A small home builder keeps a schedule on track with three things: a system where subs can see their tasks (JobTread's portal, for example), an assistant such as ChatGPT or Claude to draft the look-ahead and the chasing messages from a schedule export, and a delay log. The AI drafts the nudge; you still make the call.
Search results for this topic mostly show home-design generators, which draw floor plans and exteriors before a build starts. Once the build is under way, the trouble is a sub who did not turn up, a selection that arrived late, or a client who hears about a delay from a neighbour. Those problems need a weekly routine more than a clever tool.
The weekly loop that keeps a build on its dates
Schedules slip in small, boring ways. A start nobody confirmed. A delivery never ordered because the client had not chosen the product. A delay the builder knew about on Tuesday and mentioned to the client three weeks later. The fix is a loop that runs the same way every week, with the assistant drafting and the builder deciding and phoning.
The example here is a two-storey house with a 30-week build, 14 subcontractors and a dozen selections, run by a small firm that has one or two builds going at a time. The figures are illustrations.
| Day | Job | What the assistant drafts | What you do |
|---|---|---|---|
| Monday | Three-week look-ahead | Turns the schedule export into a table by trade and flags starts with no confirmation | Checks it against the master schedule and marks critical starts |
| Tuesday | Confirmations | Texts for each unconfirmed start | Sends them and logs every reply the same day |
| Wednesday | Deliveries and selections | Lists items due on site in two weeks with order-by dates | Phones suppliers to confirm lead times |
| Thursday | Chase and call | Second nudges for silent subs | Phones anyone still silent |
| Friday | Delay log and client update | Drafts the update from the log and look-ahead | Types the figures, approves and sends |
Each step leaves something you can see: a confirmed-by column, a delay-log line, a decision calendar, a Friday message. Keep the master schedule in one place and paste a fresh three-week copy into the assistant each Monday. An assistant has no idea what changed since last week, and a stale paste is how it ends up chasing the wrong person. For the scheduling basics behind this, AI scheduling for small builders covers subcontractor and delivery booking in more depth.
Sub chasing as a ladder, with the wording ready
Subs respond to a clear ask on a predictable rhythm. Five rungs cover most starts: a heads-up 14 days before, a confirmation request 7 days before, a phone call if there is still silence 3 days before, a final message with access details 2 working days before, and a call 30 minutes after the start time if nobody has arrived. The assistant can write the first, second and fourth rungs. The calls are yours, because a voice gets answers a text does not.
Give the assistant the table, not a description of it:
I am a home builder. Below is next week's schedule as a table: task, sub, start date, access notes, confirmed by, next trade.
Write one text message for every row where "confirmed by" is blank.
Rules: under 320 characters. First line gives the job (Lot 14) and the start date. State exactly what I need back ("Reply YES and your crew size"). Include the access note. No apology, no threat, no mention of money. Use [first name] for the name. If the row is marked CRITICAL, say which trade follows.
Table:
Rough-in plumbing | plumbing sub | Tue 16 March, 7:30 | key box on garage side door | blank | insulation, Mon 22 March (CRITICAL)
The first draft is usually close and slightly wrong in ways that matter (illustrative):
| Draft from the assistant | What I would send |
|---|---|
| Hi [first name], plumbing rough-in at Lot 14 starts Tuesday 16 March. Reply YES to confirm by 5 pm. This holds up insulation on Monday 22 March. | Hi [first name], plumbing rough-in at Lot 14 starts Tuesday 16 March at 7:30. Key box is on the garage side door. Please reply YES and your crew size by Thursday. Insulation is booked behind you on Monday 22 March. |
Four changes made the difference. The draft dropped the start time and the access note, both of which were in the table. "By 5 pm" named no day. "This holds up insulation" reads as blame to someone who has done nothing wrong yet. And it did not ask for a crew size, which is the number that tells you whether the job really will finish in three days. Fix the prompt once and the next batch inherits the fixes: add "always include start time and access note" and "name a deadline day".
The calls follow a short script, and it helps to write it down so each call takes two minutes. Ask whether they are on the way, who is coming and when work will start. If they cannot start today, ask for the first day they can and whether the crew they send can finish in the days left. Log the answer in the schedule before you put the phone down. The assistant can turn your call notes into a log line and a message to the next trade in the queue, which is a good use of it. It should not decide whether to threaten a back-charge: that depends on your subcontract, and where the clause is contested it is a question for your lawyer. Keep money out of chasing messages altogether.
A realistic slip: chasing someone who already confirmed
The commonest chasing error is not tone. It is a message to a sub who confirmed by phone on Monday while the "confirmed by" column was still blank. The sub replies, a little sharply, that you spoke yesterday. The cause is simple: the sheet was not updated the same day. Make "log every reply, including phone calls, before you finish" a rule, and tell the assistant to skip any row marked "phone". For the longer treatment of chasing and progress messages, AI for building contractors: chasing subcontractors and progress updates goes through more message types.
Three silent subs and one phone: who gets the call first
By Wednesday some starts will still be unconfirmed, and a builder with one phone needs an order for the calls. A rule you can apply in a minute: call first the start with no float, then the one with the most trades queued behind it, then the one that would wait longest for a new slot if it slipped. As a quick score, give each start its float shortfall (5 minus its float in days, never below zero), plus the number of trades waiting behind it, plus the weeks a rebooking would take. Here are three silent starts for next week:
| Sub | Task and start | Float (days) | Trades waiting | Rebook wait (weeks) | Score |
|---|---|---|---|---|---|
| Plasterer | Boarding finish, Monday | 0 | 2 | 3 | 10 |
| Electrician | Second fix, Wednesday | 3 | 1 | 1 | 4 |
| Landscaper | Driveway base, Thursday | 5 | 0 | 2 | 2 |
The plasterer gets the first call (5 plus 2 plus 3 is 10), the electrician the second (2 plus 1 plus 1 is 4) and the landscaper a text (0 plus 0 plus 2 is 2). You can hand the assistant the look-ahead table and ask it to score and rank the silent rows, but give it the formula and ask it to show its working. Otherwise it may rank by the order of the rows or by how urgent your notes sound. Then check the sums yourself. The weights are a rule of thumb, so change them if a trade you use is booked months ahead and a slip costs more than a score of three suggests.
A delay log that separates weather, materials, client and subs
A delay only moves your finish date if it sits on the critical path, the chain of tasks with no spare time. Float is the number of working days a task can slip before it pushes the finish. A roofing sub who starts three days late costs nothing if the next trade has three days of float, and costs three days if it has none. Without a log, nobody can say which delays mattered, and clients are told "everything is delayed" or "nothing is".
The log needs seven columns: date, cause (weather, materials, client decision, subcontractor, inspection or design), task affected, working days lost on that task, float available, net days lost on the finish date, and who was told when. Here are three entries from the example build:
| Working day | Cause | Days lost on the task | Float available | Net days lost on finish |
|---|---|---|---|---|
| 38 | Failed framing inspection; re-inspection two days later | 2 | 0 | 2 |
| 52 | Roofing sub late start, double-booked | 3 | 2 | 1 |
| 71 | Rain on exterior work | 2 | 0 | 2 |
| Total | 5 |
A 30-week build is 150 working days, so five net days moves the finish from day 150 to day 155, one calendar week. The roofing entry is the useful one: three days lost, two absorbed by float, one real. Without the float column the log would claim three, the client would hear three, and the number would be wrong in the other direction.
Rain is the delay builders argue about most, so decide the rule once: which tasks are weather-sensitive, how much rain or wind stops each, and who records the decision. Add a "weather-sensitive" mark to the schedule table so the assistant can flag those starts for a forecast check on Wednesday. The log entry should name the task, the hours lost and the rule applied. "Rain" on its own is not evidence, and it will not survive a client's question in week 20.
Once the log is up to date, the assistant can help re-plan. Give it the next three weeks and the log:
Here is the next three weeks of the schedule (task, trade, start, duration in working days, depends on, float in working days) and this week's delay-log entries.
Suggest up to three ways to protect the finish date. For each give: what moves, what it depends on, which trade must agree, and what could go wrong.
Use only tasks in the table. Do not invent trades or dates. Mark any suggestion that needs the client's agreement.
A typical answer offers three ideas: start interior insulation in the ground-floor rooms while exterior cladding waits for dry weather, swap the order of two interior trades, and add a second crew for one task. Read these as a colleague's brainstorm. The second-crew idea ignores that two trades cannot work in the same small room, and it assumes a supplier can deliver earlier, which only the supplier knows. The first idea is usually the real saving, because it uses work that does not depend on the weather. If you want programme-update wording for the client, AI construction scheduling: programme updates after delays goes through that side.
Selections: back-schedule from lead times to a decision date
Selections are the client's choices: windows, cabinets, tile, lighting. Their effect on a schedule is mechanical. Each item has a date it must be on site, a supplier lead time, some buffer and a sign-off step, and the client's decision has to land before all of those. Work backwards, and a vague "choose finishes eventually" becomes a dated list.
| Item | Needed on site (week) | Lead time | Buffer | Sign-off | Client decides by (week) |
|---|---|---|---|---|---|
| Windows and exterior doors | 11 | 10 | 2 | 1 | -2 (before the build starts) |
| Kitchen cabinets | 18 | 8 | 2 | 2 | 6 |
| Plumbing fixtures (rough-in specs) | 12 | 3 | 1 | 1 | 7 |
| Lighting package | 21 | 6 | 1 | 1 | 13 |
| Interior tile | 21 | 3 | 1 | 1 | 16 |
| Hardwood flooring | 24 | 4 | 2 | 1 | 17 |
The lead times are illustrative; use your suppliers' current quotes. The arithmetic is needed week minus (lead time plus buffer plus sign-off): for windows, 11 minus 13 is minus 2, meaning the decision falls two weeks before the first day on site. That is why windows belong in the contract conversation and not in week 9. Ask the assistant to produce this table from your schedule and the supplier quotes, then check each row yourself, because an assistant cannot know that a supplier's lead time changed last month. Put the decision weeks into the client's calendar at contract signing, and the contract should say what a late decision does to the finish date. The selection sheet with allowances and deadlines shows how to lay that sheet out.
Catching a wrong order before it ships
Selections go wrong twice: when the client decides late, and when the order does not match the decision. The second is where an assistant earns its keep, because comparing two lists line by line is exactly what it does well. Paste the selection sheet and the supplier's order confirmation, and ask for every difference in model, finish, quantity, size and delivery week. A sample result:
| Item | Selection sheet | Order confirmation | Difference |
|---|---|---|---|
| Kitchen sink | Undermount, matt black | Undermount, brushed steel | Finish |
| Tap | Single lever, matt black | Single lever, matt black | None |
| Pendant lights | 2 | 3 | Quantity |
| Cabinet delivery | Week 16 | Week 18 | Two weeks later than planned, which uses up the whole buffer |
A wrong finish found on a confirmation costs a phone call. The same error found on installation day can cost a reorder with a six-week lead time. Check the assistant's table against the two documents, since it can misread a code or treat a near match as identical, and tell it to list near matches instead of ignoring them.
Allowance emails that state the gap without blame
An allowance is a budgeted sum for something the client has not chosen yet. When the choice exceeds it, the email has to give the gap, the options and the date. Suppose the lighting allowance is $6,500 and the client's picks come to $8,120. The gap is $1,620. If your contract adds a builder's fee to overages (check the clause rather than assuming), a 10% fee makes the client's extra $1,782. The draft to the client might read:
Hi [first name], the lighting you have chosen comes to $8,120 against the $6,500 allowance, a difference of $1,620 (the contract's fee on overages would bring it to $1,782). There are three ways to go: keep the package as chosen, swap two fittings to bring it nearer the allowance, or move the difference to another allowance that is running under. We need your decision by Friday of week 13 so the lights can be ordered with their six-week lead time. I have attached the list with prices.
Check three things before sending. The draft must not say "you have gone over budget", which reads as an accusation when the allowance was only ever an estimate. The fee must appear only if the contract provides for it, which is a clause question and not something an assistant knows. And the date must be tied to a reason, because "by Friday" with no consequence gets ignored.
Friday's client update, built from the log
The weekly update is the delay log, the look-ahead and the decision calendar in a client's words. Give the assistant the week's log lines, the next week's look-ahead and a block of figures you have typed yourself:
Figures (typed by me; use exactly as written):
Week: 12 of 30
Net working days lost to date: 3 (failed framing inspection 2, roofing sub late start 1)
Forecast completion: Wednesday of week 31 (was Friday of week 30)
Decisions outstanding: lighting, due Friday of week 13; tile, due week 16
Allowance position: lighting $8,120 chosen against $6,500
The three days follow from the log: two net days at day 38 and one at day 52. Day 150 plus 3 is day 153, the Wednesday of week 31. The assistant turns it into plain language. The same limits apply as for any client report: it may not estimate figures, it may not call a delay "minor", and it may not blame a sub by name. A first draft shows why those limits need writing into the prompt:
Hi [first name], good progress this week: the roof is finished, the windows are in, and plumbing and electrical rough-in start on Monday. We have had a small delay of three days, mostly because the roofer let us down, so completion is now the end of week 31. Lighting needs deciding by Friday of week 13 and tile by week 16. Your lighting selections are $1,620 over budget.
It reads well and breaks four rules. "A small delay" is a judgement the figures do not support, so say "three working days". "Mostly because the roofer let us down" blames a sub and gets the cause wrong: two of the three days came from the failed framing inspection, and the late roofing start cost one. "The end of week 31" drifts two days past the forecast you typed, and next week's update will either repeat the error or appear to win back two days nobody won. And "$1,620 over budget" is the accusing wording the allowance email avoids, without the fee question; replace it with "lighting: $8,120 chosen against a $6,500 allowance, options to follow separately". The corrected delay sentence reads: "We have lost three working days so far, two waiting for the framing re-inspection and one from a late roofing start, so we now expect to finish on the Wednesday of week 31 rather than the Friday of week 30."
For a full weekly-update routine with photos and costs, see home builders' weekly buyer progress updates.
Which tools do which job, and what they cost
This is what the vendors' own pages showed on 29 October 2026, in USD. Notice how little of the routine needs anything labelled AI: the dates have to live somewhere, and the drafting can come from an assistant you already pay for.
| Option | What it covers in this routine | List price | Catch |
|---|---|---|---|
| JobTread | Tasks and scheduling with due dates; sub and vendor portals where subs handle bids, purchase orders, tasks, photos and bills; selections and allowances; customer portals. Its AI tools include daily log summaries, a writing assistant, a report builder and an autofill for its web clipper that pulls product details from supplier sites into selections | $199 a month with one internal user, then $20 per extra internal user ($16 on annual billing, which saves 20%). Vendor, sub and customer portal users are free and unlimited | No free trial (30-day money-back on monthly plans). Its AI page lists no AI feature for scheduling itself. It says every feature is included but does not name the AI tools in that promise, so confirm. |
| Buildertrend | Named here because builders ask about it | Not stated: its pages returned an error when I tried to load them this month | I am making no claim about its features, prices or AI. Ask for a demo and a written list of which features use AI. |
| An assistant plus a spreadsheet | Look-ahead tables, chasing messages, delay-log analysis, client updates | Claude Pro $17 a month billed annually ($20 monthly); ChatGPT Plus about $20 a month | You paste fresh data weekly, and subs have no portal. |
| Home-design and floor-plan generators | Nothing in this routine | Varies | They draw before the build; they do not track a schedule. |
If you run one or two builds, the spreadsheet route costs about $20 a month and does most of the work. Move to a platform when subs and clients both need their own view of the dates, because a shared record beats a faster draft. The comparison of software for custom home builders goes further on selections and client portals.
Signs the routine is working after a month
Count a few things each Friday for four weeks, against pass marks you can adjust:
| Measure | How to count | Pass mark |
|---|---|---|
| Starts confirmed 7 days ahead | Confirmed starts divided by starts due that week | 8 in 10 or better |
| Subs starting within one working day of plan | On-time starts divided by starts | 9 in 10 or better |
| Client decisions made by their date | Decisions on time divided by decisions due | All but one |
| Delays logged within a day | Log entries dated within one working day of the event | Every one |
| Client surprises | Times the client learned of a delay from someone else | Zero |
If confirmations fall short, the ladder wording needs work. If decisions are late, move the dates earlier at contract signing. If delays are logged late, the habit is the problem, not the tool. Adjust one thing a week and keep counting, because the routine only works when the numbers are honest.
Questions builders ask about AI and schedules
Can an assistant build my construction schedule for me?
It can draft a sequence of tasks from your list and lead times, but it does not know your subs, your site or your suppliers. Treat its output as a first pass to correct. The schedule you work from should live in one system or sheet you control, and each week you paste a fresh copy into the assistant.
Should AI text my subcontractors directly?
Not without a person reading each message. A draft that goes out unread can chase someone who already confirmed by phone, or sound like blame. Let the assistant write the messages, send them yourself, and record every reply the same day so the next batch starts from true information.
How do I work out when a client must choose finishes?
Start from the week the item is needed on site and subtract the supplier's lead time, a buffer and time for sign-off. The result is the decision date. Windows and cabinets often land before the build starts or in its first weeks, so tell clients at contract stage, not when the item is already late.
What if my schedule software has no AI?
Use it for what it does well, which is holding the dates, and bring the assistant in for drafting. Export or copy the next three weeks into a table, paste it with a short prompt, and ask for confirmations, chasers or a client update. The routine matters more than whether the platform advertises AI.
Further reads
- AI Tool for Construction Planning: Pre-Start Checklist and Long-Lead Items — A pre-start checklist that catches long-lead items early.
- 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.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: JobTread pricing, features and AI pages; Anthropic Claude pricing page; OpenAI plan details from vendor pages (read 29 October 2026). Buildertrend's pages did not load, so nothing is stated about its features, prices or AI.