Expert InsightAI & Business Automation
From Website Enquiry to Approved Quote: A Practical AI Workflow
Capture the enquiry, organise the lead and prepare the next response automatically — with a person approving every customer-facing message and quote before it leaves.
Turn an enquiry into organised work
A website form is only the beginning of a sales conversation. Someone still needs to read it, create or update a customer record, understand the request, gather missing details and prepare a useful response. When that work depends on copying information between inboxes and spreadsheets, good enquiries can wait longer than they should.
AI and workflow automation can take much of that preparation off your team’s desk. The useful distinction is between preparing the work and authorising contact. In the approach we would implement, every customer-facing reply, follow-up and quotation waits for a named person to approve it.
The full journey
Preparation first. Approval before contact.
- Capture → organiseWebsite enquiry → CRM lead
- Interpret → prepareAI summary → questions → quote draft
- Review → sendA named person approves the exact content.
Capture once, keep the original
The website would submit a validated enquiry to a workflow with a unique reference. Ordinary automation checks required fields, records the source and time, and handles repeated submissions without silently creating multiple leads.
The CRM step looks for an existing contact before creating a new one. It then associates the enquiry with the appropriate record and assigns an owner. CRM APIs make this practical: HubSpot, for example, supports contact creation and searches by properties such as email. The chosen CRM and account permissions determine the exact implementation. HubSpot contact API overview
The original message stays attached to the lead. A short AI summary helps someone scan the queue, but it must not become the only record of what the customer actually requested.
One reliable record
Keep the original connected
- Original enquiryMessage, source and timestamp
- Contact matchFind an existing record before creating one.
- Lead + ownerAssign responsibility and a next action.
Let AI prepare the next useful step
The AI stage can identify the requested service, extract details the customer supplied and suggest questions for missing information. For a network enquiry, that might mean the property layout, troublesome rooms, existing broadband equipment and whether cable routes are available.
Unknown information should remain unknown. A model should not invent a budget, assume an installation date or promise that a particular package will suit a building it has never assessed. Enquiry text and attachments should be treated as customer content, never as instructions giving the system extra permissions.
Next, the workflow prepares a draft reply using approved service information and your preferred tone. It can also create an internal reminder, so an enquiry awaiting review remains visible rather than disappearing into an inbox.
A clear division of work
AI interprets. Rules control.
- AI can prepareSummaries, categories and draft wording.
- Rules must controlPermissions, prices and calculations.
- A person decidesScope, commitments and customer contact.
Build quotes from approved facts
A useful quote draft combines confirmed customer requirements, an approved price list and agreed commercial terms. Calculations belong in deterministic rules or quoting software. AI can help explain scope clearly; it should not invent prices, discounts, tax treatment or contractual commitments.
If an enquiry falls outside the standard service, route it to a person for scoping. The draft should show assumptions and exclusions clearly and flag where a site visit or supplier confirmation is required. A polished document is not proof that the underlying recommendation is correct.
Quote preparation
Facts in. A reviewable draft out.
- Confirmed requirementsScope, equipment and site constraints
- Approved price rulesRates, quantities and commercial terms
- Draft quotationFlag assumptions; leave unknowns unresolved.
Approval before customer contact
The reviewer should see the recipient, original request, proposed message, quote version and any unresolved assumptions together. They can approve, edit or reject. Editing a price, attachment or recipient means the revised version needs approval again.
This should be enforced by the workflow, not left as a sentence in an AI prompt. n8n supports pausing selected AI tool calls for human review, including tools that send messages or change records. A comparable approval gate can protect a conventional workflow’s send step. n8n human approval documentation
Once approved, automation sends the exact reviewed content and records the result in the CRM. Follow-ups use the same approval rule. A customer reply, closed opportunity or stop request should prevent an outdated follow-up from being sent.
Start small and measure the result
We would begin with one enquiry route and a draft-only pilot using representative examples. Check duplicate handling, missing information, incorrect classifications, delivery failures and the approval process before enabling sending. Measure preparation time, review corrections and missed enquiries rather than assuming a headline percentage saving.
Personal data needs deliberate handling throughout. Limit what reaches the AI provider, choose appropriate access and retention settings, and assess the provider’s processing arrangements. Ireland’s Data Protection Commission highlights accuracy, retention and third-party processing as issues organisations must understand before using AI with personal data. DPC guidance
MBS Integrations can map your existing enquiry process and design a workflow that prepares the administration while keeping customer commitments under your control.
Start with a controlled pilot
Prove the workflow before scaling
- Draft-only trialOne enquiry route and representative examples.
- Test the difficult casesDuplicates, missing facts and failed delivery.
- Measure and improveReview corrections, preparation time and missed leads.
Sources & Further Reading
Technical references checked 17 October 2026. Features and device support can change.
Planning something similar?
Tell us what you already have and what you would like to improve. We will help plan the next practical step.
Discuss Your Project