Test Fin's answers before going live

Before running batch tests, make sure you have the following in place:

• Content added to Fin: You should have content sources connected so Fin can answer customer questions.
• Guidance added to Fin: You should have tone, behavior, and escalation guidance configured.

Learn how to set these up in the Launch Fin course.

When we're launching Fin, we think of ourselves as the chef. We've gathered our ingredients, our content, our guidance, and our escalation rules, but we'd never serve a dish without tasting it first. And that's what batch testing is. It's the taste test for our AI. It lets us simulate dozens of real questions at once so we can pinpoint exactly where the flavor is off. If an answer is half baked or missing a key ingredient, we can tweak our content or guidance until the recipe is perfect. So batch testing works like this. First, we provide a batch of questions, then we review Fin's responses, grading them good, bad, or just okay. This helps us identify what needs to be fixed, and we continue this loop until we're ready to go live. So this enables us to validate and test Fin's answers thoroughly so we can spot the areas that need improvement and go live with confidence. Okay. Let's jump into the product. So let's imagine that we've uploaded a CSV of our top frequently asked questions from the last thirty days. Now our job here is to go through Fin's answers and rate each one. What we're looking for here are answers that need to be improved, and it's much better to find issues at this stage before customers run into them in the wild. So our goal is simple. Get everything to go green before we go live. Okay. So over on the right, we can see Fin's actual answer, and we can see the content and the guidance that Fin used to generate it. Now let's just pretend that this answer doesn't meet our expectations, so we'll mark it as poor. Now this gives us a list of reasons to choose from, like maybe the tone wasn't right or maybe Finn used the wrong content. And based on what we select, the UI actually points us towards the fix. So when we hit improve answer, we actually get a list of the exact content sources used so we can easily update them or create new ones. Oh, and I almost forgot a feature that I really appreciate is the notes feature. You can leave yourself a clear reminder of what needs to be fixed or add notes for someone else to pick up. Now, this actually works really well, especially if your workflow is to review stuff first and then fix it later. Okay. One last thing to flag here. Now, by default, we're testing as a preview user. This gives us a broad general view of Fin's answers, but we can get much more specific by testing as a particular audience or even as a specific user or lead. For example, let's say that our shipping policies differ between the US and the UK, and we want to be one hundred percent sure that Fin is encoding a US delivery time to a customer in London. So by filtering for our UK audience, we can actually verify that Fin is pulling from the correct regional help article and using the right currency in its answer. This gives us the confidence that Fin is going to behave as we expect. Okay, so finishing your batch test means you've moved past the prep phase. You've tasted the dish, refined your ingredients and ensured everything is balanced. You aren't just crossing your fingers. You're launching a recipe that you verified yourself. And now that you have full confidence in the recipe, you're ready to open the doors and serve into your customers.