AI for Building Contractors: Chasing Subcontractors and Progress Updates

Keep one chase list that records, for every open item, who owes it, the date you need it and what it blocks. Have AI draft each reminder at the right firmness from your templates, and turn the week's site notes into a five-part client update. You send every message yourself and check every date.

AI is good at being polite and specific. It is poor at knowing your programme, your history with a firm or what the contract says, and it has a habit of smoothing bad news into something cheerful. That habit does little harm in a chase message and a lot in a client update, which is why sign-off matters more than prompting.

Follow me on Instagram@sagnikteaches

One list drives both jobs

The Monday chase list is a sheet with one row per open item. Each row carries who owes it, the day you need it, what it blocks and where you are on the chase ladder. The "blocks" column is the one that matters. It sets the priority and gives every message its reason. Friday's client update reads from the same sheet plus the week's site notes, so the list serves both jobs.

Connect on LinkedInSagnik Bhattacharya

The contractor in this example runs three jobs with eleven subcontractors; the jobs, notes and messages from here on are invented. One is a ten-week fit-out of a small office, now in week four. Its Monday list looks like this:

Subscribe on YouTube@sagnikteaches

Notice that three rows are not subcontractors at all: a designer's answer, a client's decision and a supplier's date. They belong on the list because they block the same work, and because the client rows feed the "decisions we need from you" section of Friday's update. Anything that stops a trade starting goes on the sheet, whoever owes it.

A four-rung ladder, so the tone is decided before the message

Writing a chase is rarely the hard part. Deciding how firm to be is, and that decision is better made once, in advance, than again at seven in the morning on a wet site. One possible ladder follows; the timings are examples, so set your own.

Rung four is deliberately not an AI draft. Notice wording and deadlines come from the contract and from whoever advises you on it, and a notice that misses a clause can do more harm than none. The AI's part at that stage is limited to summarising your chase log into a dated list of attempts. If the subcontract itself is the question, the clauses to check before trusting AI with a subcontractor agreement shows where drafting help stops.

Chase messages for each rung, filled in

Three templates to copy and adapt. The square brackets are the parts the AI, or you, fill from the row on the chase list.

RUNG 1 (text)
Hi [first name], quick one for [job]: we need [item] by [day] so [what it lets start]
can go ahead on [date]. Can you confirm that's on track? Thanks, [your name]

RUNG 2 (email)
Subject: [Job]: [item] now overdue (needed [day])

Hi [first name],
We still need [item] for [job]. It was due [day], and [what it blocks] is waiting on it.
[Date] is the last day we can hold the programme.
Please reply by [time, day] with either the item or a date you will deliver it. If
something is holding you up, tell me what, and I'll see if we can help from our side.
Thanks, [your name]

RUNG 3 (after the call)
Hi [first name], thanks for the call at [time]. As agreed: [item] arrives [date] via
[person]. If that changes, tell me by [time] so I can re-sequence. [your name]

Here is the difference they make. The version most of us send, in a hurry: "Any update on the certs?" The version drafted from the list, for the plasterer's insurance certificate:

Hi [first name], quick one for the office fit-out: we need your renewed insurance certificate by Friday, as site access from Monday depends on it. Can you confirm it's on its way? Thanks, [your name]

The second one can be answered in two words, and it tells the plasterer why Friday matters. Four habits keep drafts like this useful:

• One ask per message. Two asks get one answer.
• A date, never "soon" or "when you can".
• The consequence in programme terms ("glazing waits on this"), never in blame terms.
• No threats before rung four, and none in text messages at all.

The prompt that produces them takes the sheet row and returns the right rung. Keep it in a Project in ChatGPT or Claude, together with a one-page note on how you like to sound, so you paste only the row.

Draft a [rung 1 / rung 2] chase message.
ROW: [paste the sheet row]
CHANNEL: [text / email]   RELATIONSHIP: [long-standing / new / difficult]
RULES: use only the dates in the row. One ask. State what it blocks in programme terms.
No blame, no threats. Plain words. Under 50 words for a text, under 120 for an email.
Where a date is missing, write [DATE?].

Letting the AI sort Monday's list from your lookahead

Once the sheet exists, the AI can do the Monday sort. Paste the three-week lookahead, the open items and today's date, and ask for each item's days until it blocks work, a suggested rung and a one-line reason, sorted soonest first. Ask it to flag any row with no needed-by date, since those are the ones that quietly drift.

A realistic first result carries three faults, and each shows up in a different way:

• The ranking is wrong for a rule it does not know. It put the plasterer's insurance certificate sixth, because no plastering is scheduled for a week. Your rule is "no certificate, no access", which makes it urgent. Add the rule to the Project's standing instructions and it will not make the mistake again.
• It counts calendar days. A Monday item shows "3 days" on a Friday afternoon, when it is one working day away. Tell it to count working days, Monday to Friday, and to say so in the output.
• It marked the partition-frame delivery as overdue. The sheet still held the original date. The supplier had emailed a new one on Friday evening and nobody updated the row. The fault was in the sheet: garbage in the row, confident ranking out.

The sort saves ten minutes. The three corrections are why you still read the top five rows yourself. If you want the same idea applied to subcontractor paperwork, the subcontractor paperwork chasing checklist works through the documents a contractor needs before anyone starts, and AI scheduling for small builders covers the programme the lookahead comes from.

One message per firm when a sub is on two of your jobs

With three jobs running, the same firm often owes you things on more than one site. Three separate chasers to one electrician in a morning read as nagging, and the replies come back muddled: "yes" to which one? Before drafting, ask the AI to group Monday's rows by firm and write one numbered message per firm, most urgent item first:

Group these chase-list rows by firm. For any firm with two or more open items, write ONE
message: a numbered list, most urgent first, each line with the job, the item, the date
and what it blocks. Use only the dates in the rows. Under 80 words for a text.
ROWS: [paste]

The electrician also owes a test certificate on a kitchen extension, so the draft comes back as:

Hi [first name], two things this week, most urgent first. 1. Office fit-out: first-fix sign-off by Tuesday, as the ceilings and drylining wait on it. 2. Kitchen extension: test certificate by Thursday, needed for the client handover on Friday. Can you reply "1 yes, 2 yes", or tell me which one is a problem? Thanks, [your name]

The one-ask rule bends here, and the numbering is what keeps it workable: a reply as short as "1 yes, 2 Friday" is still unambiguous. Check one thing every time. In grouped drafts a model can attach one job's date to the other job's item, so read each numbered line against its own row before you press send.

When the reply is "we're stretched"

Chases do not always end in a yes. A frequent answer to a rung-two email is the polite no. The electrician writes back: "Stretched this week, can't sign off first fix until Monday." The goal has now changed from getting the item to protecting the programme, so the reply needs a different kind of draft.

Paste the reply, the row and the lookahead into the Project and ask for options, not a message: "List what a Monday sign-off changes on the lookahead, and give me two ways to protect the ceiling start." A useful answer says that grid on level 2 waits three working days, and offers two routes: a partial sign-off of the level 2 zone on Thursday, or moving the ceiling crew to level 1 in the meantime. The model has reasoned only from the lookahead you pasted. You know whether level 1 is ready, and whether the electrician charges for a part visit. Make that judgement yourself, then have the model draft the acknowledgement:

Hi [first name], thanks for letting me know. To keep the ceiling crew moving, can you sign off the level 2 zone only on Thursday, and the rest on Monday as you said? If that isn't possible, please confirm Monday in writing by 4pm today so I can re-sequence the crew. Thanks, [your name]

Tone follows the relationship. With a long-standing subcontractor that reads as a favour asked between people who work together often. For a new or difficult one, send the same message by email, copy the office, and add the cost of re-sequencing if your subcontract provides for it. The model does not know your subcontract wording, so that sentence is yours to write or to take from whoever drafted the terms.

Capturing the notes in the first place

The update is only as honest as the notes. A site manager who writes nothing until Friday afternoon rebuilds the week from memory, and memory is generous. Replace that with a 90-second voice note at the end of each day, transcribed by the phone's dictation or an app and appended to a running notes document. A fixed script keeps the notes consistent enough for the model to use:

Job, date.
Done today: ...
Not done, and why: ...
Who did not turn up: ...
Questions I asked, and who owes me an answer: ...
Client asked for: ...
Photos taken: ...

Five of those notes make Friday's input. The "who owes me an answer" line feeds the chase list directly, so Monday's list partly writes itself. For the dictation side, turning on-site voice notes into job sheets covers the mechanics.

Friday's client update from the week's site notes

A client usually needs five things: where things stand, what happened, what comes next, what they must decide and what could move the finish date. A fixed five-part layout answers all of them and makes the updates comparable week to week. The messy input, as the site manager jotted it:

Mon: ceiling grid started, level 2 only, 6 men. Tue: electrician first fix done except kitchen, waiting on drawings for pantry sockets (asked Mon, no reply). Wed: plasterer didn't turn up, said sick, back Mon maybe. Thu: client asked about moving meeting room door, told her I'd price it. Glazing frames due Fri, supplier now says Tues next week. Skip filled, 2 loads. Fri: snag walk with ceiling guys, 14 items. Photos in the folder.

Fed through a plain "write the weekly update" prompt, a model can easily return something like this first draft:

Hi [first name], a good week on site. Ceiling grid is going up on the second floor and electrical first fix is largely complete. There was a minor delay to the partition frames, which we don't expect to affect the programme. We also had a small issue with one trade, which is being resolved. Next week we will continue with ceilings and begin partitions. Let us know if you have any questions.

Read it again against the notes. "A good week" says nothing. "Largely complete" hides that the kitchen sockets are waiting on a drawing. "Minor delay" is a two-working-day slip with a stated effect, and the draft has dropped the number. "One trade" is a plasterer who did not attend for three days. The door request, which needs a price and a decision, is missing entirely. After editing:

Hi [first name], where we are: ceilings have started on level 2, and first-fix electrics are finished except the pantry sockets, which are waiting on the designer's drawing (requested Monday). Done this week: grid started on level 2, skip cleared twice, snag walk on the ceilings with 14 items logged. Plasterer was off site Wednesday to Friday; expected back Monday, and I'll confirm that first thing. Glazing frames now arrive Tuesday instead of Friday. Glazing starts in week 6, so the finish date is unchanged for now. Next week: ceilings continue, partitions begin. Decisions needed: the meeting room door. I'll send a price by Tuesday and need your answer by Friday to keep the partition layout on programme. Ceiling tile colour is also needed by Wednesday because of the three-week lead time. Photos are in the shared folder.

Every sentence in that version can be checked against the notes or the list. Nothing in it is cheerful on its own account.

Five phrases AI uses to soften news

Watch for these in any draft, and put them in the Project's standing instructions as things never to write without support.

Add two standing rules beside them: never write "on track" unless the programme dates support it, and keep what happened apart from what you expect to happen. Those two sentences in a Project's instructions do more for honesty than any amount of rewording on the day.

The sign-off: two minutes before anything goes

The person who knows the site reads the draft and the same person sends it. Ten questions, asked in order:

1. Is every date and weekday in the message in my notes or on the programme?
2. Does the status line match what I would say standing on site?
3. Is every delay stated with its length and its effect?
4. Does any decision I am asking for have a deadline and a reason?
5. Have I left out subcontractors' prices, rates and internal disagreements?
6. Does anything read as blame, or as a promise I may not keep?
7. Are the photos free of faces, neighbours' windows and security details?
8. Does it agree with what I told the client last week?
9. Are names, addresses and the client's title spelled correctly?
10. Would I be comfortable if the subcontractor or the client's other adviser read it?

Question ten catches more than it seems to. Updates get forwarded, and a sentence written about the plasterer's absence will, sooner or later, reach the plasterer.

Where your job software already does part of this

If you run jobs in a project-management app, check what it drafts before you build a parallel routine. JobTread, for example, lists "AI Daily Log Summaries", described as combining multiple daily logs into one summary of job progress that you can share with your team, customers and vendors, and an "AI Writing Assistant" for drafting and rewriting project communications. Its pricing page shows $199 a month with one internal user included, $20 a month for each extra internal user up to ten users (less per user above that) and a 20% saving on annual billing. Vendors, subcontractors and customers join its portals at no charge, and there is no free trial, only a 30-day money-back guarantee on monthly plans. The page does not say whether AI features cost extra beyond "includes every feature", so ask before assuming.

Whatever the software drafts, the sign-off applies. A built-in summary is the same smoothing machine with better access to your logs. For the update itself, writing client updates and snag lists with AI goes deeper on tone and layout, and pricing variations and change orders covers what to do with a request like the meeting room door.

A chase log you can rely on later

Log every chase in the sheet as you go: date, channel, who, what you asked and what came back. The log earns its keep on the day a subcontractor says nobody told them, or a client asks why partitions slipped. A dated sequence of "asked Tuesday by text, Wednesday by email, Thursday by phone, agreed Friday" settles most arguments before they start.

Texts and chats live on personal phones, so copy key threads into the job folder weekly. If a chase history ever has to support a formal notice or a claim, take advice from someone who knows your contract, because the wording, the deadlines and the method of service are set by the contract and not by anything in this tutorial. The AI can then help by turning the log into a plain chronology, which saves your adviser time.

Four weeks in: three numbers that show the loop works

Keep a small tally on the same sheet, filled in on Fridays in two minutes. Three numbers tell you whether the routine is doing its job:

• Items that held up work. Count the rows where the "blocks" column came true because the item arrived after the trade needed it. This is the number the whole routine exists to push down.
• Working days from first reminder to a committed date. Not to delivery, which depends on the subcontractor's workload, but to a date you can plan around.
• Client questions after each update. Every "so what does that mean for the finish date?" reply is a sentence the update should have contained.

A sample tally shows how to read them. Week one: three items held up work, a committed date took 2.5 working days on average, and the client answered the update with two questions. Week four: one item held up work, 1.5 days, no questions. The item that still slipped is the useful one. Open its row and you usually find a needed-by date entered too late, or a blank "blocks" cell, so the chaser went out without a reason anyone could act on. If nothing moves after a month, look at the sheet before the prompts: a list updated only on Mondays cannot catch a delivery date that changed on Tuesday.

What it saves in a week, and what it does not

A quick sum, with every minute an assumption to replace with your own week. Say the site manager spends five hours a week writing and sending chasers and four hours writing client updates across the three jobs: nine hours. With the routine above:

That saves about 285 minutes, nearly five hours, a week. The calls do not shrink, nor should they. The log row is new work that did not exist before, and it is worth keeping. An assistant at roughly $20 a month per person, or a Team or Business seat at $20 billed annually ($25 monthly) with a two-seat minimum on Claude or ChatGPT, costs less per month than an hour of a site manager's time.

If you would rather set the routine up with someone who has seen it fail in different places, that is what my AI implementation consultation covers.

Questions from contractors before they try it

Should the AI send the chasers automatically?

I would not. Every chase is a small relationship decision, and automation cannot see that the subcontractor rang you this morning. A draft takes about twenty seconds to read and send yourself. If you automate anything later, limit it to the first rung, a fixed template and dates you typed, with a person deciding when it fires.

How do I stop it inventing dates?

Give it the dates as a labelled list and tell it to use only those. Add a rule that any date missing from the input must appear as [DATE?] in the draft. Then search the draft for every date and weekday before sending. Models fill gaps fluently, so the habit of checking dates matters more than the wording of the rule.

Can I use the same drafts on WhatsApp?

Yes. The channel does not change the method, only the length: the first-rung text should fit on one phone screen. Keep a record by saving key threads, or screenshots of them, to the job folder, because chat history on a personal phone is a poor place to keep evidence of who was asked for what and when.

Will subcontractors mind receiving AI-drafted messages?

They will mind vague or aggressive ones, whoever wrote them. A message that names the item, the date and what it holds up is easier to answer than a chatty reminder, so most people welcome it. Put your own voice into the template once, and keep the sign-off, so every message still reads as coming from you.

Further reads

• AI Tools for Home Builders: Build Schedules, Delays and Sub Chasing — Selections, delays and sub chasing for home builders
• Five AI Apps for Construction Site Work: Photos, Logs, Checks — Apps for site photos, logs and checks
• AI Construction Contract Review Software: What Builders Should Test — Test AI contract review on an old subcontract with an answer key and planted changes, then read payment, retention, damages and indemnity yourself.
• AI Receptionist for Construction Companies: Real Costs and Call Routing — Six plans priced at 250 calls, what answering only overflow saves, and a routing rule per caller type, with urgent site issues going to a person.
• CIS Monthly Return: Software, Accountant or an AI-Built Sheet? — What you need to file a CIS monthly return, the April 2026 nil-return change, and how software, an accountant and an AI-built sheet compare.
• Home Builders AI: Tools and Cost for Weekly Buyer Progress Updates — The tools and monthly cost behind weekly AI-written buyer updates for a small home builder, from a $20 assistant to a full portal, with the routine and checks.
• Which AI Tools Should a General Contractor Buy? What to Skip — A buy-first order, a skip list with triggers, and a hours-it-must-save test for a general contractor, priced for a six-person remodelling firm.
• AI Software for Construction Scheduling: Programme Updates After Delays — One small job walked through a late delivery and two weather days: what the assistant does, what you calculate yourself, and what you tell each sub.
• AI for Electrical Project Management: Diaries, RFIs, Snag Lists — Voice note to site diary, an RFI log that gets chased, and snag lists others can close: a working setup for a small electrical contractor, with prompts.
• Best AI for Builders: Pick the First Job by Hours Lost and Risk — A scoring sheet for small building firms: rank missed calls, quote follow-up, payment chasing and site notes by hours lost and risk, then start with the winner.
• CIS Deduction Statement Template: Fill It From Your Payment Sheet With AI — The fields, the 14-day deadline and a payment-sheet method for a CIS deduction statement per subcontractor each month, with AI doing the checking.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: JobTread pricing and AI features pages (October 2026); Claude and ChatGPT pricing pages. The jobs, programme, notes, tallies and messages are invented for illustration.