What this guide is, and is not
An AI quote bot is the highest-leverage capture surface on a service-business website. Done well, it replaces a generic contact form, beats a $15k/year SaaS chat tool on attribution and voice, and converts a high-intent visitor into a structured lead inside the CRM without anyone touching it. Done poorly, it produces unqualified noise that a sales team has to clean up by hand.
This guide tells you what good looks like. It is not a build playbook. The actual question-tree configuration, the CRM field map, the attribution pipeline, and the welcome-email logic that Gibson runs are proprietary. You will finish this guide knowing how to evaluate any quote-bot build, including the one we might quote you for. You will not finish with a clone.
We use our own
The bot in the bottom-right of this site is the same engine we sell. The lead routes to our Zoho with first-touch and last-touch attribution, fires a welcome email to the prospect, and pings Albert directly. If you have submitted a quote from this site, the same engine wrote it up.
That is the proof point: we did not vendor this from a SaaS. We built it, run it, and feed our own pipeline with it.
The six questions to ask any quote-bot provider
Before signing, ask each of these. The answers separate a real quote bot from a SaaS chat widget with a custom skin.
- +Where does the lead actually land? A real bot writes directly to your CRM via API, with service-specific custom fields. If the answer involves Zapier, Make, or a shared inbox, you are buying a worse version of a contact form.
- +What attribution is captured? First-touch source, last-touch source, gclid, fbclid, msclkid, landing page, current page, device. If the provider can only show you 'referrer URL', the attribution layer is missing.
- +Does the conversation branch by service? Call tracking should ask about Google Ads and tracked numbers. Letterbox should ask about suburbs and quantity. SMS reactivation should ask about list size and last contact. A single linear funnel for every visitor is the SaaS pattern. It loses leads.
- +What happens before the submit lands? Cloudflare Turnstile or equivalent at the API layer. Rate limiting per IP. Field validation server-side. 'Honeypot field' is not anti-bot, it is theatre.
- +What fires on submit? A real engine writes the CRM lead, sends the prospect a welcome email, and notifies the owner in real time. If only one of those fires, you are losing either the prospect's trust or the owner's response window.
- +Where does the voice come from? A bot should sound like the business. Ask the provider how the voice is configured: prompts, message library, branded copy. If the answer is 'we have a default voice', the bot will sound like every other site they have shipped.
Where most setups quietly fail
After running this category for a while, the failure modes are predictable.
- +SaaS dependency. The bot is a Drift/Intercom/Tidio iframe. When the SaaS vendor raises prices, you pay. When they change their UI, your site changes. When they go down, your lead capture goes down.
- +Generic voice. Default question prompts. 'How can I help you today?' opening lines. Visitors recognise the pattern from twenty other sites and engage at half the rate.
- +No service branching. Same five questions for every visitor regardless of intent. Call tracking enquiries and letterbox enquiries land identical, qualification has to happen on the phone.
- +Lost attribution. The chat widget runs in an iframe and breaks the UTM trail. The lead lands in the CRM as 'source: chat' with no further context.
- +Email-only handoff. No real-time push to Slack or Teams. The owner discovers the lead 20 minutes later when they refresh their inbox. The prospect has already started their next search.
- +No welcome email. The bot says 'we will be in touch' and then nothing happens until a human gets to it. The prospect cools off in the gap.
- +No rate limiting. The endpoint is unprotected. Bots and competitors burn through your form thousands of times a day. Real leads get lost in the noise.
The compliance frame, non-negotiable
Australia. Spam Act 2003. Privacy Act 1988. Plus the global expectation that visitors know what is happening with their data.
- +Consent in context. The bot collects name, email, phone, and business context. The visitor needs to see a clear privacy line before they submit. Not a hidden link — visible above or under the submit button.
- +Welcome email opt-out. The auto-responder must have a working unsubscribe path. Not just legal; it stops the welcome email from triggering future complaints.
- +Data minimisation. Only ask for what the quote actually needs. DOB, exact address, financials should not be in the quote bot. Push those to a secure onboarding form post-quote.
- +Audit trail. Every submit should be logged with timestamp, IP, Turnstile token, and validation outcome. If a lead disputes the contact, you need to be able to prove how their data arrived.
What you should refuse to sign off without
If you are commissioning a quote-bot build (us or anyone else), these are the deliverables that protect you.
- +Direct API writeback to your CRM, with documented field map per service line. Not Zapier.
- +First-touch and last-touch attribution attached to every lead, with the field list visible inside the CRM record.
- +Service-specific question branching, with the question tree exportable as a single config file you can read.
- +Cloudflare Turnstile or equivalent at the API layer, with rate limiting per IP documented.
- +Three things fire on submit: CRM lead, welcome email to prospect, owner alert in real time. All three logged.
- +A monthly evidence pack: submits received, validations passed, leads landed in CRM, average time-to-owner-alert, opt-outs.
Where the value flows
A well-built quote bot changes more than the contact-form conversion number.
- +Sales: every first call starts qualified. The owner reads the brief before they pick up.
- +Marketing: attribution is intact from click to deal. Campaign ROI becomes a real number, not an estimate.
- +Operations: no SaaS subscription, no contract renewal cycle, no vendor lock-in. The bot is part of the site.
- +Customer engagement: the welcome email lands within seconds. The visitor leaves the site already feeling responded to.
Want us to build it for you?
Gibson ships this build pattern as a 48-hour working prototype. Your site, your CRM, your voice, your service tree. From $1,000 for the prototype. Scoped engagement after.
Book a 30-min diagnose call from the Sandbox page.