AI sales training pilot: a 4-week plan and decision criteria

Short answer

An AI sales training pilot should start narrow: one team or five to ten employees, one or two conversation types, two or three customer profiles, and four weeks. Record the starting position before the pilot, look at the same indicators every week, and at the end decide "continue", "continue with changes" or "stop" against criteria written in advance. The criteria must be written before the pilot, not after — otherwise any result reads as success.

The pilot's purpose is not to prove a sales increase — four weeks is not enough for that. It is to answer three questions: do employees use the practice, is the feedback useful to them, and is behaviour starting to change?

Before the pilot: four decisions

1. 1Which conversationOne or two conversation types that are lost or repeated most — say, the price objection and meeting setting. Do not test everything at once.
2. 2Who takes partFive to ten employees, both new and experienced. Picking only the most motivated distorts the result.
3. 3Who owns itOne person owns the pilot: builds profiles, runs the weekly meeting, collects notes.
4. 4Success criteriaWhat do you want to see at the end of four weeks? Written down and measurable.

The baseline

Record three things before the pilot. First, the current state of the target conversation type in real conversations — for example, in a spot-check of twenty real conversations, how many included a clarifying question after an objection. Second, employees' current practice habit — do they practise at all. Third, the time the manager spends on coaching. Without these three numbers, nothing remains at the end but a feeling that "it went well".

A four-week plan

1. 1Week 1 — setupProfiles and the standard are written, the owner tests each profile personally, employees are added and run their first session with the manager.
2. 2Week 2 — rhythmEach employee practises two or three times a week. At the end of the week, a 30-minute team meeting: the most visible mistake.
3. 3Week 3 — adjustmentProfiles are corrected based on the first two weeks' feedback. Employees stay on the same profiles so progress can be compared.
4. 4Week 4 — reviewThe real-conversation spot-check is repeated, employees answer a short survey, and results are compared with the criteria.

Success criteria: an example

The criteria below are an example; adapt them to your own goals.

• Usage: most pilot participants practised at least twice a week.
• Usefulness: most employees described the feedback in the survey as "specific and actionable".
• Behaviour in practice: progress on the target criterion is visible between week one and week four on the same profile.
• Behaviour in real conversations: in the spot-check, the target step appears more often than at baseline.
• Quality: the AI evaluation was mostly right in the samples the manager checked by hand.

The decision: continue, change, stop

• Continue: usage and usefulness criteria are met and the first change in behaviour is visible. Next step — a second team or a new conversation type.
• Continue with changes: there is usage, but the feedback is seen as unhelpful or AI errors are frequent. Fix the profiles, standard or criteria and test for two more weeks.
• Stop: usage is low and the cause cannot be removed — for example, the team has no time to practise and that cannot change.

What to collect

• Every week: number of sessions, average score per criterion, the three most frequent mistakes.
• Employees' short notes: which profile feels unrealistic, which feedback is wrong.
• The manager's notes: samples where they checked the AI evaluation, and the outcome.
• The baseline and final spot-checks of real conversations.

Data and agreement

Before the pilot starts, explain to employees what practice is for, what data is stored (name, phone number, text or recording of the conversation), who sees what, and what the results will not be used for. Do not use real customers' personal data in profiles. Check the requirements for processing personal data with a lawyer — in Azerbaijan, under the Law on Personal Data.

Illustrative example: a pilot's outcome

This is not a real customer case. Eight sales advisers at a car dealer run a four-week pilot with two profiles — a buyer hesitant about financing and a buyer comparing a competitor's model. In week two it emerges that on the competitor profile the AI customer sometimes invents a price that is not in the profile; the owner adds a fact sheet to the profile. By the end of week four the usage criterion is met and invitations to a test drive appear more often in the real-conversation spot-check, but some advisers find call practice uncomfortable because of audio quality. Decision: "continue with changes" — call practice with headsets, two more weeks.

Common mistakes

• Launching the pilot across the whole company at once.
• Writing the success criteria after the pilot.
• Not recording a baseline.
• Giving profiles to employees without testing them.
• Not naming a pilot owner.
• Expecting sales growth in four weeks.

Limitations

A four-week pilot shows usage and the first behaviour change, not long-term impact. In a small group, results are very sensitive to individual differences. Pilot enthusiasm — interest in something new — can fade later, so keep tracking usage after scaling up. A pilot result does not replace a legal compliance check.

A pilot with Vexvon AI Training

In a pilot with Vexvon AI Training the company builds its own customer profiles and good-call standard; for a company with no profiles yet, the system can generate the first one from the company's own knowledge base. Employees are added in bulk from a CSV or XLSX file, and practice runs as a Telegram chat or an AI test call. Progress is filtered per employee and group, by profile and date — which is where the pilot's weekly indicators come from. For demo questions, the platform checklist helps.

Next step

Write the four pilot decisions — conversation, participants, owner, criteria — on one page. To build profiles, see the AI customer profile; more articles are in rollout and reliability. To plan the pilot together, get in touch.