Route every channel (web forms, email, listing-site leads, WhatsApp, social messages and missed calls) into one inbox, let AI send an instant reply that answers the specific question from your own stock and price data, and send a holding reply naming a time for anything needing judgement. A person then picks those up within the hour.
The five-minute target is for the first useful reply, not an acknowledgement. "Thanks for your enquiry, we'll be in touch" sent in ten seconds doesn't count, because it answers nothing and the customer carries on messaging your competitors. The goal is a reply that moves the conversation on: yes it's available, here's the price, here are two times you could come in, and here's who will call you about the rest.
What counts as a reply, and how to time it
Before building anything, agree the measure. I'd track time to first useful reply, per channel, and look at two figures: the median (the middle enquiry) and the slowest 10%. The median tells you what a typical customer experiences; the slowest 10% tells you about the Saturday-afternoon and Monday-morning pile-ups where most leads are lost.
Here's the difference between the two kinds of reply for an independent car dealership, the example used throughout this tutorial.
Before: "Thank you for contacting us. A member of our sales team will respond to your enquiry as soon as possible." Sent instantly, and useless: the customer asked whether the grey estate car was still for sale.
After: "Yes, the 2019 grey estate (stock 4471) is still available as of 10.42 this morning, at the advertised $14,950. We're open until 6pm today and 9 to 4 on Saturday; test drives take about 40 minutes. Would 11am or 2pm on Saturday suit you? One of our sales team will call you before noon about your part-exchange question."
Both went out in under a minute. Only the second one is a reply.
Audit: where enquiries arrive and how long they wait
Spend one week logging every enquiry: channel, time it arrived, time of the first useful reply, and who sent it. A spreadsheet with those four columns is enough. Here are the illustrative results for a dealership with six staff and around 240 enquiries a month:
| Channel | Share of enquiries | Median wait | Slowest 10% | Why it's slow |
|---|---|---|---|---|
| Listing-site lead emails | 38% | 3h 40m | Next day | Arrive in a shared inbox checked between customers |
| Website form | 22% | 2h 10m | Next day | Same inbox, often buried under supplier emails |
| Phone (missed calls) | 18% | No reply to most | Never | Nobody calls back numbers that left no voicemail |
| 14% | 50m | 5 hours | On the manager's phone only | |
| Social media messages | 8% | Next day | 3 days | Checked once a day |
The audit usually reveals that the problem isn't slow staff; it's enquiries arriving in five places that nobody watches continuously. That's why the first job is plumbing, not AI.
Three reply lanes: send, draft-and-approve, person only
AI shouldn't answer everything on its own, and it doesn't need to. Sort enquiry types into three lanes and write the list down, because the automation will follow it literally.
| Lane | Enquiry types (car dealership) | What happens | Target |
|---|---|---|---|
| 1: auto-send | Is it still available; price and mileage; opening hours; test-drive slots; what documents to bring | AI answers from the stock list and policy sheet and sends straight away | Under 2 minutes |
| 2: draft and approve | Part-exchange; "best price"; delivery; holding a car; anything the data doesn't cover | Instant holding reply answering what it can; AI drafts the full answer; a person approves from their phone | Holding reply under 2 minutes; full reply under 30 |
| 3: person only | Finance; complaints; warranty and faults; anything legal; a customer who sounds distressed | Holding reply naming who will call and when; no AI draft sent | Holding reply under 2 minutes; call within 2 hours |
Finance sits in lane 3 on purpose. Credit is regulated in most places, and a friendly AI sentence about monthly payments can easily become a promise you're not allowed to make. The same logic applies in other businesses: a wholesaler would put credit terms and trade-account applications in lane 3; a courier firm would put damaged-parcel claims there.
Building the pipeline, piece by piece
One inbox that everything lands in
Pick the place your team will watch: a shared email inbox, a CRM's conversations inbox, or a multi-channel inbox tool. Forward listing-site leads and form submissions to it, connect WhatsApp and social messages if the tool supports them, and send missed-call alerts there too (the missed call text back set-up covers the phone side). If you're staying with email, the shared-inbox triage approach shows how to label and sort what arrives.
One trap with listing-site leads: they often arrive from a no-reply address, with the customer's real email and phone number inside the message body. If your automation replies to the sender, the reply vanishes. Tell the AI step to extract the customer's contact details from the body and reply to those.
The data the AI answers from
Speed without accuracy is worse than slowness. The AI needs two sources: a live stock list (for the dealership, an export from its stock system to a spreadsheet every 15 minutes, with status, price, mileage and a "reserved" column) and a one-page policy sheet (hours, test-drive rules, documents, what you won't discuss by message). For a wholesaler the stock list would be a price list with trade tiers and lead times; for a storage facility, units free and prices.
The AI step that drafts and sorts
You can run this inside a CRM that has workflows (HubSpot's workflows need a Professional plan; Pipedrive's automations start on its Growth plan) or in Zapier or Make. The core is one prompt that sorts the enquiry into a lane and drafts the reply in the same pass:
You reply to enquiries for [dealership name]. You have:
- ENQUIRY: the customer's message and contact details
- STOCK: the matching row from the stock list (may be empty)
- POLICY: our policy sheet
1. Choose a lane:
LANE 1 if every question can be answered from STOCK and POLICY.
LANE 3 if it mentions finance, a complaint, a fault or warranty,
or legal matters, or the customer seems upset.
Otherwise LANE 2.
2. Write the reply (under 120 words, plain, friendly, no hype):
- Answer every question you can answer from STOCK and POLICY.
- Never give a part-exchange value, discount or finance figure.
- Never say a car is held or reserved.
- If STOCK status is not "available", say so and suggest
the two closest available cars from STOCK.
- For anything you can't answer, say who will follow up and when:
LANE 2 "within 30 minutes during opening hours",
LANE 3 "by phone within 2 hours during opening hours".
3. Return: lane, reply, customer_email, customer_phone,
questions_for_staff (a list).Test it on real past enquiries before switching anything on. A typical first run (illustrative) on the message "Hi is the grey estate still for sale? Would you take my 2014 hatchback in part-ex, it's done 90k. Could I come Saturday?" produced this:
Lane: 2
Reply: Hi, great news, the 2019 grey estate is still available at $14,950! We'd be happy to look at your 2014 hatchback; with that mileage you'd typically be looking at around $3,000 in part-exchange. Saturday works well, we're open 9 to 4. See you then!
Questions for staff: part-exchange valuation
The lane is right, but the reply broke the most important rule by guessing a part-exchange value, which a salesperson would then have to walk back in person. It also said "see you then" without an agreed time. The fix was to repeat the valuation ban at the end of the prompt, add "offer two specific test-drive times from POLICY", and add a check step in the automation that blocks any lane 2 draft containing a dollar figure not found in the stock row. After the fix, the draft read: "Yes, the 2019 grey estate is still available at $14,950. We can do a test drive at 10am or 1pm on Saturday; which suits you? Our sales manager will call you within 30 minutes to talk about your hatchback, as we'd need a few details to value it properly."
Send rules, approvals and alerts
Lane 1 replies send automatically. Lane 2 sends the holding reply automatically, then pushes the full draft to the approver's phone (a Slack or Teams message, a CRM task, or a text) with Approve, Edit and Reject options. Lane 3 sends the holding reply and alerts the right person by name. If you're writing replies by hand for lane 2, keep them sounding like you with the approach in drafting customer replies with AI that sound like you.
Mind the polling delay
This one catches people out. Zapier checks many triggers, such as "new email", on a timer rather than instantly: every 15 minutes on the Free plan, every 2 minutes on Professional and every minute on Team. On the Free plan, an enquiry can wait a quarter of an hour before the automation even sees it, which kills a five-minute target on its own. Use a paid plan, or an instant trigger such as a webhook from your form, for anything time-critical.
A worked example: the dealership after six weeks
Take the dealership from the audit. The owner and one salesperson built the pipeline over three days: forwarding rules into one shared inbox, a stock export every 15 minutes, a Zapier flow with an AI step, and approvals sent to the sales team's phones. Six weeks later (illustrative figures):
| Measure | Before | After |
|---|---|---|
| Median time to first useful reply (all channels) | 2h 50m | 1m 40s |
| Slowest 10% | Next day | 6 minutes (weekend evenings) |
| Lane 1 share of enquiries | n/a | 54% |
| Lane 2 median time to approved full reply | n/a | 24 minutes |
| Test drives booked per month | 31 | 43 |
| Wrong facts sent (found in weekly review) | n/a | 2, both from a stock list that hadn't updated |
The quick sum on running costs is worth doing, because task-based pricing adds up. Each enquiry runs about seven Zapier tasks (stock lookup, the AI step at one to three tasks, sending the reply, creating the CRM record, the staff alert). At 240 enquiries a month that's roughly 1,700 tasks, well over the 750 included in Zapier's Professional plan, so the dealership needs a higher task tier; Zapier's Team plan, for example, includes 2,000 tasks for $103.50 a month billed monthly ($69 billed annually). The AI itself costs almost nothing: using a cheap API model such as gpt-6-luna at $0.10 per million input tokens and $0.50 per million output, 240 enquiries of about 3,000 tokens in and 400 out cost well under $1 a month. Make, which starts from about $9 a month on credits, would run the same flow for less if someone on the team is comfortable with it. The full cost picture, including set-up help, is in what it costs to automate lead follow-up with AI.
Twelve extra test drives a month from faster replies is plausible, but be careful attributing all of it to speed; the dealership also started offering specific times instead of "come in any time", which on its own tends to get more bookings.
Where fast replies go wrong
- The stock list is stale. A car sold at 3pm, the export ran at 4pm, and at 3.30pm the AI told a customer it was available. The customer drove an hour to see it. Fix: export more often, add a "reserved" status staff set the moment a deposit is taken, and word lane 1 replies as "available as of [time]".
- Two replies to one customer. The automation replied, then a salesperson who hadn't seen it replied too, with a different price. Fix: log every automated reply in the CRM or inbox thread before it sends, and train staff to check the thread first.
- Bots replying to bots. A customer's out-of-office reply triggered your auto-reply, which triggered theirs. Fix: filter out messages whose subject starts "Automatic reply" or "Out of office", and cap automated replies at one per contact per hour.
- Answering scams quickly. Fast replies are also fast for fraudsters, such as a buyer who wants to pay more than the asking price and have you send back the difference. Fix: put any unusual payment request in lane 3, and never let AI discuss payment methods.
- Replies that sound like a machine. Exclamation marks, "great news!", identical wording to every customer. Fix: a short style card in the prompt (plain, warm, no exclamation marks, use the customer's words back to them) and a weekly read of ten replies.
Channels that need their own rules
WhatsApp. The free WhatsApp Business app offers greeting and away messages, which are acknowledgements, not answers. For AI-written replies you either use Meta's own Business Agent, which has been charged per token since 1 August 2026 (about 4 to 5 cents a message by Meta's estimate), or connect a tool through the WhatsApp Business Platform. From 1 October 2026 the platform starts charging for free-form service replies after the first 1,000 per business number each month, at rates that vary by the customer's market, so check Meta's pricing page before routing high volumes there. WhatsApp Business app versus the API explains which you need.
Phone. Missed calls are enquiries with no text attached. A missed-call text back that asks "How can we help?" turns them into written enquiries the pipeline can handle.
Social messages. Instagram and Facebook messages can feed most multi-channel inboxes. Keep lane 1 narrow here: people often ask vague questions ("price?") on a post, and the AI needs to know which post they're on to answer.
Trade and B2B enquiries. A wholesaler's customers often send a list of product codes and quantities. That's ideal for lane 1 if your price list is clean, and it's covered in depth in replying to trade enquiries in 60 seconds.
Keeping the five-minute promise after launch
Automations drift. The stock export fails silently, a new listing site sends leads in a different format, the approver goes on holiday. Three habits keep it honest:
- A weekly speed report. Median and slowest-10% reply times per channel, plus the percentage under five minutes. If one channel's figure jumps, something has broken upstream.
- An approval rota. Lane 2 only works if someone is always on approvals during opening hours, including Saturdays. Put it on the staff rota like any other duty, with a named backup.
- Ten replies read every Friday. Pick them at random across lanes. You're checking for wrong facts, wrong lane choices and tone. Each problem found becomes one line added to the prompt or the policy sheet.
Out of hours, decide honestly what happens. Lane 1 can keep running around the clock if your stock data is reliable. Lanes 2 and 3 should say "first thing tomorrow" with a real time, and someone should work through them before the doors open, because the enquiry that arrived at 9pm is the one your competitor answered at 8.05am.
Questions about five-minute enquiry replies
Will customers mind getting an instant reply written by AI?
Most don't, provided the reply actually answers their question and says who will follow up. What irritates people is an instant reply that says nothing, or one that gets a detail wrong. Sign automated replies from the team rather than a named person who didn't write them, and make it easy to reply and reach a human. Read a sample of replies each week to check the tone still sounds like your business.
Is a five-minute reply realistic for a one-person business?
Yes for the first reply, not always for the full answer. A sole trader can let AI send factual answers automatically and send a holding reply to everything else that names a realistic time, such as 'I'm on a job until 3pm and will call you then.' That still beats most competitors. The part that needs a person, such as pricing a custom job, can follow within a few hours.
What if two customers enquire about the same item at once?
Your stock data decides. If the item is marked available, both get a truthful reply that it's available now, and the first to book or pay a deposit gets it. Update the stock status the moment something is reserved, and have the AI say 'available at the time of writing' rather than promising it's held. Holding an item should always be a human decision or a deposit, not an automatic promise.
Further reads
- One AI Inbox for WhatsApp, Instagram and Facebook Enquiries — Pulling WhatsApp, Instagram and Facebook messages into one AI inbox.
- What Is the Cheapest Way to Handle Enquiries Out of Hours? — Low-cost options for evenings and weekends when nobody can approve.
- How to Automate Sales Follow-Ups With AI Without Being Pushy — What to send when the first fast reply gets no answer.
- How to Add an AI Chatbot That Captures Leads on Your Website — Capturing website visitors in chat before they fill in a form.
- What Is Lead Nurturing and How Can AI Automate It? — Keeping slower buyers warm after the first conversation.
- When Does Zapier Get Too Expensive? Finding the Tipping Point — When task counts from busy enquiry flows outgrow your plan.
- Can AI Help a Small Business Grow Revenue, or Only Cut Costs? — When AI grows revenue and when it only trims costs: diagnose your bottleneck, pick from five revenue levers, and check two worked examples with numbers.
- How Independent Hotels Win More Direct Bookings With AI — Five places direct bookings leak to online travel agents, what AI can fix at each, and the commission sum that tells you what it's worth.
- How a One-Person Travel Agency Can Reply to Enquiries Instantly — A two-speed reply system for solo travel agents: a useful answer within a minute, then an AI-drafted personal reply you check and send the same day.
- How Wedding Suppliers Can Reply to Enquiries in Minutes With AI — A facts sheet, five reply types and one prompt that let photographers, florists, bands and make-up artists send a personal first reply in minutes.
- How Tattoo Studios Handle Enquiries and Deposits With AI — How a tattoo studio can use AI to sort and answer enquiries, collect what artists need to quote, and take deposits safely through booking software.
- How Coaches Use AI to Qualify Discovery Call Bookings — Set fit criteria, ask four questions before the calendar, let AI sort and brief each booking, and send people who aren't ready somewhere useful.
- Answering Pupil and Parent Enquiries With AI at a Driving School — Build an answer sheet from your last 50 enquiries, put AI on web chat or WhatsApp, and hand test dates, complaints and progress questions to a person.
- How Removal Firms Turn Customer Videos Into Quotes With AI — A seven-step workflow for removal firms: from the customer's phone video to an inventory, a checked volume and a priced quote with clear assumptions.
- Tutoring Centre Admin With AI: Enquiries, Timetables, Invoices — Enquiry replies from a fact sheet, AI-proposed timetables checked for clashes, and invoices built from attendance: a tutoring centre's admin, step by step.
- Signs Your Business Is Ready to Automate Sales Follow-Ups With AI — A 16-point readiness checklist for automating sales follow-ups, each item with how to verify it, plus a filled-in example and a scoring table.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Zapier pricing and task-counting help pages (polling intervals, AI by Zapier task use); Make pricing page; OpenAI API pricing page; Meta Business Agent and WhatsApp Business Platform pricing notices. Figures checked September 2026.