AI tools for office work: estimates, emails, job notes

Your staff are probably using AI already

Federal Reserve analysts estimated that about 41% of US workers reported using generative AI for work in November 2025, while only about 18% of firms had formally adopted AI^[1]. That gap means a lot of AI use happens without a company policy: an office manager drafting a tricky email on a personal account, a technician cleaning up notes on a phone. Small businesses report the same direction of travel, with 58% saying they used generative AI in 2025^[2].

You do not need a long policy. You need a short set of rules everyone can remember, a business account so you control the settings, and a shared library of prompts that produce consistent, on-brand drafts. The rest of this guide gives you all three.

Five rules for AI in the office

1. Use a business account with data controls switched on, not personal accounts. Check whether your inputs are used to train the vendor's models and turn that off where you can.
2. Never paste payment card numbers, bank details, Social Security numbers, alarm codes, gate codes, lockbox codes or medical details a customer mentioned.
3. Use first names or job numbers instead of full names and addresses when the draft does not need them.
4. A person reviews every output before it goes to a customer, a crew or a supplier. The reviewer owns what is sent.
5. If an output contains a price, a date, a model number, a warranty term or a legal statement, check it against the source every time.

These rules also protect you against a quieter risk: AI tools sometimes produce confident, specific details that are simply wrong. A draft that says a part is under a ten-year warranty when your manufacturer gives five creates a promise your company may have to keep.

A starter prompt library

Save these as templates in a shared document. Replace the bracketed parts each time. Each one tells the tool what role to play, what facts to use and what not to invent.

1. 01

   Estimate cover letter

   Write a short, friendly cover note for a [trade] estimate. Use only these facts: customer first name [name], work [summary], options [good, better, best], what is included [list], what is not included [list], how long the estimate is valid [days]. Do not mention prices, warranties or dates that are not in these facts. Plain language, under 150 words.

2. 02

   Reschedule or late technician email

   Write a brief apology to [first name] because their technician is running [time] late due to [reason]. Offer two choices: keep the appointment with a new arrival window of [window], or rebook for [options]. Ask them to reply with their choice. No discounts or promises unless I add them.

3. 03

   Clean up job notes

   Turn these technician notes into a clear job summary with headings: problem reported, findings, work done, parts used, recommendations, follow-up needed. Do not add findings that are not in the notes. Flag anything unclear as a question for the technician. Notes: [paste notes without customer name or address].

4. 04

   Maintenance reminder

   Write a reminder email for customers on our [plan name] that their [seasonal service] is due. Include how to book: [phone, link]. Keep it under 120 words, no hype, no invented statistics.

5. 05

   Training checklist from a procedure

   Turn this procedure into a numbered checklist a new CSR can follow on the phone. Keep our wording for safety instructions exactly as written. Procedure: [paste].

6. 06

   Job posting

   Draft a job posting for a [role] in [city]. Use these facts only: duties [list], schedule [details], requirements [list], pay range [range], benefits [list]. Do not add requirements I did not list.

Which office tasks fit and which do not

Translation deserves a note. AI tools can produce a readable Spanish or Vietnamese version of a service notice in seconds, but technical terms such as condensate line, breaker or termite bait station are easy to get wrong. Have a fluent staff member read anything customer-facing before it goes out.

The two-minute review checklist

• Facts: names, dates, times, addresses, model numbers and quantities match the source.
• Promises: no warranty, discount, guarantee or arrival time that you did not approve.
• Tone: sounds like your company, not like an advertisement.
• Privacy: no customer details that the recipient should not see, and nothing from another customer's job.
• Length: short enough to read on a phone.
• Disclosure: if the message will be read as personal advice from a named person, make sure that person actually agrees with it.

Disclosure matters to customers. In a 2025 Pew survey, 76% of US adults said it was extremely or very important to be able to tell whether pictures, videos and text were made by AI or by people^[3]. You do not need to label a reschedule email. You should never pass off AI-written review replies, testimonials or expert advice as something a person wrote from experience when they did not.

Mistakes that undo the time savings

Offices that give up on AI writing tools usually hit one of a few predictable problems. Each has a simple fix.

• Drafts that all sound the same. Fix it by pasting one of your own real emails as a style example and asking for the same tone.
• Long, padded messages. Add a word limit to every prompt and ask for plain language a customer can read on a phone.
• Reviewers who skim. Make the review checklist part of the habit, and have the owner spot-check a few sent messages each week.
• Prompts kept in one person's head. When that person is out, quality drops. Keep the library shared and named by task.
• Using AI to answer questions it should not, such as code requirements, permit rules or tax treatment. Send those to the right professional or the official source.
• Forgetting the customer record. A draft email is not a note. Copy the key facts into the job record so the next person sees them.

None of these is a reason to stop. They are reasons to treat AI drafting like any other office process: written down, owned by someone and checked.

What the time is worth

Office time in the trades is not cheap. Customer service representatives in the US earned a median of $21.53 an hour in May 2025^[4], and receptionists and information clerks at building equipment contractors earned a median of $41,030 a year^[5]. Saving an hour a day of drafting is real money, but only if the saved time goes to work that needs a person: calling back unsold estimates, handling complaints or keeping the schedule tight.

Example: a fictional electrical contractor's office manager spends about six hours a week writing estimate cover notes, reschedule emails and job summaries. With a prompt library and a review habit, the drafting time drops to about two and a half hours, including review. The owner asks her to spend the recovered time calling every estimate that has been open for more than five days. The value shows up in closed estimates, not in the hours saved.

Setting up the shared library

1. 01

   Collect your best real examples

   Find five emails, notes or letters your team is proud of. Remove customer details. These become the style examples you paste into prompts.

2. 02

   Write one prompt per task

   Use the starter library above. Keep the instruction not to invent facts in every prompt.

3. 03

   Test each prompt on three past jobs

   Compare the drafts with what you actually sent. Adjust until a reviewer would change only small details.

4. 04

   Store prompts where everyone can find them

   A shared document or your field service software's template library. Name each prompt by task.

5. 05

   Review the library every quarter

   Update prompts when prices, policies, service plans or warranty terms change.

Once the office habit is solid, you can look at AI for customer-facing work such as answering calls, which needs its own testing. Our AI receptionist evaluation checklist is the place to start.