How I test AI sales assistants

Ten tests, the same for every assistant, each graded from 1 to 5 on a fixed scale. The overall grade is their average plus my verdict, which can move it by up to one point. Every test is run by hand on an account I use, against the same set of sales calls, emails and CRM records, and the timings, screenshots and transcripts behind each grade go in the full test results published with the review.

Written by Miles Foster, Owner and tester, RankDude.

The RankDude mascot holding a laptop with a check on its screen

Each assistant gets the same sales week. I connect it to a test CRM (a HubSpot and a Salesforce instance I control, each holding the same 30-deal pipeline), the same sales inbox and the same calendar. Then I run the same material through it: ten recorded sales calls read from scripts by real people, each with 20 checkable facts, six next steps and a planted coaching problem; ten follow-up emails to draft; ten deals to update; a list of 50 named prospects to research; and three questions about the pipeline with known answers. What the assistant did is checked in the CRM, the inbox and the transcript, not in its own dashboard. Tools that do only part of the job, such as prospect data or outbound email, are graded on the tests they attempt and get no grade on the rest. Nothing is graded from a feature list, a case study or a demo.

Each review states how the account was obtained: bought at the public price, used on a free tier, or provided by the vendor. A vendor-provided account never changes a grade, and a vendor never sees a review before it is published.

How I rate

Every tool gets one score out of 5.

First, I run the same 10 tests on every tool and give each one a grade from 1 to 5. The average of the graded tests is the starting point; a test that cannot be graded is left out of it, and the review says how many tests the score is based on.

Then I add my verdict: what the tests cannot measure, such as how the tool holds up after the test weeks, how billing and cancelling work in practice, and how it compares with the others in the category. My verdict can move the score up or down by one point at most.

Every review shows how much my verdict changed the score, and why, in its scores table.

What a grade means

5	Excellent
4	Good
3	Average
2	Below average
1	Poor

The final score is the average of the graded tests (ten when every test can be graded), to one decimal place, plus my verdict, which can move it by up to one point. If a test couldn't be finished properly it gets no grade, and the review says how many tests the score is based on. A prospecting tool that does not record calls, or a note-taker that does not research prospects, is graded on the tests it attempts, and the review says how many.

The 10 tests

1.Call Notes Accuracy

The ten calls are read from scripts by real people on video and phone so every tool hears the same thing. Each script holds 20 checkable facts (budget, timeline, decision makers, objections, competitor named, figures) and six next steps with an owner and a date. The summary and the action list are marked against the script. Facts stated wrongly and next steps given to the wrong person are counted as errors, not misses.

5: 184 or more facts and 55 or more next steps, no wrong owners
4: 170 to 183 facts, or 48 to 54 next steps
3: 140 to 169 facts, or 36 to 47 next steps
2: 100 to 139 facts, or a wrong figure in a summary
1: under 100 facts

2.Follow-up Drafts

After each call I ask for the follow-up email. A draft passes when it restates the agreed next steps correctly, includes the two facts from the call that the prospect asked to see in writing, invents no pricing, feature or date, matches the length and tone of a 1,500-word sample of the rep's past emails, and could be sent as written. Drafts that needed a factual fix or that promised something not said on the call are counted separately.

5: 9 or 10 pass
4: 7 or 8
3: 5 or 6
2: 3 or 4, or an invented promise
1: 2 or fewer

3.CRM Accuracy

For each of the ten deals the assistant is asked, or set, to update six fields from the call: stage, amount, close date, next step, the new contact with title and email, and the competitor field. Each field is read in the CRM record itself and marked right, wrong or missing. Fields it filled with something not said on the call are counted separately. Whether it wrote directly or queued the change for approval is reported.

5: 57 to 60 right, nothing invented
4: 51 to 56 right
3: 42 to 50 right, or one invented value
2: 30 to 41 right, or two or more invented values
1: under 30

4.Prospect Research & Data

The list holds 50 real people at 25 companies in the US, UK and Germany whose current title and work email I have verified beforehand. The assistant is asked to find each one's work email and title, three facts about the company (size, funding or ownership, recent news) and a reason to reach out now. Emails are verified by a bounce check; titles against the company site or profile; facts against the source. Invented facts and invented people are counted. Tools without a prospect database get no grade here.

5: 45 to 50 right on email and title, nothing invented
4: 40 to 44
3: 30 to 39, or one invented fact
2: 20 to 29, or invented facts in several records
1: under 20, or an invented person

5.Deal Coaching

Each scripted call has one planted problem: the rep talks for 80% of the call, a pricing objection goes unanswered, no next step is agreed, a competitor is mentioned and ignored, a decision maker is never identified, a discount is offered unasked. The coaching output, or the answer to "how did this call go", is checked for whether it names the problem. A note that flags a problem the script did not contain counts as a false alarm.

5: 9 or 10 found, no false alarms
4: 7 or 8 found, at most one false alarm
3: 5 or 6 found
2: 3 or 4 found, or more false alarms than finds
1: 2 or fewer

6.Pipeline Answers

The test pipeline has known answers: which five deals have had no activity in 14 days, what the weighted forecast for the quarter is on the CRM's own probabilities, and which deals closing this month have no next step. Each answer is marked right only when the figure and the named deals match the CRM exactly. A fourth question with no good answer in the data is asked to see whether the assistant says so or invents one.

5: all three right, and it declines the fourth
4: all three right, but invents an answer to the fourth
3: two right
2: one right
1: none right

7.Connected Tools

Each connection is set up and then used: Salesforce, HubSpot, Pipedrive, Gmail, Outlook, Google Calendar, Zoom, Microsoft Teams, Google Meet, Slack, LinkedIn, and a phone line or dialler. A point counts when a test task completed through that connection: a call recorded, a record written, a message posted, a meeting booked.

5: all twelve
4: 10 or 11
3: 7 to 9
2: 4 to 6
1: 3 or fewer

8.Data & Permissions

Tried in the product and read in the policy: asks before sending any email in the rep's name, announces recording to call participants, lets an admin limit which records and fields it can write, keeps a log of every action, supports single sign-on or two-factor sign-in, states whether customer data trains its models, states a retention period and a way to delete recordings, publishes an independent security attestation (SOC 2 or ISO 27001), offers a choice of data region, and names the model providers that receive transcripts. One point each.

5: 9 or 10 points
4: 7 or 8
3: 5 or 6
2: 3 or 4
1: 2 or fewer, or it emailed a prospect without asking

9.Setup Time

The clock runs from the sign-up screen to the first correct call summary and the first correct CRM write, including every connection, permission, calendar hook, admin approval and onboarding call. Time waiting for a sales call or a manual onboarding is included. The number of settings changed to stop wrong behaviour in the first two days is counted.

5: under 1 hour
4: 1 to 4 hours
3: 4 hours to 2 days
2: 2 to 10 days
1: over 10 days, or a sales call is required before anything works

10.Pricing & Value

Every plan is read from the checkout or billing page, not the marketing page, and where there is none the quote I was given is used and the review says so. The figure graded is the yearly cost for five reps of the cheapest plan on which every test could be run, with platform fees, minimum seats, credits, per-call or per-contact charges and onboarding fees included. Renewal prices and what the free tier can do are listed with the payment methods seen at checkout.

5: under $3,000 a year for five seats, pricing published, no platform fee
4: $3,000 to $7,499 a year, pricing published
3: $7,500 to $14,999 a year, or a platform fee, or a credit cap hit during the week, or under $7,500 with the price only from a quote
2: $15,000 or more a year
1: poor value, or costs not disclosed before signing

Documented but not graded

Live coaching during calls, outbound sequencing and deliverability, role-play training, forecasting, languages, mobile apps and manager dashboards are checked and described in every review. When they work well they count in the written assessment, but they never add points to the overall grade.

Rules that apply to every test

Real use of the product on a real account, with real calls, a real CRM and real emails, never promotional descriptions, case studies or demo environments.
Identical scripts, pipeline, prospect list, tests, thresholds and sample sizes for every assistant in the category.
Authorised accounts only; no password sharing, no bypassing of plan limits or bot checks, and no real prospect contacted: every counterpart on the test calls and emails is a person who agreed to take part.
Every grade has evidence: a timing, a screenshot, a transcript or a table, in the full test results.
A grade changes only when a retest produces new evidence; the review records the date and the reason.

Questions about a specific grade go to the contact page. Makers can submit an AI sales assistant for testing. No AI sales assistant has been through these tests yet; every one I follow is listed on the AI sales assistants page, and reviews appear there as tests finish.