Questions to ask any sales AI vendor (including us)

Every sales leader is evaluating AI right now, most for the second or third time, and the demos all look the same. The call gets summarized. The email drafts itself. The dashboard glows.

Demos are the one thing every vendor in this category has mastered. So the demo can't be how you decide.

What separates a system your team runs on from a science project you'll quietly retire in Q3 only shows up under specific questions.

These are the ones we'd ask. They're also the ones we'd want to be asked, so where it's useful, we've included how we'd answer. Ask every vendor the same list and compare notes.

1. What happens for the rep who never opens it?

Most sales AI is opt-in by architecture: it works for whoever opens the tab, builds the habit, writes the prompts. Which means it works for your two most curious reps and does nothing for the other twenty, and you've bought another island.

A system answers differently: the work happens whether or not anyone remembered to show up. Prep arrives before the meeting. The CRM updates after the call. The follow-up drafts itself. If the vendor's answer to "what about the rep who ignores it?" is "training and change management," you're buying adoption risk and it's priced at zero in the proposal.

Our answer: leaders push workflows across the whole team; the reps who never log in anywhere still get their after-call work done in Slack, where they already live.

2. Show me what the leader sees

Not the rep experience. The team view. Which deals are slipping, who's following your sales methodology, which workflows are running, what changed since last week.

Rep copilots make individuals faster and leave the leader exactly as blind as before, and if you're a leader buying for a team, that blindness is the problem you were trying to solve. If the leader view is a usage dashboard (seats active, prompts run), that's telemetry, not visibility. You want to see the work, not the login stats.

3. What exactly can it write to my CRM, and can it delete anything?

The objection your RevOps person will raise first, and they're right to.

You want three specific answers. Writes are human-approved. Deletes are impossible. And every change is attributable in the audit trail. Anything vaguer than that ("it syncs automatically!") is a data-integrity incident on a delay.

4. What governance and control do we actually get?

This is the section most evaluations skip, and it's where AI purchases go wrong quietly. Sprawl taught every leader the lesson: AI without governance becomes ungovernable usage you can't see, standardize, or switch off. Buying a system is supposed to fix that, so make the vendor prove it does. Four sub-questions:

Who controls what runs? Leaders and ops should be able to define the workflows, push them across the team, see exactly what every workflow sends, and refine or kill any of them centrally. If each rep configures their own AI independently, you've bought your sprawl problem back with a nicer interface.

Can we see what it's doing? Every output, every action, every CRM change should be inspectable after the fact. Not aggregate usage stats, the actual work. If something produces a bad answer in front of a customer, you need to know what it said and why.

What are the boundaries? What is the AI allowed to do autonomously versus with human approval? Can it contact a customer without a human in the loop? (The right default: no.) What data can it reach, and can access be scoped by role or object?

What's the security posture? Where does your data go, is it used to train models, who can access it, and what certifications back that up. Ask for the security documentation, not a reassuring sentence.

Any serious vendor should answer this whole section without flinching, in writing.

5. What breaks when the underlying models change?

This one sorts real products from wrappers. The model layer is shifting under everyone, constantly. A vendor with real engineering absorbs those changes and your team never notices. A thin wrapper quietly degrades: prompts that worked stop working, outputs drift, and nobody notices until a number is wrong.

Ask how they test, how they catch regressions. If the answer is a blank stare, the maintenance burden you were trying to outsource is still yours; it's just hidden.

6. How does it handle a real pipeline, not a demo pipeline?

Demos run on ten clean deals. Your team has hundreds in flight, with messy stages, half-written notes, and edge cases that would embarrass everyone involved.

Ask the vendor to talk concretely about scale: analyzing hundreds of deals in one pass, flagging risk across a whole team's pipeline, handling the deal that's been open for 14 months with 9 stakeholders. Tools that answer off partial data are worse than no tool, because a forecast built on a partial read looks right and quietly isn't.

7. What proof do you have, with names attached?

Every vendor claims results. Ask which customers will say it on the record, with numbers.

8. What does my team stop paying for?

A fair question in both directions. New line items need to earn their place. Sales AI done properly should consolidate either tools or jobs that are being done in manual or expensive ways, for example: the standalone notetaker goes, or parts of the manual RevOps load go.

9. How will you support us during onboarding, and beyond?

The category is new enough that every rollout is partly a partnership, and this question exposes whether the vendor knows that.

Get specific on both halves. For onboarding: who actually works with us, how fast is time-to-first-value, and what does "live" mean? The answer should be measured in days, with real workflows running for the whole team inside the first week, not a 90-day implementation plan run from a slide. Sales teams don't have a spare quarter, and a tool that pays off "eventually" gets abandoned before it eventually arrives.

For beyond: what happens after week one, when the launch energy fades? You want a named human who knows your account, check-ins tied to the numbers you baselined rather than QBR theatre, and a vendor who proactively brings you new workflows as the product evolves instead of waiting for renewal season to reappear.

And ask the question almost nobody asks: what will you build for us? The AI category moves too fast for a rigid roadmap to be the whole answer. Great AI vendors behave like partners: when your team hits a workflow the product doesn't cover yet, the good ones build it with you, and it usually ships to every customer after. Ask for an example of something they built because a customer needed it.

10. If we leave, what do we keep?

An honest vendor answers this one plainly, and it tells you two things: whether your data is portable (it should be; it's yours), and where the product's real stickiness comes from. If retention depends on contractual lock-in, that's a warning. If it comes from the product doing so much work that leaving means hiring for it, that's a system earning its seat. Reps fighting to keep a tool is the version of this you can actually observe during a pilot: watch what happens when you try to turn it off.

How to run the evaluation itself

Three rules, whoever you pick.

Baseline first. Before the pilot starts, write down the numbers you want moved: days-in-stage, percentage of deals with a live next step, ramp time, forecast accuracy. If you don't have a before, the pilot review will be a vibes meeting.

A representative sample, not just volunteers. Piloting with your keenest rep tests the rep, not the tool. Pick a slice that looks like your org: a top performer, a few from the middle, someone skeptical, a newer rep. If it only works for the enthusiasts, you'll find out in the pilot instead of after the rollout.

Pick one number per use case. "Make us better at AI" isn't measurable. "Cut after-call admin to zero and lift next-step coverage above 90%" is.

Then, at the end, apply the take-away test: tell the team the pilot is ending and watch the reaction. Relief means you rented a reporting tool. A fight means you found something real.

Run this list on every vendor you're evaluating. Run it on us too. Book a demo and bring the hardest versions of these questions; the vendors worth buying from are the ones who enjoy them.