Choose AI sales coaching software by running a test in the demo: bring one real discovery call and your own scorecard, and ask each vendor to score it live. Then ask six more questions, on coverage, your CRM, what happens after a low score, proof of improvement, total cost and setup. Each one is below with a weak answer and a strong one.
I work at Triple Session, which sells AI sales coaching, so I wrote the tests to be passed with your calls and your numbers, by us as much as anyone. The facts about HubSpot and Gong come from their own pages, read on November 3, 2026.
Do you need scoring or a recorder?
If your managers already hear most of their team's calls, buy a recorder and stop there. A recording with a transcript is enough when someone is listening.
The problem starts when nobody is. Take a team at Corvane, a made-up software company: eight sellers, four calls a day each and 20 working days in a month. That is 640 calls. A manager who listens to two calls per person per week hears 64 of them, one in ten. Those 64 are rarely a random sample, because they tend to be the calls someone flagged or a deal forced.
Those numbers are assumptions, so put your own in. If the share comes out near one in ten, you need something that scores the calls for the manager, and the seven tests below are how to compare what is on offer.
The seven tests
| Test | Ask the vendor | Weak answer | Strong answer |
|---|---|---|---|
| 1. Your call, live | "Score this call against our MEDDICC scorecard, right now." | "We'll set that up after the contract." | Your steps, scored live, with the moment in the recording for each one. |
| 2. Coverage | "What share of yesterday's calls got scored?" | "Managers choose which calls to review." | A number, and it is every call that was recorded, with nobody picking. |
| 3. CRM | "Show me the record in my CRM after a call." | A note pasted into the timeline. | A named property on a named record, written with nobody typing. |
| 4. After a low score | "My person scored 0 on economic buyer. What does she get today?" | A dashboard her manager might open Friday. | Feedback on that call, plus a practice round on that exact skill. |
| 5. Proof | "How will I know she improved?" | Her average call score. | Her economic buyer score on the next ten discovery calls. |
| 6. Cost | "What does year one cost for 25 people, all in?" | "Let's loop in procurement first." | A number on the call, with seats, fees, onboarding and add-ons itemized. |
| 7. Setup | "Who builds the first scorecards, and when is the first scored call?" | "It depends on your setup." | A named person, a start date and a date for the first scored call. |
Test 1: score your call, live
A demo call is chosen by the vendor, so it is the call that makes the product look good. Yours are different.
Bring a discovery call you know well and your own scorecard. Pick one where you know a step was missed, for example nobody asked who signs. Then watch what the tool does with it. A scorer that passes the step you know was skipped will pass everything.
That risk is documented. Gong's help center, in its guide to writing scoring questions, says that "AI has a bias to answer 'yes'", and tells admins to define what counts as a yes and what counts as a no (Gong Help Center, "Gong AI for scoring", updated September 13, 2026, read November 3, 2026). Ask the vendor how they handle it. A good answer shows the moment in the recording that earned each score, so you can check it by ear.
Before the demo, make sure your scorecard is something two of your managers would score the same way. Sales call scoring: how to build a scorecard your team uses has a template and a calibration routine.
Test 2: coverage
Ask what share of yesterday's calls were scored without anyone choosing them. Anything under all of them puts you back to guessing, because the calls that go unscored are the ones nobody picked.
If you use HubSpot, there is a limit upstream of the vendor. HubSpot's knowledge base says call transcripts need a Sales Hub or Service Hub Professional or Enterprise subscription, and that only recordings made by users with those seats are transcribed automatically (HubSpot Knowledge Base, "Review call recordings and transcripts", updated October 9, 2026, read November 3, 2026). The share of calls a tool can score cannot be higher than the share that was recorded and transcribed in the first place. So ask for two numbers, calls held and calls scored, and ask what the gap is made of.
Test 3: what lands in your CRM
"We integrate with HubSpot" can mean anything from a synced deal property to a note pasted into the timeline. Ask the vendor to open a record after a real call and show you where the score is.
It helps to know what is there to write to. HubSpot's developer documentation lists calls as their own type of record, with the type ID 0-48, and describes them as a type of activity that represents phone calls associated with your records. The calls guide lists properties such as the call title, body, duration and recording URL (HubSpot developer docs, no date on the pages, API version 2026-09 in the examples, read November 3, 2026). Ask three things:
- Which record holds the score: the call, the deal or a note?
- Which property, with which name?
- Who writes it, the tool or a person typing?
A note in the timeline is easy to read and hard to filter or report on. A named property can be used in a list, a view or a workflow.
Test 4: what a low score triggers
Ask what a person gets today after a score of zero on economic buyer. If the answer is a dashboard her manager might open on Friday, you are buying a recorder.
A transcript nobody acts on does not change the next call. What does change it is specific feedback on that call, naming the moment, and a chance to practice that skill before the next live call. Ask whether the practice is scored on the same scorecard as the real calls, because otherwise you are measuring two different things.
Test 5: how you will know she improved
Ask how you will know the coaching worked. The weak answer is an average call score going up, which tells you little: it moves with the mix of calls and buyers in the week.
A better answer is the step you coached. If the focus was economic buyer, her economic buyer score on the next ten discovery calls is the number to watch. Ask whether the tool can show a score for one step by person over time, not only the total.
Test 6: total cost for your headcount
Ask for year one for the number of people you have, all in. The pieces to itemize are seats, any platform fee, onboarding, add-ons, and the CRM tier the tool needs. On the last one, check the HubSpot subscription requirement from test 2 against what you pay today.
Some vendors publish prices and some do not. Gong's pricing page says licenses are priced per user and that there is a platform fee based on the number of users supported, with no dollar amounts, only a form for a customized proposal (gong.io/pricing, read November 3, 2026). That is a common way to sell, and the cost of it is a sales cycle spent finding out. If a price is not on the site, ask for the itemized total in writing before you commit time to a pilot. For how to estimate one from public data, see Gong pricing in 2026.
Test 7: who builds the first scorecards
Ask who builds the first scorecards and when the first scored call happens. There are two honest answers. Your enablement team builds them, which costs their hours, or the vendor builds them with you, which may cost money or time. Either can be right for your team.
What matters is that the answer has a name and a date. "It depends on your setup" means nobody has planned it yet.
How to compare what you find
Score each vendor 0, 1 or 2 on each test: 0 for a weak answer or none, 1 for a partial one, 2 for a strong one backed by something you saw. A 0 on test 1 or test 2 ends the comparison, since a tool that cannot score your call or cover your calls cannot coach. For the rest, add up the totals and read the gaps, not only the sum.
If you want to see the whole practice end to end first, AI sales coaching: what it is and how to run it walks through capture, scoring, coaching, practice and verification.
Sources
All read on November 3, 2026.
- Gong Help Center, Gong AI for scoring, published and updated September 13, 2026.
- HubSpot Knowledge Base, Review call recordings and transcripts, last updated October 9, 2026.
- HubSpot developer docs, Understanding the CRM and the calls guide, no date shown.
- Gong pricing, no date shown.