Vendor AI demos are a tax on MSP time. Someone from a channel team books a thirty-minute slot, brings a polished slide deck, and shows you a chatbot answering a planted question about return policies. You smile, ask for a follow-up deck, and then spend two weeks explaining to your account manager why you are not putting that thing in front of a dental office that still prints every invoice.
You can end the demo early without being a jerk. The move is not theatrical. It is a short, boring checklist you run while they still have the screen shared. Twenty minutes is plenty if you stop watching the show and start poking the product where your clients actually live.
What you ask before the demo even starts
Send three questions in the calendar invite reply. Do not wait until you are on the call. Ask who owns the data when the contract ends, what the per-seat and per-token economics look like at fifty users, and whether the product can sit behind your client's existing identity provider without a second login. If the reply is vague, you already know the demo will be theater. Keep the meeting anyway if the relationship matters, but go in with a stopwatch and a notepad, not hope.
I keep a one-page sheet for these calls. Column one is the claim they make. Column two is the client scenario that would break it. Column three is whether they showed a real fix or a slide. After five vendor meetings, the sheet fills itself. You start spotting the same soft answers: "roadmap item," "depends on the tier," "our professional services team handles that." Those phrases are exit ramps. Write them down in front of the rep. People get more honest when they see notes forming.
A regional MSP I know used this sheet on a white-label helpdesk bot pitch. The vendor promised password resets through Teams. On paper that sounded useful. The sheet forced a second look: the client's SSO was Okta with conditional access, half the staff used personal phones for MFA, and the helpdesk already had a self-service portal that nobody used because the link lived in a SharePoint site nobody bookmarked. The demo never got past minute twelve because the bot could not clear the MFA prompt without a human. That is a useful twelve minutes. You learned the product fails where your tickets actually pile up.
The twenty-minute kill sequence
Minute one through five: refuse the happy-path script. Ask them to open a fresh tenant, not the golden demo environment. If they cannot, ask why. A demo tenant that already knows the answers is a magic trick. You want the cold start your client will get on a Tuesday after a failed Windows update.
Minute five through ten: feed it a ticket that looks like your queue. Use a real phrasing from a client, scrubbed of names. Something like "VPN connects but Outlook keeps asking for a password and the printer on the third floor is offline again." Watch whether the tool asks clarifying questions, invents a root cause, or dumps a generic article. If it invents a root cause with confidence, you have your reason to stop. Confident wrong answers create more tickets than they close, and you are the one who will eat the callback.
Minute ten through fifteen: ask about handoff. When the bot fails, how does the ticket land in your PSA? Does it preserve the conversation? Does it tag the device? Does it create a duplicate for the same user who already opened three tickets this week? Most products dump a blob into email or create a naked ticket with no asset link. That is not automation. That is a new inbox with worse search.
Minute fifteen through twenty: money and exit. Ask what happens when usage spikes during month-end for an accounting firm, or when a manufacturing client dumps a hundred SOPs into the knowledge base and the retrieval quality falls off a cliff. Ask for the offboarding steps in writing: export of prompts, conversation logs, embeddings, and custom workflows. If they cannot describe the exit, treat the product as a hotel you can check into but never leave. MSPs have been burned by that pattern with backup vendors and RMM tools. AI vendors are not magically better at contracts.
You can keep the relationship while you kill the demo. Thank them for the time. Tell them exactly which three gaps blocked a pilot. Offer a date three months out if those gaps close, and mean it. Channel reps remember the partner who gives a clean no with reasons more fondly than the partner who ghosts after asking for pricing. Your AE still has a number to hit. Help them by being specific. "We need PSA write-back with ConnectWise Manage custom fields, Okta group-based access, and a hard cap on token spend we can bill as a line item" is a gift. "Not a fit right now" is a shrug.
A few demos deserve to die earlier than twenty minutes. If the product requires your client to paste customer data into a consumer chatbot with no BAA and no residency controls, end it. If the only differentiator is a skin over a public model you already resell through your own stack, end it. If the pricing assumes every seat in the company will use the tool daily when your helpdesk data shows eight percent of users open more than two tickets a month, end it. You are not obligated to sit through the roadmap section when the commercial model already fails.
What you take back to the office
Write the outcome in your internal channel notes the same day. Include the vendor name, the gaps, and whether you want another look later. Future you will forget which chatbot promised ConnectWise sync and which one only emailed a PDF transcript. Your techs will also stop getting ambushed when a salesperson forwards a "quick intro" without context.
Use the failed demo as sales material with your own clients, carefully. You do not need to trash a vendor by name. You can tell an SMB owner that you evaluated three tools this quarter and rejected the ones that could not hand off to your board cleanly. That story sells your judgment. Clients hire MSPs to be the adult in the room when a shiny AI pitch shows up in their inbox from a cold LinkedIn message.
If a vendor later fixes the gaps, re-open the conversation without apology. Markets move. Products improve. Your job is not to hate vendors. Your job is to refuse unpaid science projects that land on your service desk at 4:55 p.m. on a Friday. A clean kill in twenty minutes protects the relationship, protects your margin, and keeps your calendar free for the pilot that might actually close tickets.