AI Chatbot for CRM Systems: Send Website Enquiries to Job Software

Pick one of three routes. Send the visitor to your job software's own enquiry form, have the chatbot post what it captured through Zapier, or email a summary to an address the software reads. Your software and plan decide which routes exist, so check that first, then map six fields and test for duplicates.

Most generic guides assume a general-purpose CRM with an open API. Trades software behaves differently. Some job apps take enquiries natively on every plan, some keep Zapier and API access for a top tier, and for two popular ones I couldn't find a Zapier listing at all. Your plan usually decides the route before the chatbot does, so start with the software, not the bot.

Follow me on Instagram@sagnikteaches

Which routes your job software allows

This table comes from each vendor's own pricing and help pages, plus Zapier's app listings, read in October 2026. Where a cell says "check", the vendor doesn't publish the answer, which is not the same as a no.

Connect on LinkedInSagnik Bhattacharya

The Housecall Pro row matters most for a small firm. Paying $240 a month more than the Basic plan just to connect a chatbot makes no sense for a one-van business, so that firm should let the chatbot hand visitors over to the online booking page instead. Jobber's gap is smaller: its pricing page showed Core at $49 and Connect at $139 a month on monthly billing, so check whether you were going to move up anyway for the follow-up and Zapier features.

Subscribe on YouTube@sagnikteaches

If you are still choosing software, Jobber, Housecall Pro and ServiceTitan AI features compared goes further than the intake question, and AI receptionist built into job software or standalone covers the phone-and-text side of the same decision.

Six fields, in the order the bot should ask for them

A chatbot that captures too little gives you a name and a mystery. One that interrogates loses the visitor at question four. For a trades firm I'd settle on six fields, asked in this order, plus two admin fields the bot fills in itself.

The order is deliberate. Job type and postal code come first because they let the bot refuse a job it can't do before it has taken anyone's phone number. Contact details come once the visitor has seen that the bot can help. Email is optional if there is a mobile number, because a missing email costs you less than a visitor who closes the window.

Here is the instruction I'd give a bot for a security installer, written as plain rules rather than clever wording:

You are the website assistant for [business], a security installer.
Your jobs: answer questions from the approved sheet, and collect an enquiry.
Ask for these one at a time, in this order:
1 job type (menu), 2 postal code, 3 full name, 4 mobile number (digits only, read it back),
5 when they need it (today, this week, flexible), 6 optional photo.
Before step 3, check the postal code against the service-area list. If it is not on the list,
say we do not cover that area yet, recommend nobody, and stop.
Never promise a price, a visit time or a guarantee.
If someone describes a break-in or danger happening now, tell them to call the emergency services
first, then offer a callback.
When all fields are collected, repeat them back and ask "Is that right?"
Send the enquiry only after a yes.

A short sample exchange shows how the fields come through, with the final structured message the bot hands to your software:

Visitor: Hi, I need cameras at the front of my shop.
Bot: Happy to help. Is this a new camera install, a fault with cameras you already have, a quote for a bigger job, or something else?
Visitor: New install.
Bot: Thanks. What's the postal code of the shop?
Visitor: [postal code on your list]
Bot: Good, we cover that. What's your full name?
Visitor: [first name] [last name]
Bot: And a mobile number, digits only please?
Visitor: [mobile number]
Bot: When do you need this: today, this week or flexible?
Visitor: This week.
Bot: A photo of the front of the shop would help the installer. Send one here or skip. Then I'll read everything back.

{
  "source": "website chat",
  "job_type": "New camera install",
  "postal_code": "[postal code]",
  "name": "[first name] [last name]",
  "mobile": "[digits only]",
  "when": "this week",
  "summary": "Shop front, wants night view of the car park. Photo link: [link]",
  "marketing_consent": "no"
}

Route A: hand the visitor to your own enquiry form

This is the lowest-effort route and the right one when your plan has no Zapier access. The chatbot answers questions, confirms the job type and postal code, then gives the visitor the link to your job software's enquiry form and says it takes about two minutes. Everything the form collects becomes a record without any integration.

Jobber shows how well this can work. Each request form has its own link, and its help centre describes an HTML embed code plus WordPress and Wix plugins for putting the form on a page. The default form has an image upload field that accepts up to 10 files, each under 10MB, in JPEG, PNG or HEIC format. A service-area setting makes clients enter an address inside your area. New people who submit are added to the client list as leads.

Two practical tips. Create a separate form just for the chatbot, so the count of requests from that form tells you how many chats converted. Jobber's own page warns that Google Analytics 4 tracking may not work reliably on an embedded form, because it depends on third-party cookies, so a separate form is a more dependable counter than analytics.

The weakness is the hand-over itself. The visitor retypes details they've just given the bot, and some won't. The chat transcript doesn't travel with the request either. A hybrid fixes both: let the bot ask for a mobile number before sharing the link, and post that bare lead through Route B, so a visitor who abandons the form still leaves something you can follow up.

Route B: post the captured fields through Zapier

This route moves the bot's answers into the software without the visitor lifting a finger. It needs three things: a chatbot that can send data out, a job software that Zapier can write to, and a paid Zapier plan, because Zapier's Free tier allows only two-step Zaps and this needs more. Professional lists at $19.99 a month billed annually or $29.99 monthly for 750 tasks.

On the chatbot side, check four things on the vendor's pages before you commit:

• Can it send data out? Tidio's Zapier page describes a "Send to Zapier" action you drop into a flow, and says the integration is available on all Tidio plans, including the free one. Its pricing page doesn't mention Zapier, so confirm in your account. Chatbase's actions page lists a collect-leads action and custom actions that call your API. Botpress's pricing page lists HubSpot, Salesforce, Zendesk and Intercom among its integrations and doesn't mention webhooks.
• Can it ask in a fixed order and read back? Free-text capture is what produces phone numbers with letters in them.
• Can it accept photos, or at least hand out a link? See the photos section below.
• What does a plan cap? Botpress's free plan allows 100 conversations with no top-ups or overages. Chatbase lists no free plan, with Hobby at $40 a month or $32 on yearly billing for 700 message credits. Tidio's Free plan gives the first 50 AI conversations free, for life.

With Jobber, the Zap looks like this. The names are the ones Zapier lists, and the order matters.

1. Trigger: the chatbot's "send to Zapier" step posts the six fields (no task used).
2. Filter: stop unless there is a name and at least a phone or an email. Jobber's own request form needs a contact method, so I'd hold the Zap to the same standard (no task).
3. Formatter: strip spaces and brackets from the phone number (Formatter steps don't use tasks).
4. Find or Create Client: one step that searches by name first and only creates a client if none matches (1 task).
5. Create Request: needs the client and takes a title only, so build the title as "Job type, postal code" (1 task).
6. Add Note to Request: paste the summary, the preferred timing and any photo links (1 task).
7. Alert: a text or email to whoever answers new enquiries, optional (1 task).

That is three or four tasks per lead. At 30 leads a month it is 90 to 120 tasks, a small slice of the 750 in the entry tier. Zapier's help page confirms that triggers and Filter steps use no tasks.

The ServiceM8 version is shorter: Create Client then Create Job. Zapier's list shows no Find action for ServiceM8, so the Zap can't check for an existing customer. Housecall Pro is the awkward one. Zapier showed Create Customer as its only action when I looked, and the connection sits on the Max plan, so for a small firm on a lower plan Route A, handing over to its online booking page, is the realistic choice. Zapier's help pages say it turns a Zap off by itself only when the Zap has run more than 20 times in seven days and 95% of those runs errored. A quiet enquiry Zap that handles a few leads a week can therefore fail for days without tripping that rule. Look at the Zap history weekly, and make sure the error emails reach someone who reads them.

The step-by-step for the chatbot side of this is covered in Tidio and Zapier: route chat leads and handovers automatically, and how to add an AI chatbot that captures leads on your website covers the chatbot basics this tutorial skips.

Route C: email the summary to an address your software reads

If Zapier is out of reach, an email can carry the same fields. ServiceM8's pricing page lists an Inbox from the Growing plan, described as an email address for your account that lets you "easily convert emails into new job cards". Tradify's page describes an enquiry email address on its Pro and Plus plans that brings enquiries straight in. The chatbot, or your chat tool's transcript email, sends to that address.

Be clear about what you get. ServiceM8's wording suggests you convert the email yourself, so this route saves the retyping but not the step of creating the record. Treat the lead as real only once it's converted, and give someone the job of converting new emails each morning and afternoon.

To make a converted email useful, standardise the message the bot sends. Compare these two sample messages:

Before: "Chat transcript from website. Visitor wanted cameras. Said shop. Number is in the chat somewhere."
After: "NEW ENQUIRY, new camera install, [postal code], [first name] [last name]. Mobile: [digits]. When: this week. Notes: shop front, night view of car park. Photo: [link]. Marketing consent: no."

The second version takes twenty seconds to turn into a job card and can be searched by postal code later. The first one is the reason lead-handling goes wrong.

Stop one customer becoming three records

Duplicates arrive from three directions: the same phone number typed two ways, one household enquiring under two names, and returning customers who don't say they are returning. A duplicate looks harmless until two people call the same customer, or a past job's history sits on a different record from the new request.

Where the check can happen depends on the route. In Route B with Jobber, the Find or Create Client step does it, but Jobber's help pages don't say how a new request is matched to an existing client, so test rather than assume. Zapier describes the step as finding the client that "aligns most closely with the search criteria", and the field it searches is the client's name. A name spelt two ways can therefore still create a second client, while two different customers who share a name can be matched to one record. In Route B with ServiceM8, Zapier lists no Find action, so a duplicate is likely unless you add your own lookup. In Routes A and C the check is human. Whichever route you use, the bot should do the cheap parts: ask "Have you used us before?", take the number as digits and read it back.

Test with five cases and write down what the software does:

1. The same person twice, with the number written "555 0142" and then "5550142". You want one client and two requests, or a warning.
2. A known customer, using their real email and a different spelling of their name.
3. A husband and wife who share a number and an address but use different first names.
4. A business enquiring once through the owner and once through the office manager.
5. Two different customers with the same name and different mobile numbers. You want two clients; one client holding both requests means the name search has merged strangers.

Whatever happens, schedule ten minutes a week to scan the 20 newest clients for pairs, and check whether your software has a merge tool before you need one.

Photos that arrive in the record, not just in the chat

For trades work a photo often settles whether a visit is needed. There are three ways to get it into the job record, each with a failure to test for.

• Through the job software's own form. Jobber's request form accepts up to 10 files under 10MB each, in JPEG, PNG or HEIC. Its help centre says photos can be attached to line items and carry forward when the request becomes a quote or job. HEIC matters because many phones save photos in that format.
• Through a separate upload form. Jotform's free Starter plan lists 100MB of upload space and 100 submissions a month, with 500 stored. If a typical phone photo is 3MB, that is about 33 photos, which two or three customers can use up. Its Bronze plan lists 1GB, at $39 a month or $408 a year. Check what your own phone's photos weigh.
• Through the chat widget. Some widgets let visitors send an image. If yours does, test whether the file reaches the Zap, or whether you only get a link in the transcript. A link in the note is fine, as long as the link doesn't expire.

Whichever you choose, tell the bot not to interpret the photo. It collects it for the installer, who is the person qualified to judge what the photo shows.

A security installer's chat-to-Jobber build, in one afternoon

A four-person security installer fitting alarms, cameras and access control makes a good test case. Its website chat gets about 35 conversations a month, and roughly 30 turn into real enquiries. Today the owner copies each chat summary from the inbox into Jobber, which takes about five minutes a time. The figures are illustrative, but the build sequence is a realistic one.

1. 20 minutes: write the menu (new install, fault, bigger quote, something else), the service-area list and the six questions.
2. 30 minutes: make a separate Jobber request form for the chatbot, with an image upload field and the service-area setting on. It acts as Route A for visitors who prefer a form.
3. 40 minutes: build the chat flow with the instruction shown earlier and a "send to Zapier" step at the end.
4. 45 minutes: build the Zap: filter, phone formatter, Find or Create Client, Create Request, Add Note to Request, alert to the owner's mobile.
5. 30 minutes: run the duplicate and photo tests.
6. 45 minutes: run the twelve test enquiries below.

That is about three and a half hours. Running costs: Zapier Professional at $19.99 a month on annual billing, plus whichever chatbot plan was already chosen. At four tasks a lead and 30 leads, the Zap uses about 120 tasks a month. The owner saves 30 times five minutes, or two and a half hours a month of retyping. The bigger gain is that a lead exists in Jobber the moment the chat ends, so a technician can be asked to ring back that afternoon rather than whenever the inbox is next cleared. The firm needs Jobber on Connect or above for this, since Jobber's Zapier page names Connect and Grow, and Plus includes every Grow feature.

A dozen test enquiries before you go live

Run these from your own phone, write down what happened in each, and don't switch the bot on until every one passes. Several are where a mismatch hides for months.

1. A full enquiry. Give all six fields and attach a JPEG. You should see one new client and one request, every field in the note and the photo reachable.
2. The same enquiry, number typed differently. Repeat the first test with spaces in the number. A pass is no second client, or a clear warning.
3. Outside the area. Give a postal code you don't cover. The bot should refuse politely, and nothing should be created.
4. Email only. Refuse to give a number. Either the request is created with the email, or the bot explains it needs one contact method.
5. A known customer. Use an existing customer's details. The enquiry should attach to the existing client.
6. An iPhone photo. Send a HEIC image. It should arrive, convert, or reach you by a working link.
7. A 12MB photo. The visitor gets a clear message, and no Zap run fails silently.
8. Awkward names. Use a surname containing an apostrophe and a first name containing an accented letter. The record should hold the exact characters.
9. Danger happening now. Type that a break-in is under way. The emergency-services message comes first, then an alert to you.
10. After hours. Submit at 11 pm. The record exists and the visitor is told when to expect a call.
11. Consent left blank. Don't agree to marketing. The consent flag stays off in the record.
12. Junk input. Paste gibberish and a web address. The request is blocked or flagged, not created.

Consent, disclosure and the first month's checks

Tell visitors they are talking to an automated assistant. That is good practice anyway, and chatbot-disclosure rules in some markets expect it. Disclosure wording for your chatbot gives sample lines. Say plainly what you do with the details, and with any photo of someone's home or business, which is more personal than a name. Ask a data-protection adviser what your notice must say.

For the first month, review every lead by hand. Look for fields in the wrong place, a Zap error, a job type the bot invented, and leads that sat unanswered. If more than a handful need correcting, tighten the instruction rather than adding tools. When the numbers settle, check once a week: new leads against chats started, any duplicates, and Zap failures. That ten-minute habit is what keeps "every enquiry becomes a lead record" true six months on. For deciding when a person should take over from the bot, when an AI chatbot should hand over to a human sets the rules.

Questions about wiring a chatbot into job software

Do I need a CRM, or is a spreadsheet enough?

A spreadsheet is enough to prove the idea. Send the six fields to a sheet for a month and see how many chats become real enquiries. Move to the job software once you are retyping rows into it, because the record you want, with the customer, the property and the later quote on one screen, only exists there. The same fields map across, so nothing you built is wasted.

Can the chatbot book the job straight into my calendar?

Usually not safely. A booking commits a technician, a van and a time, and a chatbot cannot see travel gaps or a job that overruns. Let it create the lead and offer the visitor a callback window, or send them to your software's own online booking page, which does know your availability. Keep the human decision on the visit itself.

What if my job software has no Zapier connection?

Fall back on the routes that need none. Use the software's own enquiry form with the chatbot handing over the link, or email the captured fields to an enquiry address if the software provides one. Either way you lose the automatic matching of existing customers, so schedule a short weekly check for duplicates and unassigned enquiries.

Should the chatbot collect marketing consent?

Only as a separate, optional question that defaults to no. Do not infer consent from someone asking for a quote. Jobber's own request form includes marketing-consent tick boxes beside the phone and email fields that cannot be removed, which shows how vendors treat it. Mirror that: record the answer in the lead, and never tick the box on the visitor's behalf.

Further reads

• AI Chatbot for Plumbers: Answer Questions, Collect Photos, Refuse Jobs — What a trade chatbot should answer, and when a person takes over.
• AI Customer Support Chatbot for Home Services: After-Job Questions — Extend the same bot to after-job questions.
• How Much Does a Website AI Chatbot Cost, and Will It Pay Off? — Work out whether a website chatbot pays for itself.
• AI for Self Storage: A Unit-Size Chatbot for "What Fits in a 10x10?" — How to build a self-storage unit-size chatbot: size guide, item-to-unit logic, sample conversation, lead capture and a test set.
• Roofing Chatbot Platforms With AI Compared: Qualifying Storm Enquiries — No roofing chatbot wins on its vendor page alone: the ten questions a storm bot must ask, nine platforms compared, and a ten-enquiry test to run first.
• Wix AI Builder Pricing: Which Plan a Small Trades Firm Needs — Light, Core, Business or Elite? Wix plan prices read on the day, what each gives a trades firm for forms, bookings and deposits, and what to skip.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Jobber pricing page and help centre (Request Basics, Requests and Bookings Settings, adding forms to a website), Jobber Zapier page, Zapier app pages for Jobber, ServiceM8 and Housecall Pro, Zapier help pages on task usage and error handling, ServiceM8 pricing page, Housecall Pro pricing page, Tradify and Fergus pricing pages, Tidio Zapier page, Chatbase pricing and actions pages, Botpress pricing page, HubSpot chatbot builder page, Jotform pricing page. Pages read October 2026.