to 4, lifting the total from 58 to 66, still short of 75, which says price it only if the crew would otherwise be idle. Tender C scores 53 and fails on client and risk. A scope that says "all works necessary" with pay-if-paid terms is a bid where you can lose money while doing everything asked. You could still ask for the scope to be pinned down, but pass unless the answer is excellent.
Do not read the totals as truth. The scorecard exists to force the conversation before the estimate, and a bid that scores 58 but fills a gap in the programme may still be worth pricing. The point is that you decide on purpose, with the reasons written next to the number.
A bid brief: have AI read the tender pack into one page
Reading a 200-page tender pack for the six facts that decide whether to bid is exactly the task AI does quickly and people do badly at four in the afternoon. The prompt below asks for the same ten items every time, requires a page or clause reference for each, and gives the model a safe way to say it found nothing. That last rule matters most.
You are helping a [trade] subcontractor decide whether to price a tender. Read the attached documents and fill in this brief. For every item give the page or clause where you found it. If you cannot find it, write NOT FOUND. Do not guess and do not fill gaps from general knowledge.
1. Project and client: who is the client, and who takes questions?
2. Dates: bid due (date and time), last date for questions, mandatory site visit or pre-bid meeting.
3. Scope for [trade]: sections that name my trade; anything marked "by others" or "by owner".
4. Contract type: lump sum, unit price, cost plus or other.
5. Money terms: payment terms, retainage and when it is released, penalties for late completion, bonds or guarantees, insurance limits.
6. Programme: start, duration, key milestones.
7. Selection: price only, or scored on price and quality, with weightings if stated.
8. Red flags: unusual indemnity, pay-if-paid or pay-when-paid wording, no payment for delay, phrases such as "all works necessary", conflicts between documents.
9. Addenda: list every addendum and what it changes, and say which earlier answers an addendum overrides.
10. Missing: documents that are referred to but not attached.Here is an excerpt of what comes back for the school refurbishment in tender B (illustrative, with bracketed placeholders for real figures).
2. Dates: bid due [date] 14:00 (Instructions to Bidders 4.2, p.6). Questions by [date] (p.6). Mandatory site visit: NOT FOUND.
3. Scope for drywall and ceilings: Section 09 21 00 partitions (Spec p.212-219); Section 09 51 00 ceilings (p.220-224). Painting and fire stopping by others (Spec p.209).
4. Contract type: lump sum (Form of Tender p.2).
5. Money terms: payment 60 days from valid invoice (Conditions 14.2); retainage 10% until completion (14.5); daily penalty $[amount] for late completion, no cap stated (9.3); bid bond NOT REQUIRED (Instructions 5.1); insurance $[limit] public liability (clause 18).
7. Selection: lowest compliant price (Instructions 7).
8. Red flags: no cap on late-completion penalty (9.3); scope phrase "all works necessary to complete" (Spec p.209).
9. Addenda: Addendum 1 moves the bid date by five days (p.1). Addendum 2 is listed but not attached.
10. Missing: drawings D-101 to D-140 referred to but not attached; Addendum 2.Now a realistic mistake, and how the format catches it. The brief says "bid bond NOT REQUIRED", citing the instructions. Suppose Addendum 2, which the firm has not uploaded, adds a 5% bid bond. If the model had been asked only for a summary, nothing would warn anyone, and the firm could price a bid it was not eligible to submit. The "Missing" line flags the gap, and a quick look at the portal's addenda list would confirm it. The habit that protects you is simple: do not rely on any answer that an unattached or unread addendum could override, and always load every addendum before you ask the questions.
Use the page references. They let you check any figure in a minute, and you should spot-check five of them on every brief, starting with the money terms and the dates. If two of the five are wrong, treat the whole brief as a rough guide and read the document yourself. The tenders and RFP tutorial covers what happens after the decision to bid, and the construction tender writing tutorial covers drafting the response against the scoring.
What AI misses in a tender pack, and the file limits that cause it
Four gaps show up again and again. First, drawings and specifications disagree, and notes on a drawing ("by others", "allow for") may never appear in the text you pasted. Second, scanned pages are read less accurately, so a figure that looks odd deserves a look at the original. Third, a model should not count items from drawings; quantities come from a takeoff, not a summary. Fourth, anything not uploaded is invisible, which is how the addendum slipped through above.
The file limits explain a lot of the misses. OpenAI's file uploads page lists a maximum of 512 MB per file and 2 million tokens for text and document files, and says that when you attach a PDF directly to a chat, ChatGPT can analyse its images as well as its text, including charts and diagrams. Files saved in a Project are handled differently, and embedded images may not be included automatically. Anthropic's help article on uploading files (updated 23 July 2026) lists up to 20 files per chat and PDFs of up to 1,000 pages. It says PDFs of 100 pages or fewer get full analysis of text and visuals, PDFs of 101 to 1,000 pages are processed for text only, and Projects extract text only for most files.
The practical consequences are easy to follow. Attach the pack directly to a chat rather than storing it in a Project. Pull out the drawing sheets and specification sections that name your trade into a single PDF of fewer than 100 pages, so the visuals are read as well as the words. And never assume a 300-page specification book was "seen" in the sense you mean; it was read as text. If you want drawings compared against the specification, that comparison is a job for you, with AI as a second reader on the extracts.
What it costs to price a bid, and the hit rate you need
The screen earns its keep by comparing the cost of an estimate with what winning is worth. Imagine a small firm whose estimator spends 14 hours on a typical bid, at an internal cost of $48 an hour. That bid costs $672 before a single dollar is won. Suppose a won job of this size leaves a $2,400 margin. Divide the cost by the margin and you get the break-even hit rate: $672 divided by $2,400 is 28%. If you win fewer than 28 in every 100 bids of this kind, a little more than one in four, pricing them loses money on average.
| Type of bid | Hit rate from your log | Expected margin per bid | Minus the $672 cost |
|---|---|---|---|
| Repeat client invitation | 1 in 3 | $800 | +$128 |
| Invited tender, four to six bidders | 1 in 5 | $480 | -$192 |
| Open bid with many bidders | 1 in 12 | $200 | -$472 |
Those hit rates are made up for the illustration. Yours come from your log, and AI cannot supply them: it has no idea how often you win. What the table shows is the shape of the problem. Bids with good odds can pay for a lot of estimating, and bids with poor odds can quietly cost more than they return. A 20-minute screen that stops you pricing one open bid in a month saves $672, which pays for a $99 bid-matching plan several times over. Add the admin time and any printing or fees to the bid cost for a truer sum.
Keep a bid log, and retune the weights every quarter
A spreadsheet with one row per bid is the whole system. Record the date, project, client, source, score and band, estimating hours, price, outcome, margin if won, and the reason for a loss if the client will tell you. Some clients will give you the gap to the winning price when you ask politely; that one number is worth more than any scorecard weight.
Start with a back-test. Score your last ten bids as they looked at the time, then compare with the outcomes. If the jobs you won tend to score high and the ones you lost low, your weights fit. If a pattern is missing, change the weights until it appears. After about 20 live bids, review by band. This is what a review might show (illustrative).
| Score band | Bids priced | Won | Hit rate |
|---|---|---|---|
| 75 to 100 | 9 | 4 | 44% |
| 55 to 74 | 8 | 1 | 13% |
| Below 55 | 3 | 0 | 0% |
Against the 28% break-even from earlier, the top band pays and the middle band does not. The sensible response is to raise the clarify band's bar, so that mid-scoring bids are priced only when the clarification lifts the score, and to stop pricing anything below 55. You can hand the log to an AI model for a first pass at this analysis, but check it against a pivot table, because language models can slip on arithmetic.
Here is my bid log as CSV. Calculate the hit rate and the average margin by score band, by client and by source. Flag any band where the hit rate is below [28]%. List the three biggest differences between won and lost bids. Show the counts you used so I can check them.
[paste the CSV]Sibling tutorials cover what happens either side of the screen. The bid invitations tutorial covers triaging invitations and chasing after a submission, and the bid levelling tutorial covers comparing the subcontractor quotes that feed a general contractor's price.
Which tender documents you can paste into a chat
Public tender documents are normally published for anyone to read, so summarising them is rarely a problem. Private bids are different: the pack may carry confidentiality terms, a non-disclosure agreement or client drawings you have no right to share. Check the terms before you upload anything, remove client names and site details if you are unsure, and use a plan where your content is not used to train the model by default. Business plans such as ChatGPT Business and Claude Team work that way according to their vendors, while consumer plans give you a setting to check yourself. Confirm the position on the vendor's page for the plan you use.
How to tell the screen is working
Six numbers tell you within a quarter. Count how many bids reach you, how many fail a knockout, how many you price, your hit rate on priced bids, your estimating hours per win, and the margin you achieve against the margin you estimated. The screen is working when the number you price falls, the hit rate on what you price rises above break-even, and hours per win trend down. If you price fewer bids and win the same number of jobs, the screen has simply given you back estimating time.
Start this week with the back-test. List your last ten bids, score them, and see which ones the screen would have stopped you pricing. The hours that would have saved are the first figure to put in the log, and they are usually all the motivation the estimator needs.
Bid screening questions contractors ask
Can AI find construction bids for me automatically?
Partly. Bid-matching tools search portals and plan rooms and score bids against your trade, but their coverage is not complete, and a tool cannot know your capacity or your client history. Use them to widen the funnel and to set alerts, then run every bid through your own screen before you spend estimating hours.
How many bids should a small contractor price each month?
Enough that estimating time stays inside your capacity and your hit rate is measurable, and no more. Count the estimator hours available, divide by the hours a typical bid takes, and leave slack for urgent jobs. If the screen sends you more than you can price, raise the threshold instead of rushing estimates.
Is bid-matching software worth paying for?
Compare its monthly price with the cost of one wasted estimate. At an illustrative 14 hours and $48 an hour, one bid costs $672 to price, so a $59 or $99 monthly plan pays back if it stops you pricing even one wrong bid a month. Start with a free tier or trial and check that the portals you need are covered.
Further reads
- AI Tools for Construction Estimating: Bid Checks After the Takeoff — Bid checks to run once you decide to price.
- Groundworks Tenders: Build a Tender Pack and Method Statements With AI — Build a tender pack and method statements for civil work.
- Construction Estimating With AI: Turn Supplier Quotes Into a Rate Sheet — Turn supplier quotes into a rate sheet you can price from.
- Construction Retainage: Track It and Chase Release With AI Reminders — Track retainage so a long hold does not surprise you.
- AI Estimator for Contractors or a Human Estimator? How to Choose — A cost sum, a volume-and-complexity grid and a hybrid workflow for choosing between an AI estimating tool and a part-time estimator, with a named sign-off.
- AI Marketing for a Construction Company: Case Studies, Bids and LinkedIn — A commercial builder's workflow: project fact sheet and photos to a case study, a bid-ready sheet and a month of LinkedIn posts, with client consent first.
- Commercial Landscaping Proposal: Bid Property Managers With AI Help — A four-building office park walked from site notes to a tiered, checked bid, with AI drafting the scope and you keeping control of the numbers.
- How to Bid Concrete Jobs: Quantities, Pump Costs and AI Checks — A concrete bid built line by line: volume, waste, pump, labour, finishing and washout, with AI prompts that recheck your arithmetic and catch missing lines.
- How to Bid Drywall Jobs: Plan Review and Scope Checks With AI — A commercial drywall bid from spec to exclusions: wall-type takeoff, fire-rated assemblies, finish levels, alternates, and where AI helps or a person must read.
- How to Bid Flooring Jobs: Commercial Takeoffs and Tenders With AI — A commercial flooring bid from finishes schedule to tender questionnaire: room-by-room takeoff, waste and prep, an AI answer bank and hand-checked counts.
- How to Use AI as a General Contractor: Four Weeks, Four Jobs — A four-week start for a general contractor: documents, site notes, bid comparison and review, with plan costs, owner hours and checks for each job.
- Masonry Bidding: Review Plans and Specs for Scope Gaps With AI — A bid-review routine for masons: AI tables each requirement, flags who supplies and who installs, and turns gaps into RFIs, exclusions and clarifications.
- AI Chatbot for a Construction Company: Quote Requests and Service Areas — Four fact sheets for a construction company's chatbot (areas, project types, quote requests, careers), plus routing, lead capture and no price promises.
- AI for Mechanical Contractors: Can It Read Specs and Drawings? — What AI reads reliably in a mechanical bid set, where it fails on drawings, and a 12-point hand check plus a 45-minute test to run on a past bid.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: PlanHub pricing page; constructionbids.ai home and pricing pages; Autodesk BuildingConnected product page; Anthropic help centre article on uploading files to Claude (updated 23 July 2026); OpenAI help article on uploading files to ChatGPT. Read on 1 November 2026. Worked-example figures are illustrative.