UAT test site · your organization
01Overview

What this page is for

This is a hands-on practice space for your team to try your Chat AI assistant, exactly as your members will, before it goes live on your own website.

Ask real questions, read the answers, open the links Chat AI provides, and tell your Interface AI contact about anything that looks off. Nothing you do here affects your production website or any member account.

Everything happens inside the chat bubble in the bottom-right corner. This surrounding page is a plain backdrop so the assistant has somewhere to sit while you test.

02Scope

What to test, and what to skip

This round covers two things. Mostly it is a knowledge test: whether the assistant gives correct, well-sourced answers to questions answerable from your public website content. Alongside that you can try a preview of account actions, which run against the interface.ai test core using a mock member, never your core and never real member data.

In scope: ask about this
  • Rates, fees, and product details published on your website
  • How to open an account or apply for a loan (the information, not the transaction)
  • Branch and ATM locations, hours, and contact details
  • Ambiguous questions ("best rate"): it should ask you to clarify
  • The everyday "how do I" questions your members ask
  • Transactional queries preview (on the interface.ai test core; mock member credentials are in your Sandbox Testing Handbook, chapter 08): Check account balance, Transfer money, Checking loan due amount, and Check transaction history.
Out of scope: skip for now
  • Asking to be transferred to a live agent or staff member
  • Any transactional queries apart from the in-scope ones.
  • Anything the assistant would need a live system or API lookup to answer (for example an ATM or branch locator that queries a locations service, or real-time rate feeds, or scheduling or booking)
Why skip those: Live-agent handoffs and anything that requires a live system or API lookup connect to other systems not part of this test, so they are tested separately. Rule of thumb: if the answer is published on your website, it is in scope; preview transactional operations (listed above) are also in scope, and these are available on interface.ai's test core; if the assistant would have to call a separate live system or API to get it (even for locations, hours, or rates), it is out of scope. If you ask for those here, the assistant may not respond the way it will in production, and that is expected, not a defect. If you would like to try out these features, please reach out to your Interface AI contact and we will enable them for you.
03Start here

What to test first

Start with these; they show the assistant at its best. Use your own words too.

  • What are your checking account features?
  • What are your current auto loan rates, and what do I need to apply?
  • How do I open an account?
  • Can you just give me a better rate than what's posted?
  • What is your best rate?

Also try the account actions:

  • Check the balance of my savings account
  • Transfer $20 from my checking account to my savings account
  • Tell me my recent transactions
04How to use it

Using the assistant

  1. Open it

    Tap the round chat bubble in the bottom-right corner. The assistant opens with a short greeting.

  2. Ask a question in your own words

    Type a question in the box at the bottom and send it. Phrase it the way a member really would. A short typing indicator appears while the assistant prepares its answer.

  3. Ask follow-up questions

    Keep the conversation going. The assistant remembers what you already asked, so you can say things like "what about a used car?" without repeating yourself.

  4. Use the reference links

    Answers often include a reference link to the page or document the answer came from. Open it to check the source. See section 06 for how links behave here versus on your live site.

  5. Open the options menu

    The three-dot menu in the top corner of the chat window holds extra options, such as clearing the chat and leaving feedback.

  6. Start a new conversation

    To test a fresh question with a clean slate: open the three-dot options menu in the top corner of the chat window, then choose "Clear chat" (this wipes the current conversation and starts a new one). If you only want to step away for a moment, minimize the window instead by tapping the minimize or close control on the chat header; your conversation is kept and reopens where you left off, whereas clearing the chat starts over from scratch.

05Interpreting

Reading the answers

A few parts of each answer tell you where the information came from and how much to rely on it.

Reference links and citations
Point to the exact web page or document the answer is based on. They are the assistant showing its work, so you can confirm the answer against the real source.
Suggested-question buttons
Tappable shortcuts for common questions and sensible next steps, for example "Compare options" or "Branch hours". Tapping one simply sends that question for you.
Where sources appear
The assistant answers first, in plain language, then shows the link or document it used. When it opens a document, it jumps to the relevant part and highlights it so you do not have to search.
When it is unsure
If the assistant does not have a good source, it should say so and offer to help another way rather than guess. That is correct behavior. Note the question so the knowledge base can be improved.
07Upcoming Improvements

Known areas we're improving

A few areas are still getting polish and we are actively improving them during your evaluation. Test them freely; they are already on our list. Spot something beyond this list? Log it, that is what the sandbox is for.

  • Conversational warmth: answers are accurate but may not acknowledge your context first; we are tuning tone and empathy.
  • Guardrails can be over-cautious: some valid questions (for example "is my money insured?") may get a "sorry, I can't help with that"; we are tuning the balance.
  • Answer length: the key point can sometimes be buried among related detail; we are tightening responses.
  • Clarifying questions: the assistant sometimes answers with more detail than needed instead of asking which product you mean.
  • Response styling: rate tables and interactive elements are getting formatting polish.
  • Reference links and citations: link opening and passage highlighting are being refined.
  • Branch locations: moving to progressive disclosure (a short summary first, then details on request).
  • Fine-tuning of transactional experiences, like supporting special filter queries for transaction history on date, transaction type, or so.
08Getting the most from it

Tips and reporting

  • Ask real member questions, phrased naturally, not keywords.
  • Try a few wordings of the same question to see if answers stay consistent.
  • Include the everyday and the tricky questions your members really ask.
  • Open the reference links to confirm each answer against its source.
  • Test on both a computer and a phone, since the layout can differ.
  • Note anything wrong: a wrong answer, a missing or broken link, or poor formatting.

How to report an issue

When something is not right, send it to your Interface AI contact. We will fix it and make it available for you to test again the next day. The full ticket-logging steps are in your handover deck and in chapter 06 of your Sandbox Testing Handbook. To help us reproduce it quickly, please include:

  1. The exact question you asked (copy and paste it).
  2. What the assistant answered, and what you expected instead.
  3. A screenshot if you can, plus whether you were on a computer or a phone.
  4. The conversation ID from Assemble (see below), so we can trace exactly what happened.

Finding the conversation ID: sign in to Assemble, open the Conversations area (your conversation history), find your test conversation by date and time, open it, and copy the conversation ID shown there. Your Interface AI contact can point you to the Conversations area and the full steps in the user guide.