Start with one useful job
Choose a question visitors ask before contacting your team. For a service business, that might be what the service includes or how to request a consultation. For a shop, it might be your return policy or delivery area. Give the assistant a reliable answer and a clear next step before adding more topics.
Free access is most useful when it helps evaluate your own customer journey. A generic demo can show the controls, but your knowledge determines whether answers fit your business. Prepare a short checklist of questions and check replies in the workspace tester.
What no-code setup means in practice
The dashboard handles knowledge, greetings, colours, placement, and contact options. You do not implement a chat interface or put the AI provider key in your page source. Your website still needs to permit the widget JavaScript, and someone with website access must install the snippet.
Approve exact domains rather than wildcards, publish the code once, and test a live page. Ask a question, submit a test enquiry with your own details, and confirm the record in your dashboard. Check a mobile screen so chat does not obstruct a menu or checkout control.
Understand free limits before publishing
The current Free plan appears on the pricing page. Its allowances cover conversations, AI messages, leads, knowledge resources, widgets, and other plan-dependent capacities. These limits let you evaluate the product; free does not mean unlimited traffic or unrestricted data storage.
Watch usage while testing. One conversation may contain several messages, and imported pages or documents consume their own resource allowance. A larger plan can provide capacity, but it does not make incorrect source information accurate or guarantee a higher conversion rate.
Capture an enquiry your team can act on
A useful lead contains a contact method and enough context to respond. Ask for the information needed for the next step rather than every possible field immediately. The saved conversation helps your team understand what the visitor meant without repeating the exchange.
Appointment requests are an option when a visitor needs a discussion. Your team must review and confirm the request. Keep the process honest about availability and do not train the assistant to promise approvals, prices, or time slots that require a person.
Keep human routes and privacy choices visible
AI can misunderstand a question or lack a relevant policy. Provide an easy contact option and review replies before going live. Team members can respond from the dashboard when a visitor needs a person. Build a follow-up routine before expecting automation to improve your sales process.
Optional visitor analytics requires consent. Visitors who decline can still use chat. Reports therefore do not count every visitor, and missing location information remains Unknown. Review the privacy notice before using real customer information in testing.
Decide whether to continue with evidence
After your trial, look for specific improvements: were common questions answered, did saved enquiries contain useful context, and could your team follow up? Record unclear replies and missed requests. Honest observations are more valuable than assuming every chat becomes a sale.
Continue on the Free plan while it fits your allowance and process. Compare paid plans when you need verified capacity or additional plan-dependent features. Native WhatsApp automation and unconfirmed CRM connectors should not be assumed to appear after an upgrade.
A first-week evaluation you can actually complete
Before installation, write five questions customers regularly ask and the answers your team would give. Include one question about an exception, such as a service area or a policy limit. Add the relevant knowledge, then test all five questions without sharing sensitive customer details. Note the specific correction needed when an answer is unclear.
On the live website, repeat the trial on desktop and mobile. Open and close the chat, navigate to another page, and check that the widget is installed only once. Submit a test enquiry using your own details. Confirm that the lead and conversation are available to the person responsible for follow-up.
Then ask someone unfamiliar with the setup to describe what they think the chat can do. If the greeting implies a confirmed booking or a guaranteed quotation, revise it. Useful automation begins with accurate expectations. A visitor should understand when the business will need to review a request.
At the end of the trial, decide whether the product fits the job you selected. Keep a record of useful replies and remaining gaps. If you need a capability that is not present, ask the agency before paying. The aim is to make one visitor journey work reliably, not to collect the largest possible list of enabled options.
Prepare knowledge without importing everything
Choose the pages that answer customer decisions, not every page your crawler can reach. A homepage may describe the offer while a policy page explains the conditions. Check both before adding knowledge. Navigation text, repeated footers, expired promotions, and staff-only documents can obscure the information the assistant needs.
List the facts you want to test and where each is maintained. If a service price appears in several sources, update them together or keep one approved reference. When your team cannot agree on a policy, resolve that question before asking the assistant to answer it. AI should not determine your business terms from conflicting material.
Use the tester for a normal question, an exception, and a request outside your offer. Keep the approved reference answer beside each test so reviewers assess facts rather than tone alone. A polite reply that changes a return condition is still a problem. Correct the source and rerun the affected questions.
Do not use private customer transcripts as test material without a proper reason and authority. Synthetic enquiries can reveal most setup and knowledge problems without involving another person. Use your own contact details for form tests and identify those records as tests when reviewing lead outcomes. This keeps your evaluation practical and prevents demo data from being presented as customer performance.