Donkey

AI Tool Vendor Evaluation Questions

Walk into your AI vendor demo with sharp, tailored questions that expose weak answers before you sign.

What you get: A tailored question list for evaluating a specific AI tool or vendor. It is grouped by topic, and each question explains why it matters and what good and red-flag answers sound like. It also includes a top-10 must-ask list, a scoring sheet, and a short follow-up message you can paste to the vendor. Price: $19 flat. Time: usually ready in minutes. Done by AI; you review it, with one free change.

We're not taking orders yet. Join the waitlist and we'll email you once when we open.

Real example

This is an actual result our AI made in a test run, from sample details. Yours is made from your details.

Show the example

Vendor Evaluation Questions: Intercom Fin AI Agent for Outdoor Gear E-commerce Support

Summary

This list is for evaluating Intercom's Fin AI Agent. It would answer first-line customer questions in your website chat and by email, and pass conversations to your 4 human agents when it's unsure. You're a 25-person outdoor gear retailer in Colorado with some EU customers, and Fin would see customer names, emails, shipping addresses, order history and return requests. Your three biggest risks are:

  1. Wrong policy answers. Fin could state return windows, conditions or exceptions incorrectly. That can cost you money and customer trust.
  2. Costs you can't predict. You'd pay per outcome, so a holiday surge could drive up the bill, and it matters exactly what counts as a billable outcome.
  3. Lock-in and data control. You need to know how Fin connects to Shopify, how it handles EU customer data, and whether you can leave cleanly with your conversations, help content and settings.

What the research found (limited): I was cut off after one search, so I only confirmed pricing. According to Intercom's public site (as of research):

  • Intercom offers three plans: Essential ($29/seat/month), Advanced ($85/seat/month) and Expert ($132/seat/month), all billed annually.
  • Fin is billed separately at $0.99 per "outcome," counted when Fin "successfully delivers value in a conversation."
  • Intercom also mentions spend caps and points to fin.ai/pricing for full pricing.

I could not confirm anything about Intercom's security certifications, where EU data is stored, or how the Shopify integration works. The questions below ask the vendor to prove those points. Some third-party blogs make claims I couldn't verify, so they are phrased as questions:

  • Fin counts an "assumed resolution" when a customer goes quiet.
  • Intercom has had a recent change of ownership.

If you'd like any of these items researched further, you can use your one free change request.

Assumptions

  • You use Shopify as your store platform and want Fin to look up live order and return status, not just answer general FAQs.
  • You don't have an in-house IT or legal team, so evidence and plain-language answers matter more than technical depth.
  • You're in the US, so US and Colorado state privacy laws may apply. EU privacy law (GDPR) may also apply because you have EU customers. Confirm both with a professional.
  • Holiday season (roughly November to December) brings a large, short spike in chat and email volume.
  • Your return policy may have exceptions, such as used gear, sale items, warranty claims and oversized shipping.

How to use this list

  • Send the written sections ahead of time (data, security, pricing, exit) and ask for written answers before the demo.
  • Ask the Top 10 live in the demo call, and ask them to show you rather than just tell you. Example: "Ask Fin about a return on a used tent right now."
  • Ask for evidence, not verbal assurances. Get documents, screenshots, sample invoices and contract clauses. If they say "yes, we do that," follow up with "where is that written down?"
  • Score the answers within 24 hours using the scoring sheet below, while the call is fresh.

Top 10 must-ask questions

  1. Exactly what counts as a billable "outcome"? If a customer stops replying, or leaves and comes back later with the same problem, are we charged? Please show us the written definition.
  2. Can we set a hard monthly spend cap? What does Fin do when the cap is reached: hand off to humans, stop, or keep going and bill overage?
  3. Please give us a cost estimate using our real numbers: [X] conversations a month normally and [Y] in December. What share would you expect Fin to resolve?
  4. Show us live how Fin answers return-policy questions with exceptions, such as used gear, final-sale items, 31 days after delivery, and EU customers. What stops it from making up a policy?
  5. Can we test Fin on our real past conversations or help content before going live, and see which answers it got wrong?
  6. What does the Shopify integration actually do? Can Fin look up a specific customer's order status, tracking and return eligibility? What can it not do, such as start a return or change an address?
  7. How does handoff to our 4 agents work in chat and email? Do agents see the full conversation, and what happens after hours?
  8. Where is our customer data stored? Can EU customer data be kept in the EU, and how do you help us respond to EU (GDPR) and US state privacy requests to access or delete data?
  9. Is our customer data or conversation content used to train AI models, yours or the model provider's? Where is that stated in the contract?
  10. If we cancel, what exactly can we export (conversations, help articles, Fin settings, reports), in what format, and how long until our data is deleted?

Full question list by section

A. Use-case fit and real-world performance

# Question Why it matters Good answer sounds like Red flag
A1 Can you show Fin answering 10 of our real return and exchange questions, including exceptions (used, sale, warranty, late returns)? Wrong policy answers cost money and trust A live demo using your actual policy, with correct handling of the exceptions Only canned demo data; "it'll learn once you're live"
A2 When Fin isn't confident, what does it do? Can we set topics where it always hands off, such as refunds over $X, damaged items or warranty claims? You want humans on the risky cases Configurable rules for handoff topics and thresholds, shown in the admin screen Only a general "confidence" setting you can't control
A3 What keeps Fin from inventing policy details that aren't in our help content? Made-up answers (often called "hallucinations") are the main accuracy risk Fin answers only from approved sources, cites them, and declines if the answer isn't found "It rarely happens"; n

Questions

How does the ai tool vendor evaluation questions work?

You answer a few questions, pay $19, and an AI worker does the job. The result appears on your private page, usually within minutes.

Can I get changes?

Yes. Every job includes one free change request: tell the donkey what to fix and it redoes the job with your notes.

Is it really done by AI?

Yes, fully. No human writes it. AI can make mistakes, so please review the result before you rely on it.

What happens to my information?

It is used only to do your job, is never used to train AI models, and is deleted after 30 days.

What if it fails?

Failed jobs are retried once, then refunded automatically.

Need something different?

Can't find it? Describe anything.

Our AI checks it's something we can do safely, then shows you exactly what you'll get and a flat price. Nothing is charged until you order.

More in Operations & SOPs