The First 30 Days of AI in a Trades Business

The first 30 days of AI in a trades business should cost under $50 and cover three jobs: replying to enquiries, turning site notes into quotes, and chasing invoices or reviews. Spend days one to three logging where admin time goes, days four to seven setting up one assistant, weeks two and three using it, and week four measuring.

The trap is buying before you know what you need. Owners who start with an AI receptionist, a new job-management system or an estimating package usually spend the month configuring software and end it with no hours saved. The exception: if your log shows you miss more than ten new-job calls a week, that problem jumps the queue, and it gets its own section below.

Follow me on Instagram@sagnikteaches

Before day one: three decisions that shape the month

Settle these before you open any AI tool, because they stop the month drifting.

Connect on LinkedInSagnik Bhattacharya

• Who runs it. One person owns the experiment. In most small firms that is whoever does the quotes and invoices, often the owner or an office partner. Site staff feed in voice notes and photos the way they already do.
• The budget ceiling. $50 for the month. That covers one paid individual assistant plan (about $20 a month for ChatGPT Plus or Claude Pro at list price) with room to spare. Nothing else gets bought until day 30.
• The send rule. For all 30 days, nothing AI-drafted goes to a customer until a person has read it. This is the single rule that prevents embarrassing mistakes while you learn what the assistant gets wrong.

Throughout, the main example is an illustrative four-person electrical contractor: the owner, who quotes and does some jobs; two electricians; an apprentice; and a partner who does the admin three days a week. The numbers are made up but realistic for a firm that size.

Subscribe on YouTube@codingliquids

Days 1-3: log where the admin hours actually go

Most owners guess their admin time and guess wrong. Quotes feel like the biggest drain, but the log often shows it is chasing, rewriting texts, or phone tag. For three working days, whoever does the admin jots each task in a notebook or a phone note: what it was, how long it took, and whether it was repetitive.

At the end of day three, scale it up to a week. Here is the contractor's log, scaled:

Pick your three jobs from the "yes, repetitive" rows with the most hours. For this firm that is enquiries, quotes and invoice chasing, with reviews as a quick add-on because it costs almost nothing. Certificates stay out of the month: they are formal documents with their own rules and software, and a chat assistant adds risk there without saving much.

Days 4-7: set up one assistant with a business brief

Pick one assistant and stick with it for the month

ChatGPT, Claude and Gemini all handle this work. Pick based on what you already use: if the firm runs on Google Workspace, Gemini is already inside Gmail on the business plans; otherwise a paid individual plan of ChatGPT or Claude at about $20 a month is the simplest choice. Switching tools mid-month wastes the learning. Before anything else, open the settings and turn off the switch that lets the provider train on your chats (in ChatGPT it is "Improve the model for everyone" under Data controls).

A business plan such as ChatGPT Business or Claude Team adds admin controls and keeps business content out of training by default, but both require at least two seats. For month one, with one person running it, an individual plan is enough. Revisit this on day 30.

Write the business brief once

The brief is a one-page document the assistant reads before every task. Both ChatGPT and Claude let you create a Project, a workspace where you save instructions and files, so you paste the brief in once rather than every time. This is the most valuable hour of the month. An illustrative brief for the contractor:

BUSINESS BRIEF - read before every task
Who we are: 4-person electrical contractor, domestic and small commercial.
Work we do: rewires, consumer unit upgrades, EV charger installs,
  fault finding, lighting, sockets, landlord safety checks.
Work we don't do: solar panels, alarms, anything over 3 storeys.
Area: within roughly 25 miles of the yard. Further = politely decline.
Call-out: $95 first hour, $65 per hour after, parts at cost plus 15%.
Quotes: free for jobs over $500; site visit usually within 5 working days.
Payment: 30% deposit on jobs over $2,000, balance within 14 days of
  completion. Card or bank transfer.
Guarantee: 12 months on workmanship. Manufacturer warranty on parts.
  Never promise anything beyond this.
Tone: friendly, plain, short. No jargon unless the customer used it.
  Sign off as "The team at [firm name]".
Never: give a firm price without a site visit, promise a date,
  diagnose a fault by message, or mention a competitor.

Every line in that brief exists to stop a specific mistake. The "work we don't do" line stops the assistant cheerfully quoting for solar. The guarantee line stops it inventing a longer one. The "never" list stops the riskiest drafts before they are written.

What never goes into the assistant

Write this on the same page: customer alarm codes, key-safe numbers, door codes, bank details, and anything about a customer's health or vulnerability. An apprentice pasting a whole customer text thread into a chat is how an alarm code ends up in a chat history. Tell the team the rule on day four, not after it happens.

Telling the team on day four

Site staff hear "AI" and assume either that they are being watched or that their jobs are going. Neither is true here, and a short message on day four heads off both worries. An illustrative version the contractor sent to the team chat:

"Quick one. For the next month we're trying an AI assistant for the office side: replying to enquiries, typing up quotes and chasing invoices. Nothing changes on site. The only ask: after a quote visit, send me a voice note with what you saw, same as now but a bit more detail (sizes, board type, access). Please don't paste customer messages or codes into any AI app yourselves. If a quote comes back wrong, tell me, that's the point of the trial."

Two things make this work. It asks site staff for one small change, not a new app. And it tells them their corrections are wanted, which turns the most sceptical electrician into the best tester you have. Expect the voice notes to be patchy in the first week; by the second, once people see their notes turn into tidy quotes, they usually get more detailed on their own.

If one person refuses to use it at all, let them. The month is about the admin seat. Forcing site staff onto a tool before it has proved itself is the fastest way to have the whole team dismiss it.

Week 2: enquiry replies and quote drafts

Enquiry replies in under a minute

Paste the customer's message into the Project and ask for a reply. Here is a real-looking before and after.

The enquiry: "hi do u do ev chargers, we have a tesla coming next month, house is semi detached drive at front, how much roughly and how soon"

What the owner used to send, at 10pm: "Yes we do. Price depends. Can come look next week."

The AI draft (illustrative), after one edit: "Hi, thanks for getting in touch. Yes, we install EV chargers regularly. For a semi-detached house with a front drive the install is usually straightforward, but the price depends on the charger you choose, the distance from your fuse board to the drive, and whether your supply needs any upgrade. We quote for free after a short site visit, and we can usually get out within five working days. Could you send a photo of your fuse board and tell us which charger you have in mind? The team at [firm name]."

The edit the owner made: the AI's first draft said "we can install before your car arrives", which is a date promise. It came out. That is exactly the kind of fix you are learning to spot in week two. For a faster routine built around this, see replying to trade enquiries in 60 seconds.

Quotes from site notes

The owner records a voice note in the van after each visit, or types rough notes. The assistant turns them into a quote the owner then prices. The prompt:

Using the business brief, turn these site notes into a customer quote.
Structure: short intro, scope of work as a numbered list, what's excluded,
price table with lines for labour and materials (leave the numbers as
[PRICE] for me to fill in), payment terms and guarantee from the brief.
Do not invent any work not in the notes. List anything unclear at the end
under "Questions for the customer" instead of guessing.

Notes: [paste voice note transcript]

An illustrative voice note: "Mrs P, kitchen extension, wants six downlights, two double sockets on the island, cooker circuit moved, board is old rewireable fuses so needs a new consumer unit, plasterer is in on the 14th so first fix before then."

And the useful part of the illustrative output: a scope list of four items, an exclusions line ("making good plaster and decoration"), and under "Questions for the customer": "Is the island floor-fed, and who is lifting the floor?" and "Do you want to quote the consumer unit as a separate line so the customer sees it?" Those questions are the reason the prompt tells the AI not to guess. The owner fills in prices from his own rates, which is the part AI should never do unsupervised.

Leaving prices blank is deliberate. Chat assistants do not know your rates, supplier prices or how long a job takes you, and a draft with made-up prices is worse than one with gaps. The workflow for recording on site is set out in turning on-site voice notes into job sheets and quotes.

Week 3: invoices, reviews and the missed-call question

Invoice chasers that go out on time

Chasing is put off because it is awkward, not because it is hard. Ask the assistant once for a three-stage set, then reuse it. An illustrative second-stage reminder:

"Hi Mr D, just a reminder that invoice 1047 for the rewire at your property, $3,840, was due on the 3rd. If you've already paid, thank you and please ignore this. If not, you can pay by card using the link below or by bank transfer. Any questions about the invoice, just reply here. The team at [firm name]."

The first stage is a friendly nudge on the due date, the second follows seven days later, and the third, at 21 days, mentions your terms plainly. Keep all three in the brief so every chaser sounds the same. More on the tone and timing is in polite AI invoice reminders for trades.

Review requests after every job

The log showed the firm asked for reviews almost never. A short text sent the evening a job finishes changes that. Ask the assistant for three versions and pick one:

"Thanks for having us today, Mrs P. If you're happy with the downlights and sockets, a quick review really helps a small firm like ours: [link]. And if anything isn't right, reply here and we'll sort it."

The second sentence matters. It gives an unhappy customer a private route before a public one. In week three the contractor sent six requests and got three reviews, which was three more than the previous month.

The missed-call question

Fourteen missed calls a week, five of them new jobs, three lost. That is the number that tempts owners to buy an AI receptionist. Before you do, work out what the missed calls cost: if three lost jobs a week average $600 each, that is real money, and a fix is worth paying for in month two.

There are two tiers of fix. A missed-call text-back sends an automatic text ("Sorry we missed you, we're on a job. What can we help with?") and costs little or nothing on many phone systems. An AI receptionist answers the call, takes details and books a visit. Jobber's Receptionist, as one example, costs $29 a month for 30 conversations and $0.79 per conversation after that, and some other job-management apps offer something similar. The costs and trade-offs are laid out in what an AI receptionist costs a tradesperson. In month one, just count: how many missed calls, how many were new work, how many called back.

Week 4: count the hours and decide what to pay for

On day 30, redo the admin log for three days and compare. Here is the contractor's scorecard:

Cost for the month: one $20 plan. The owner values his evening time at about $50 an hour, so five hours a week is worth around $250 a week to him, before counting the faster payments and the reviews. That is an illustration, but it is the calculation to do with your own numbers. If your total saving is under two hours a week, the month still taught you something: either the jobs you picked were not the right ones, or the brief needs work.

Thresholds for spending more

• Saved three or more hours a week: keep the paid plan. Consider a second seat or a business plan if a second person should use it.
• Missed more than ten new-job calls a week: trial a text-back or AI receptionist in month two and measure calls answered and jobs booked.
• Copying the same data between two apps every day: that is a job for an automation tool. Zapier's free plan covers 100 tasks a month on two-step automations; its paid Professional plan starts at $19.99 a month billed annually for 750 tasks.
• Saved under two hours: do not buy anything. Revisit the brief and pick a different job.

Five mistakes from month one and how each showed up

These are illustrative, but each happens in the first few weeks at firms that skip the send rule or the brief.

1. An invented guarantee. Before the brief had a guarantee line, a draft reply promised "a 5-year guarantee on all work". The customer would have held the firm to it. Found by reading every draft; fixed with one line in the brief.
2. A placeholder sent to a customer. "Dear [Customer Name]" went out in a text because the office partner copied the draft too quickly. The fix was a habit: search each draft for "[" before sending.
3. A quote for work the firm doesn't do. An enquiry mentioned solar panels and the assistant drafted a helpful reply offering a site visit. The "work we don't do" line now stops it, and the reply politely declines.
4. A sensitive detail in the chat history. The apprentice pasted a whole customer thread, including a key-safe code, to ask for a reply. The code was deleted from the chat and the customer was told. The team rule now covers it explicitly.
5. Too-polished replies. Customers used to short, friendly texts got three paragraphs. The brief now says "short", and the owner asks for "two sentences, like a text" for anything by SMS.

How the plan bends for a sole trader or a bigger crew

The same 30 days work for different sizes, with a few changes.

A sole-trader decorator does everything from the van, so the log is just a note on the phone and the assistant lives in the phone app. The biggest wins are usually enquiry replies and quote write-ups from photos. Voice input matters more than any other feature: talk the quote through while standing in the room, and have the assistant turn it into a tidy document that evening. A two-seat business plan makes no sense here; an individual plan does.

A two-person roofing firm spends much of its admin time on quote follow-ups, because roofing quotes are large and customers sit on them. Swap invoice chasing for quote follow-ups in week three: a polite message at day three, day ten and day twenty-one after the quote goes out.

A ten-person plumbing and heating firm with an office manager should run the same month, but the log will usually point at scheduling and job-sheet admin rather than quotes. Check the AI features in the job-management software first: many now include message rewriting, suggested replies and report builders, and using those avoids adding a separate tool the team has to learn.

Rules to write down before month two

By day 30 you know enough to write a short rules sheet. Keep it to one page and pin it where the team will see it:

1. One named person approves anything AI-drafted before it goes to a customer.
2. No prices, dates or guarantees unless they come from the business brief or the owner.
3. No codes, bank details, health information or anything about vulnerable customers goes into the assistant.
4. The business brief gets updated the same day a price, service or policy changes.
5. Certificates, safety paperwork and anything you sign off professionally are written in the proper software and checked by the qualified person, never generated by a chat assistant.
6. Anyone who spots an AI mistake adds it to a shared note, so the brief can be fixed.

What month two usually looks like

Month two is where money starts to make sense, because now you know which job the AI does well for you. Common next moves, in the order they usually pay off for small trades firms:

• Give site staff voice-note-to-job-sheet access if quotes and job sheets were the biggest saving.
• Fix missed calls with a text-back or receptionist if the count justified it, measuring jobs booked rather than calls answered.
• Automate one handoff, such as a new web-form enquiry creating a job in your software and drafting a reply for approval.
• Look at estimating tools only once your quoting process is written down, because an estimating tool automates a process; it cannot invent one.

Whatever comes next, keep the admin log habit. Three days of logging every quarter tells you whether the AI is still saving time or whether the team has quietly gone back to the old way. If you would rather plan month two with someone who has seen how other firms set this up, that is the kind of thing my AI implementation consultation covers.

Questions from trades owners starting with AI

Do my engineers or electricians need their own AI accounts in month one?

Usually not. In the first month one person, normally whoever does the admin, runs the assistant and everyone else feeds it voice notes and photos through the usual channels. Give site staff their own access in month two if the scorecard shows quotes or job sheets are the bottleneck, and at that point a business plan with admin controls is worth pricing.

Is the free version of ChatGPT good enough for the first 30 days?

For week one, yes. You can test enquiry replies and quote wording on a free plan. Free tiers have tighter limits and fewer features, and the model-training switch needs turning off in settings. Most owners who use it daily move to a paid individual plan, around $20 a month, by week two.

What if I already use Jobber or similar job software?

Check its AI features before adding anything. Jobber, for example, has a Rewrite button for messages, an AI Receptionist and suggested automations, so some of this plan may already be built in. Run the same time log, then decide whether the built-in features or a separate chat assistant saves more time for the job you picked.

How do I stop AI promising things on my behalf?

Keep a human send step for everything customers see in month one, and put your real guarantees, payment terms and exclusions in the business brief. Then add a line to your instructions telling the assistant never to state a guarantee, price or date that is not in the brief. Check every draft for promises before it goes out.

Further reads

• Free vs Paid AI Tools for a Small Trades Business — Decide which AI tools are worth paying for once month one is done.
• What Tradespeople Actually Use AI For: 12 Real Jobs It Handles — Twelve jobs tradespeople hand to AI, for picking month two's target.
• Jobber, Housecall Pro, and ServiceTitan AI Features Compared — Compare the AI built into the main job-management apps.
• Seven AI Mistakes That Cost Tradespeople Jobs — The mistakes that lose jobs, so month two avoids them.
• How to Set Up Missed Call Text Back for a Service Business — A cheaper first fix for missed calls than a full AI receptionist.
• When Should a Trades Business Bring In an AI Consultant? — When it makes sense to bring in outside help.
• How to Add AI Steps to Zapier: Classify, Summarise, and Draft — How to add AI to an automation once the manual version works.
• AI Tools for Roofing Contractors in 2026: Estimating and Leads — Roof measurement, instant estimates, lead replies and photo documentation tools for roofers, with current prices and three stacks by firm size.
• How a Plumber Can Save Five Hours a Week on Admin With AI — An illustrative time log for a sole plumber, six changes that add up to five hours a week, a four-week rollout and what it costs.
• How Small Builders Use AI to Price Variations and Change Orders — Turn a site conversation into a priced, signed variation: voice-note capture, rate-sheet pricing, knock-on cost checks and a variation register.
• AI Scheduling for Small Builders: Subcontractors and Deliveries — Use AI to build the trade sequence, turn lead times into order-by dates, run a Friday look-ahead and re-plan when a subcontractor drops out.
• Nine Everyday AI Uses for Electricians Beyond Estimating — Nine electrician-specific jobs AI handles, from explaining inspection findings to landlords to logging intermittent faults, each with a worked example.
• Why Most Landscapers Haven't Adopted AI Yet, and Where to Start — The honest reasons landscaping firms hold off on AI, the three office jobs to start with, and a first-month plan for a four-crew firm.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: OpenAI and Anthropic pricing pages; Jobber AI features page and Receptionist help article; Zapier pricing page; OpenAI help pages on data controls.