From business knowledge to a useful customer conversation
A practical guide to preparing and testing V7 Agents for your business website. The examples below are fictional demonstrations, not customer testimonials or measured results.
For online retailers, service businesses and teams managing multiple branches.
Start with facts your team can maintain
Prepare your current products or services, prices, FAQs, delivery rules, opening hours and contact details. Include conditions and exceptions: where delivery is available, when a quote needs review, and which questions your team must handle.
Use the business workspace to maintain this information. Keep private credentials and unnecessary customer details out of the knowledge you supply. Choose a conversation goal your team can fulfil, such as helping a visitor compare products or requesting a callback.
Connect public knowledge and useful next steps
If useful, import up to 20 public pages from the HTTPS website saved in your business profile. Check the excerpts before relying on them and refresh the import when those pages change. This supplements your managed catalogue, FAQs and policies; it does not continuously synchronise the whole website or connect private order records.
Add approved destinations for product details, checkout, booking or support, with a clear label and purpose for each. Enable only the quote, callback or consultation request forms your team can fulfil. Test where each link opens and what a successful submission looks like. A suggested link or an opened form is not a completed purchase, submitted request or confirmed appointment.
Retail demonstration: narrow the choice
Fictional catalogue: a compact desk lamp costs £35 and has adjustable warm light. A customer asks, ‘Which lamp fits a small desk?’ A useful response explains the known size and lighting features, then asks about the customer's needs before recommending it.
Test ‘Can you deliver tomorrow?’ separately. If your policy does not confirm a delivery date, the assistant should explain the published delivery rules and offer the configured contact path. A product recommendation must not invent stock availability or promise an unverified arrival date.
Continue the fictional conversation with ‘My budget is now £25’ and then a question about returns. Check that a cheaper option is suggested only if your catalogue contains one, the changed budget is respected, and the returns answer uses your policy rather than restarting the product list. If the visitor chooses an item, check the approved product or checkout link before asking them to complete the purchase on your website.
Service demonstration: qualify an enquiry
Fictional service: a design studio offers website consultations, with final prices provided after reviewing scope. A visitor asks, ‘How much is a new website?’ A useful conversation asks about the work needed and explains the quote process instead of inventing a fixed price.
Configure the details needed for follow-up and explain what happens next. A request for a consultation is not a confirmed booking unless a working booking integration actually confirms it. Test the handoff using a contact route your team monitors.
Multi-branch demonstration: identify the location
Fictional business: one branch opens on Sundays and another does not. When a visitor asks about Sunday opening, the assistant should identify the branch before answering from its supplied hours.
Check ambiguous location names, temporary opening changes and enquiries outside the supplied service area. Maintain branch-specific facts so a visitor receives the policy for the location they need.
Test before embedding on your website
Use the widget section to test a known fact, missing information, an out-of-scope request and a request needing your team. Compare answers with your source material. Try follow-up questions, a changed product choice and a declined callback. Check the welcome message, brand colours, mobile layout and contact links. Select a supported model and check the prepaid AI credit balance when testing model-powered replies; saved-knowledge fallback is available when paid processing cannot run.
The workspace provides generated embed instructions and guidance; optional one-time assistance is a paid add-on. Your team or website developer installs the widget; website editing and custom development are separate. Configure the approved website origin. Test on the actual website before launch. Knowledge configuration does not automatically connect payments, inventory or booking systems. WhatsApp and call handling are included in the Scale plan and need their own provider setup. Phone numbers are not included; number rental and WhatsApp provider charges are separate and are not covered by prepaid AI credit.
Review actual conversations after launch
Review available conversation and enquiry records in your authorised workspace. Look for unanswered questions, unclear next steps and outdated facts, then update the relevant business information and retest.
Start with what you can observe. These demonstrations are acceptance criteria, not evidence of increased sales. Publish customer stories or reviews only when they exist and you have permission to use them.
What to check before launch
AI responses can be incorrect. Arrange authorised workspace access and a working provider configuration before launch. The examples are fictional and do not represent existing customers, testimonials or guaranteed outcomes.