Starter trial and plans: pick the right level

Make the most of your 14-day Fynd Starter trial

An AI product advisor can only make useful recommendations when it understands your assortment, shopper questions, and commercial goals. That is why a Fynd Starter trial is more than a quick product demo. Treat the first 14 days as a focused implementation and learning period: add the embed widget, make your catalogue searchable through catalog RAG, configure your bots with the right context, and learn from real shopper questions.

This does not need to become a large IT project. The trial is a practical way to establish whether Fynd helps shoppers find suitable products with less effort. At the end of it, you can choose a subscription based on demonstrated value and actual use, rather than selecting the longest feature list.

What the trial should prove

Start with a single, clear hypothesis. Perhaps shoppers who are unsure about variants will get an answer faster. Perhaps people will find a suitable product without browsing an entire category. Or perhaps your team will learn which questions are missing from product and category pages. A good trial does not answer the abstract question of whether AI is interesting. It answers whether your shop gets less search friction and more relevant product interaction.

Assign an owner before you begin. That can be someone in ecommerce, content, customer support, or merchandising. They do not need to build the widget themselves, but they should keep the work moving: review the catalogue, test bots, check the funnel each week, and follow up on unanswered questions. Without an owner, a trial can stop at “the widget is live,” which misses the learning loop that makes it useful.

Decide which pages you want to improve first. A wide catalogue is fine, but begin with commercial intent. Consider a category with many choices, products with detailed specifications, a page where visitors often leave, or a campaign landing page. This lets you later interpret activity in the context of why visitors reached that page.

Days 1 and 2: embed the widget and validate the basics

Your first task is adding the Fynd embed widget to your storefront. Treat it like any other conversion tool: review desktop and mobile behaviour, different browsers, and every page where it should appear. Do not only check that the icon is visible. Open a conversation, send a message, and make sure the widget does not cover key navigation, consent controls, or checkout elements.

Set up a simple measurement check at the same time. Fynd uses the funnel open → message → product click → cart → purchase. An open is the start of an interaction; a message shows that a visitor actually wants help; a product click means a recommendation was relevant enough to explore. The last two steps connect advice to shopping behaviour. You do not need conclusions in the first two days, but you do need to know the events arrive correctly.

Use this compact implementation checklist:

• Show the widget on agreed product, category, and campaign pages.
• Test one mobile session and one desktop session.
• Send one broad question and one specific buying question.
• Open a product from a bot response.
• Add that product to the cart and confirm the journey appears in analytics.
• Identify internal test conversations so they do not later look like shopper demand.

Keep the rollout deliberately small. Do not enable every bot, prompt, and page at once. A working, measurable foundation is more valuable than a broad deployment nobody can evaluate.

Days 3 to 5: make catalog RAG useful

Fynd uses catalog RAG to make product information available to a bot. In this context, RAG means the bot retrieves relevant information from your catalogue before answering. That distinction matters: a product advisor should not guess about size, compatibility, or product properties from generic marketing copy.

First, check whether your core product data can answer shopper questions. Clear titles, descriptions, prices, stock status, categories, and variant data are typically more useful than lengthy promotional copy. For technical, fashion, or bundled products, attributes such as material, dimensions, fit, audience, use case, and compatibility are especially valuable.

Build a list of twenty real pre-purchase questions. Source them from customer support, on-site searches, reviews, or conversations with your store team. Include simple questions as well as questions that require comparison or nuance:

• “Which backpack will fit as carry-on luggage?”
• “Is this suitable for sensitive skin?”
• “What is the difference between these two models?”
• “Will this accessory work with my existing product?”
• “Which size should I choose for a wider foot?”

Test each question as if you were a shopper. Check three things: does the bot retrieve relevant products, make an understandable recommendation, and admit when information is missing? An answer that hides uncertainty is less useful than one that asks a follow-up question or states that the catalogue cannot confirm the detail.

If answers are too general, a longer prompt is not automatically the solution. Look at the source first. Is a required attribute absent from the catalogue? Is a description ambiguous? Are variants not described clearly? Catalog RAG works best when your product data would also let a human answer the question reliably.

Days 6 and 7: configure bots around purchase intent

One bot does not need to handle every visitor in the same way. Configure bots around the purchase intents that matter most in your shop. A category advisor can help shoppers choose from a large selection. A comparison bot can clarify differences. A product-page bot can answer fit, use, or compatibility questions. You can also create a bot for a limited campaign, as long as its knowledge and purpose remain focused.

Give every bot a clear job, tone of voice, and boundary. The job might be “help visitors find a suitable product.” The boundary matters just as much: do not invent stock availability, make delivery promises that are not in your data, or offer technical advice beyond the available knowledge. Add custom knowledge where it helps, for example for returns, care instructions, size guides, or product bundles.

Test more than happy paths. Ask about a product that does not exist, an ambiguous comparison, and a question outside the catalogue. You want the bot to remain helpful without pretending it knows everything. Use the results to improve bot instructions and knowledge, not to teach it a rigid script.

Week two: learn from real conversations

From day 8 onward, observation becomes more important than constant configuration changes. Review conversations briefly each day and schedule one more thorough review every week. Pay special attention to questions without a good answer, questions that do not produce a product click, and questions that recur. Fynd surfaces these signals as unanswered questions. They are not failures; they are a backlog built directly from shopper intent.

Turn that backlog into a fix loop:

1. Group similar unanswered questions.
2. Decide whether the fix belongs in catalogue data, custom knowledge, bot instructions, or a better product page.
3. Add only verified information.
4. Re-test with the original question and a variation.
5. In the next week, see whether the question is unanswered less often and produces more product clicks.

For example, “is this dishwasher safe?” may need a product attribute. “How can I combine this with my existing set?” may need a short guide in custom knowledge. When visitors ask to be contacted or submit their email address, treat that as a lead and ensure your follow-up process is explicit. Fynd can make leads visible, but value appears only when someone owns a timely, relevant response.

Which funnel metrics matter during the trial?

Do not focus only on total widget opens. An open may indicate curiosity, but it may also be an accidental click. The useful signal is movement through the funnel:

• Open: how many visitors begin a widget interaction.
• Message: how many opens become a real question.
• Product click: how many conversations send someone to a product detail page or product link.
• Cart: how many of those product interactions result in an add-to-cart action.
• Purchase: how many eventually result in an order.

Compare steps you can influence. Many opens but few messages can suggest an unclear welcome message or poor placement on the page. Many messages but few product clicks can mean answers are too general, catalogue details are missing, or the bot is not making a confident recommendation. Product clicks without carts are not automatically bad: shoppers may be comparing items, checking availability, or returning later. Look for patterns over several days instead of judging a single session.

Choose a plan after day 14 from actual use

After the Starter trial, choose Starter, Growth, or Pro from your Fynd dashboard. Do not begin by asking which plan is “best.” Choose the plan that fits your current traffic, bot count, catalogue complexity, desired analytics, and working model.

Starter is often appropriate when you are keeping one or a few focused use cases live and establishing a reliable rhythm for catalogue reviews and unanswered questions. Growth becomes more appropriate when several teams, bots, pages, or optimisation efforts need attention at once and you use the funnel routinely to steer decisions. Pro is for when product advice becomes a broader, business-critical programme, with greater scale, governance, integrations, or support appropriate to your operation. Always check the current limits and options in the dashboard; those are the source of truth for your account.

Think of limits as a design prompt, not merely a ceiling. If you approach a usage threshold, first ask which conversations, bots, and pages demonstrably create value. Remove outdated configurations, merge overlapping knowledge, and focus on the intents with the strongest commercial relevance. Upgrade when growth in use, teams, or proven use cases is sustained, not simply to reserve capacity in advance.

Make a decision without guessing

Close the trial with a short review. Which questions does Fynd solve well? Where is catalogue context still weak? Which bots generated relevant product clicks? Which unanswered questions become concrete content or data work? Who will maintain the fix loop after the trial?

A successful 14-day trial does not have to prove that every visitor buys immediately. It should prove that you can improve product advice in measurable ways: the widget is correctly placed, catalog RAG uses trustworthy information, bots help with real choices, analytics shows the route toward purchase, and leads and missing knowledge have an owner. With that foundation, selecting Starter, Growth, or Pro in your dashboard becomes an operational decision rather than a leap into the unknown.