Pair one tool type with each job. For quotes, use a takeoff or quoting tool plus a chat assistant for scope wording. For programmes, use the scheduling in your job software, or a spreadsheet with an assistant drafting the sequence. For site paperwork, use a photo and daily-log app with voice notes. AI drafts; you price, sign and certify.
A five-person building company is usually better served by one platform and a general chat assistant than by three specialist subscriptions, and that is the shape I would aim for. Every price below is a USD list price read from the vendor's own page in October 2026, so confirm it before buying. The job is invented for illustration, and I rank tool types rather than brands. If you have not yet chosen which job to automate first, score them by hours lost and risk before you buy anything.
The job: a rear extension, priced, programmed and documented
Every example below comes from one imagined job. A five-person building company (an owner who estimates, a site manager, two carpenters and a part-time administrator) wins a single-storey rear extension of 28 square metres, about 300 square feet: a steel beam to open the back wall, bifold doors, a roof window, new electrics and heating runs, and a kitchen-diner. The contract price is $92,000 and the build is planned for 14 weeks. Groundworks, electrics, plumbing and heating, and plastering are subcontracted.
That one job produces three kinds of work: a quote to write before the contract, a programme (the schedule, if you prefer that word) to build before the first delivery, and paperwork to keep for 14 weeks. They need different tools because they fail in different ways. A quote fails when a quantity is wrong, a programme when a lead time is missed, and paperwork when it is not written at the time.
Quotes: four tool types and what each did on the extension
| Tool type | What it does | Examples and list prices | Where it let the extension down | Pick it when |
|---|---|---|---|---|
| AI takeoff tool | Reads uploaded drawings, then measures and counts items | Kreo Pro $175 per user a month on annual billing (one to four users; Essential at $100 has no AI). BuildVision Pro $299 a month with a token allowance | Counted two legend symbols as downlights; measured the roof in plan, not along the slope | You price several jobs a month from drawings |
| Quoting module in job software | Turns cost lines into a client quote and tracks it | JobTread: $199 a month for one internal user plus $20 for each extra user up to ten, every feature in one plan. Buildertrend sells by quote | Its AI names and scales plan sheets and counts repeated items, but the vendor claims no AI measuring of lengths or areas; budget drafting still needs your cost data | You want quotes, programme and logs in one place |
| General chat assistant | Writes scope, exclusions, allowances and cover emails | ChatGPT or Claude at about $20 a month; Gemini or Copilot if bundled in your office suite | Rewrote allowances as "included" work | Always; it is the cheapest layer |
| Spreadsheet plus assistant | Holds your rates and checks quantities | Excel with the Copilot side pane, or Google Sheets with Gemini | Chat arithmetic drifted; formulas did not | You already keep a cost database |
Takeoff: ten minutes of hand-counting before you trust it
A takeoff tool gives you quantities, not prices. The two pricing models differ in a way that matters for a small firm. Kreo charges per user and bills annually. BuildVision charges per month with a token meter: the vendor says a processed page costs 500 tokens and each detection 20, and gives a 50-page set with 1,000 detections as about 45,000 tokens. Its 14-day trial includes 50,000 tokens and takes a card up front, so one realistic plan set uses nearly all of it. The page does not show the price of the larger monthly allowance beside the option, so ask before assuming $299 covers a busy month.
A quick sum before paying: $175 a month, with the owner's time valued at $40 an hour, is 4.4 hours of owner time. If the tool saves 1.5 hours a quote, it breaks even at about three quotes a month. Those are assumptions; put in your own.
Whatever tool you use, count a sample by hand. This is what the check looked like on the extension:
| Item | Tool result | Hand count | Verdict |
|---|---|---|---|
| Floor slab | 28.0 m² | 28.0 m² | Matches |
| Foundation run | 22.4 m | 22.0 m | Close enough; rounding |
| Downlights | 16 | 14 | Two legend symbols counted as fittings |
| Double sockets | 9 | 9 | Matches |
| Roof area | 34 m² | about 36 m² on the slope | Tool measured the plan view, not the pitched surface |
Two of five were wrong, and both errors would have flowed straight into the price. Before accepting any total, also check the scale by measuring one dimension you know, such as a door opening, because a 1:50 drawing read as 1:100 doubles every length and quadruples every area.
Why not upload the drawings to a chat assistant instead?
You can, and it is useful for the parts of a drawing set that are words: the notes, the specification, the schedule of openings. Ask it to list everything the notes say the contractor must supply, then compare that list with your cost lines. It is a poor way to get quantities. General assistants read drawings as pictures, and their limits on visual reading are tighter than their limits on text (Claude, for instance, analyses a PDF's page layout visually only up to 100 pages). They also have no scale control, which is the feature a takeoff tool is built around. Use the assistant for the notes and the tool for the measuring.
Scope wording: allowances must stay allowances
A chat assistant is best at the words around the numbers. This prompt kept the figures out of its hands:
Below is the cost breakdown for a single-storey rear extension. Lines marked ALLOW are allowances, not fixed prices.
Write (1) a scope of works in plain English, 250 words at most, (2) a list of exclusions, (3) a table of allowances with the amount for each.
Do not change, round or add any figure. If a line is unclear, ask me instead of guessing.
[paste the cost lines here]The draft was good apart from two things I would fix every time. It described the $4,500 electrics allowance as "a full electrical installation", which turns an allowance into a promise. And it added "all necessary approvals are included", which was in nobody's cost line. Both come out. If you work from a saved project holding your standard exclusions, the second error stops appearing, because the assistant copies your wording instead of inventing its own.
Programmes: let the assistant find the order-by dates
A programme is only as good as its lead times, and the assistant's most useful job is turning them into dates. Give it the task list with durations, the week each material is needed on site, and each supplier's lead time. For the extension:
| Task | Weeks | Material needed on site | Lead time |
|---|---|---|---|
| Groundworks, slab and drainage | 1 to 3 | None | None |
| Steel beam installed | 4 | Week 4 | 4 weeks |
| Bricks and blocks, shell built | 4 to 6 | Week 4 | 3 weeks |
| Roof structure and covering | 7 to 8 | Week 7 | 1 week |
| Roof window | 8 | Week 8 | 6 weeks |
| Windows and bifold doors | 8 | Week 8 | 8 weeks |
| First fix, then plaster and drying | 9 to 11 | None | None |
| Second fix | 12 | None | None |
| Kitchen fitted (client's supplier) | 13 | Week 13 | 6 weeks |
| Decorating, snagging, handover | 14 | None | None |
Here is a task list for a 14-week build with durations, the week each material must be on site and each supplier's lead time.
For every material, give the latest week to place the order (week needed minus lead time; week 0 means before the build starts).
List the orders that must be placed before the build starts. Then list the five assumptions in this programme most likely to be wrong, ranked by how many weeks each would cost.The order-by arithmetic is simple enough to verify in a minute, which is why I trust the assistant with it: steel is week 0 (4 minus 4), bricks week 1, the roof window week 2, roof timber week 6 (7 minus 1), windows and bifold doors week 0 (8 minus 8), and the kitchen week 7. The result changes how the job starts. Two items, the steel and the windows and bifold doors, must be ordered before anyone has dug a trench, which means the deposit and the design sign-off have to be settled at contract signing. An experienced owner knows that in their head. The list makes it visible to the site manager and, more usefully, to the client.
The draft's weak spot was plaster: it allowed three days to dry and booked second fix straight after, and its list of risky assumptions never mentioned drying at all. Drying times come from your plasterer and the product's data sheet, not from the assistant, so the plaster line in the programme becomes two weeks and the list of assumptions gets a new first item.
Where to keep the programme is a separate decision. Options are the task scheduling in your job software (JobTread lists scheduling and daily logs in its one plan), a task app such as Fieldwire (Basic is free for 5 users, 3 projects and 100 sheets; Pro is $39 per user a month on annual billing), or a spreadsheet with a bar-chart view. The assistant drafts the sequence and asks the awkward questions; the dates live in the tool the crew opens. One warning for spreadsheet users: Excel's =COPILOT() worksheet function was retired on 14 September 2026, so templates that depend on it can show errors when they recalculate. Use the Copilot side pane instead, and for a deeper look at keeping subcontractors and deliveries aligned, see AI scheduling for small builders.
Site paperwork: logs, photos, variations and checklists
Paperwork fails when it is written from memory on Friday evening. Tools help by making the record cheap at the time: a photo with a tag, a voice note, a form that is already half filled in. AI adds two things, summaries and first drafts. The apps differ mostly in how much of that sits inside the base plan and how much is metered.
| Need | Tool type | What the vendor pages say |
|---|---|---|
| Daily log and its summary | Log module in job software | JobTread lists AI summaries that turn several daily logs into one shareable summary; its page does not say whether AI is metered |
| Organised photos | Photo app | CompanyCam: Core $63 a month on annual billing with a two-user limit, Crew $129, Scale $199, extra users $29. The plan table shows AI credits of 10, 100 and unlimited while its FAQ says AI starts at Crew, so ask which applies to Core |
| Plans, tasks and punch lists | Field task app | Fieldwire Basic is free (5 users, 3 projects, 100 sheets); Pro is $39 per user a month on annual billing |
| Inspections and checklists | Checklist app | Mitti, the app formerly sold as SafetyCulture, is free for up to 10 seats with up to 300 AI credits per seat a month and 5 active templates; Premium is $24 a seat annually or $29 monthly |
Week 6: five daily logs become one client update
A weekly client update is the paperwork builders most often skip, and it prevents the most phone calls. The prompt that worked:
Here are five daily logs from week 6 of an extension build. Write a client update of 120 words or fewer: what was done, what happens next week, and anything we need from the client by Friday.
Use only what the logs say. Do not judge whether we are on programme.
[paste the five logs]The draft read well: brickwork up to eaves level on three walls, steel delivered and set, roof timbers booked for Monday, and a request for the client's decision on external light positions by Friday. The first version of the draft, before the last instruction was added, also said "the build remains on programme" even though Wednesday's log recorded a day lost to rain. Status is a judgement, so the owner writes that sentence; the assistant only collects the facts.
A variation recorded the day it was asked for
Variations (change orders, in some markets) turn into arguments when nobody wrote down who asked for what and when. A filled-in record for the extension:
| Field | Entry |
|---|---|
| Reference and week | V-03, week 9 |
| Request | Extra double socket by the sink, asked for by the client on site |
| Price and time effect | $240 from the rate sheet; no change to the programme |
| Approval | Client's email reply of the same day, saved to the job |
| Evidence | Photo of the marked position, taken before first fix |
The assistant's part is the dull part: turning a diary line into the record and drafting the confirmation email. The price comes from your own rate sheet and the "yes" from the client. My tutorial on pricing variations and change orders with AI goes further into the numbers.
Checklists and method statements need a competent person
Mitti and similar apps can draft an inspection template from a description. That is fine for a pre-start checklist, as long as the site manager edits and approves it. Ask an assistant for a pre-start checklist for setting the steel beam and a typical draft covers access, lifting and protective equipment but says nothing about temporary support or the occupied room below, which are the points that matter on that job (the kind of gap to expect, and the reason I never treat a draft as a finished document). For safety documents the rule is simple: someone competent writes or approves it, and the AI saves them typing. My tutorial on writing a method statement with AI help covers what to check.
What three set-ups cost a five-person firm
| Set-up | What is in it | Monthly cost (list, USD) |
|---|---|---|
| A: assistant plus free tools | One ChatGPT or Claude plan, Fieldwire Basic, Mitti Free | About $20 |
| B: one platform | JobTread for five internal users ($199 plus four at $20) plus one assistant plan | About $299 on monthly billing; the vendor advertises 20% off annual billing, which would bring the platform to roughly $223 |
| C: separate specialists | Kreo Pro $175, CompanyCam Core for two users $92 ($63 plus one extra user at $29), Fieldwire Pro for two users $78, one assistant plan $20 | About $365, with annual billing on most of it; photos for the whole crew would mean CompanyCam Crew from $129 instead |
Two cautions on the sums. JobTread has no free trial, only a 30-day money-back guarantee on monthly subscriptions, so start monthly. And check how any platform counts crew who only clock time, because a per-user price is easy to underestimate. Set-up A looks cheap, and for a firm that mostly needs quote wording and a place for photos it is enough. It is the weakest on one thing: nothing in it links a variation to a programme change, so the owner is the integration.
Which set-up suits which firm
| Firm's biggest problem | Start with | Why |
|---|---|---|
| The owner prices two or three jobs a week from drawings | Assistant plus a takeoff trial | Hours sit in measuring, and a trial costs one month |
| Jobs slip because materials or subcontractors arrive late | Programme in job software plus the order-by prompt | The fix is visibility of lead times, not smarter AI |
| Arguments about extras and defects | Daily logs, photos and a variation record | Evidence written at the time settles most disputes |
| The administrator is buried in chasing | Assistant plus reminders in job software | See the first-job scoring in my companion tutorial |
For the extension firm I would pick set-up B's platform but trial it for one month on monthly billing, with the takeoff tool tested on three live quotes before any annual commitment. Its pain is late deliveries and unrecorded extras, and both are visibility problems that one system solves better than three. If the owner priced five extensions a month rather than one, the takeoff tool would move up the list. My tutorials on AI estimating software for small builders and on AI apps for photos, logs and checks on site compare more products in each area.
Where AI stops and you start
| Job | The AI drafts | You decide |
|---|---|---|
| Quotes | Takeoff counts, scope wording, exclusions, the cover email | Rates, margin, which lines are allowances, and the final figure |
| Programmes | The task sequence, order-by dates and a list of risky assumptions | Durations such as drying times, subcontractor availability, and whether to promise the client a handover date |
| Site paperwork | Log summaries, variation records and draft checklists | Status ("on programme"), variation prices and approvals, and every safety sign-off |
Pin that table where the team can see it. It is the reason a draft from the assistant never goes out unread.
Five questions to put to any vendor before you pay
- Which AI features are metered? BuildVision counts tokens, CompanyCam counts credits, and JobTread's page lists its AI tools without saying whether any are charged separately. Ask what happens when the allowance runs out.
- How are users counted? A per-user price grows with every carpenter who clocks in.
- Where do my drawings and photos go? Ask whether they are kept after a project closes and whether they are used to train models.
- Can I export everything if I leave? Logs and photos are your evidence in a dispute, so they must come out in a usable form.
- What does cancelling look like? Monthly or annual, refund window, and what happens to the data afterwards.
What to leave out for now
- An "AI project manager" that promises to run the job. The work is feeding it accurate data, which is the same work as running the job.
- AI-written risk assessments or method statements nobody competent has signed. A draft is a draft.
- Apps that need every subcontractor to log in. Ask each sub first; a platform half the crew ignores records nothing.
- An annual prepayment before a live-job trial. BuildVision's trial takes a card and bills on day 15, and its FAQ promises a full refund only if you ask within seven days of that first charge, so diarise the cancellation date.
- Prices taken from this page. They were correct when I read the vendors' pages and several vendors change them without notice.
Checking the set-up on the next job
After the extension, compare four numbers with the last comparable job: hours spent building the quote, the number of late deliveries, the share of working days with a log and photos, and the share of variations approved before the work started. If none of them moved, the tools are costing money without changing behaviour, and the cheapest thing to do is cancel the monthly plan. If you want a second opinion on which of these the firm should buy first, that is what my AI implementation consultation is for.
Further reads
- JobTread vs Buildertrend for a Small Builder: Price, AI and Fit — If one platform looks right, compare the two names builders hear most.
- AI Solutions for Construction Companies: One Platform or Point Tools? — The platform-or-separate-tools question argued at more length.
- JobTread Pricing for a Five-Person Builder: What You Pay and What AI Adds — What a five-person builder pays for JobTread, and what its AI adds.
- AI Software for Construction Scheduling: Programme Updates After Delays — Reissue the programme after a delay without redoing it by hand.
- AI Software for Contractors: An Office Stack for a Five-Person Firm — An office stack for a five-person firm, tool by tool.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Kreo, BuildVision, JobTread, Fieldwire, CompanyCam and Mitti pricing pages and the JobTread AI features page (checked October 2026); Microsoft's announcement retiring the Excel COPILOT function.