7 minute read
Get your evenings back: three admin jobs AI can take off your plate, and where it goes wrong
See how AI could help with enquiries, quote follow-ups and review requests, with practical checks and clear human control.
AI stands for artificial intelligence. In this guide, it means software that uses patterns in data to draft text or answer questions. If you spend the day on jobs, the useful question is simple: what part of the paperwork or messaging could it help with? Start there. You do not need a talking box on your website just because the tool is available.
AI can get things wrong, so you stay in charge. The National Cyber Security Centre explains that AI tools can present false information as fact and can be manipulated by what people send them. Its AI and cyber security guide is a useful starting point.
The jobs below are possible uses, not promises about a particular product. Each needs a clear setup, testing and someone responsible for checking what happens.
In this guide
- Answer common questions and collect enquiries
- Help follow up a quote
- Help ask for honest reviews
- Keep control of the information
- Test before customers rely on it
Answer common questions and collect enquiries
A customer might ask whether you cover their town, take on a type of work or offer a call back. An AI chat tool could use an approved list of services and areas to draft answers and collect the details you need to respond.
Give it a short, current set of facts. Include what work you do, what you do not do and how the next step works. If you only arrange visits after a phone conversation, say that in the instructions. Make the same information clear on the website so a visitor does not have to use chat.
For a plumbing business, a sensible enquiry might collect a name, town, contact number and short description of the job. It should explain that a person will assess the request. Do not let the tool turn “I can ask about a visit” into “your plumber is booked”.
The failure to watch for is a confident answer that sounds like your business but is not true. It might invent an appointment, expand your service area or claim a qualification you have never listed. An answer can sound friendly and still be wrong.
Keep a clear route to a person. Set the tool to stop and pass on questions it cannot answer from the approved information. Do not use it to diagnose hazards, give repair instructions for dangerous situations or decide that a job is safe. Agree a reviewed response for urgent messages before you switch it on.
Make it clear that the visitor is using an automated assistant. Keep phone and email contact available alongside it. If you cannot review the conversations and correct errors, a straightforward enquiry form may be easier to manage.
Help follow up a quote
An AI tool could draft a short follow-up after you have sent a quote. The useful part is helping with the wording or organising which requests need attention. A simple reminder system may be enough if the message rarely changes.
Begin with drafts that you approve before sending. Check the customer, the job and whether a follow-up is still needed. A suitable message might ask whether the customer has questions about the work discussed. It should not pressure them to decide or pretend that a slot has been reserved.
Keep the agreed scope out of the tool’s control. It should not change what is included, extend a commitment or accept a job without your approval. If the customer asks for something different, send that reply to you for a decision.
The main mistakes here are poor timing and stale information. A reminder could go to someone who already accepted, declined or asked not to be contacted again. It could refer to an earlier version of the job. It could send the same message twice after a connection problem.
Decide what stops a follow-up before deciding what starts one. Record when the customer replies or the job changes. Give yourself a visible way to pause messages. Check the queue before sending rather than assuming that every record is current.
Agree what happens if the system loses access to your records. It should leave uncertain messages for review. A missing answer is a reason to check, not a reason to guess about the customer’s decision.
Help ask for honest reviews
A workflow could prepare a review request after a completed job. You may not need AI for this at all: a standard message and a clear completion record can be enough. AI might help draft wording, but it must not write the customer’s review.
Google offers a link or QR code for customers to leave feedback. Its review request guidance explains the process. Ask for an honest account of the customer’s experience, with no suggested rating or reward.
Do not build a system that sends the Google link only to happy customers. Google’s content policy prohibits selectively asking for positive reviews. A feedback question must not be used as a gate that hides the review route from unhappy customers.
Use the same fair request rule for completed jobs. Keep service complaints with a person who can help. Do not make removing or changing a review a condition for resolving an issue.
Watch for messages sent before the work is complete, requests sent repeatedly or a reply that reveals customer details. Check any drafted response before publishing it. The tool should not invent a memory of the job or make promises about a complaint you have not investigated.
Keep control of the information
Before putting customer details into a tool, find out who provides it, what information it receives, where that information goes and how long it is kept. Ask whether it uses conversations to improve its models and what controls your account provides. Record the answers for the actual service you choose.
Collect only what you need for the task. Do not ask for gate codes, payment details or sensitive household information in a general chat. Use made-up details during testing. Keep information from one customer out of answers to another.
Your arrangements also need to reflect data protection requirements. The ICO’s AI guidance is the official starting point for working through that responsibility. A chat tool does not remove it, and a friendly message is not a substitute for a clear privacy notice.
Limit what the tool can do. A tool that drafts a reply does not need permission to alter every customer record. Keep a person responsible for changes, access and switching the system off when something goes wrong.
Test before customers rely on it
Write realistic questions, including ones the tool should decline. Try a town outside your area, a service you do not offer, a request to confirm an appointment and a complaint. Check what happens when the tool has no answer or a connection fails.
Test the whole handover. Can you find the enquiry? Is the contact number intact? Does the customer know whether the message arrived? Can you stop a pending follow-up after a reply? Keep notes of failures and fix them before widening use.
Start with a task where you review the result before it reaches a customer. Add more only when the checks and daily workload make sense. AI can get things wrong, so you stay in charge.
See how we take the admin off your plate for the jobs Tovi can discuss with you, or what we do for plumbers. A clear site and a working enquiry route are a useful place to begin.