Last checked: 12 Sep 2026.
A prospect put the question to us in plain terms in August 2026: "Is our sales team AI ready, AI competent?" Not "can you run us some AI training", but can we measure it. They wanted a bespoke programme with, in their words, "something that comes out of it": a certification the business could assess people against, not a library of generic videos. And they wanted it to be bigger than what they already called nano-learning in house, on the grounds that the pace of AI change makes a one-off session obsolete within months.
That is the right instinct, and a harder brief than it sounds. Competence is a claim about behaviour, and behaviour has to be observed. This post sets out how to find out where a sales team actually is, how to assess it, and what a certificate should and should not say.
Why generic video training does not produce competence
Most AI training on the market is a recorded course: an hour on what a large language model is, an hour of prompt tips, a quiz at the end. It produces completion data, not competence, for three reasons.
First, it teaches a tool, and the tools move. OpenAI's Assistants API was officially sunset on 26 Aug 2026 and is no longer available, per OpenAI's own migration guide. Anthropic's pricing documentation now states that the planned 1 Sep 2026 increase for Sonnet 5 will not occur. Those are two changes in a single fortnight, in two vendors, on things a course would have taught as fixed.
Second, a quiz tests recall, not judgement. The question that matters in a sales seat is not "what is a token", it is "this draft reply cites a contract clause, do I send it".
Third, one session leaves no retention curve. Our HubSpot training already runs on spaced learning, short sessions of around 45 to 60 minutes with deliberate gaps between them, because spreading learning across separated sessions produces far better long-term retention. The same applies here, and more so, because the subject matter changes underneath the learner.
The operator matters more than the tool
John Kelleher wrote on LinkedIn on 9 Sep 2026 that on the same task, with the same model, a skilled user gets an excellent result and a poor user struggles. He was a qualified teacher and an assistant headteacher before SpotDev, and the observation is the one every teacher recognises: the teacher matters more than the school.
The commercial consequence is direct. If results vary wildly across a team on identical tools, the answer is not a different vendor. It is that the spread of skill is wide and nobody has measured it.
A four-level competence framework for a sales team
Four levels, each defined by behaviours somebody else can watch. The spine is conscious competence: the goal is to get every relevant person to the point where they can do the work deliberately and reliably, not to the point where they have heard of it.
| Level | Observable behaviours | How it is assessed | Evidence produced |
|---|---|---|---|
| 1. Safe use and data handling | Uses the company account, not a personal one. Knows which client data may be pasted into which tool and which may not. Recognises when a request needs sign-off. Reports a near miss rather than hiding it. | Scenario test with five to eight judgement calls, marked pass or fail on each. Spot check of tool accounts in use against the approved list. | Scenario score, a signed acceptable-use acknowledgement, and an exceptions list for the compliance owner. |
| 2. Prompting and verification | Gives context, constraints and a worked example rather than a one-line request. Checks claims against the source before use. Rewrites rather than accepting the first output. Says when the tool is the wrong tool. | Two timed practical tasks (a prospect research brief and a reply to an awkward client email), marked against a published rubric by a second assessor. | The two artefacts, the rubric scores, and the assessor's note on what was corrected. |
| 3. Working with CRM-connected agents | Understands that a connected assistant can write, not only read. Checks what an action changed before moving on. Knows which record types are in scope and which are excluded. Uses the audit trail. | Supervised task in a sandbox portal: run a connected action, then find and explain the resulting record changes and the audit entry. | A before-and-after record diff, the audit log entry, and the assessor's observation sheet. |
| 4. Designing and owning an agent's goal | States the goal, the trigger, the guardrails and the exit condition in writing. Defines what "done" looks like. Owns the review cadence and can say when the agent should be switched off. | Written design for one real pipeline job, defended in a 30-minute review with an engineer and the sales lead. Reviewed again after 30 days of live running. | The signed-off design document, the 30-day review with actual numbers, and a named owner on the record. |
Level 3 is where most teams are quietly exposed. HubSpot's ChatGPT connector, per HubSpot's knowledge base article dated 16 Jul 2026, creates and updates contacts, companies, deals, tickets, line items, products and engagements, handles up to ten records per bulk action, attributes activity in the Audit Log, and does not apply custom validation rules. A salesperson who assumes an assistant only answers questions will not check what it changed. That is a training gap, not a software fault.
How to assess it
Four components, in order.
A baseline audit of current use. Before designing anything, find out what is already happening: which tools the team uses, on which accounts, for which tasks, and with what data. Expect personal accounts and tools nobody approved. This is what tells you whether you have a training problem, a governance problem or both.
Practical tasks scored against a rubric. Real tasks from the team's own work, timed, with the rubric published in advance so people know what good looks like. Two assessors on a sample, because a single assessor drifts. A rubric two people cannot apply consistently is an opinion.
A before-and-after survey. We run the same competence survey before the first session and after the programme ends on our HubSpot training, which gives a comparable measure of what changed. Do the same here, and treat the self-assessment as a confidence measure sitting alongside the assessed scores, not as a substitute for them. The gap between the two is itself useful: a team that rates itself high and scores low is the one most likely to send something it should not.
Re-certification every six months. Not an administrative ritual. Model behaviour, pricing, retention terms and connector scopes all changed materially in the twelve months to September 2026, and a certificate that ignores that is claiming something it cannot support. Six months is short enough to stay true and long enough to be worth passing.
A modular programme over 6 to 12 months
Structure it as a sequence of short sessions with gaps, not a course. An outline that works for a sales team of ten to forty people:
- Weeks 1 to 3, baseline. Usage audit, acceptable-use position agreed, survey one, level 1 scenario test for everybody.
- Months 1 to 3, levels 1 and 2. Six to eight spaced sessions on safe handling, prompting, verification and the team's own repeat tasks. Assessment at the end of month 3.
- Months 3 to 6, level 3. Connected tools in a sandbox portal, record changes, audit trails, escalation. Assessed in the sandbox, not on live data.
- Months 6 to 9, level 4 for a subset. Usually the sales lead and two or three others. One real agent design each, defended and then run.
- Months 9 to 12, review and re-certify. Survey two, re-test levels 1 to 3 against the current tool set, 30-day reviews on the level 4 designs, update the material for whatever the vendors changed.
Budget for the material to be rewritten at least twice inside a year. That is a feature of the subject, not a failure of planning. For reference on adjacent work, our published rate for HubSpot training is stated on the training page: "HubSpot training starts at £500 plus VAT for four sessions. Bespoke packages can be designed to your needs." An AI competence programme is a different piece of work and is scoped and priced separately.
What the certificate should and should not claim
A certificate is a statement of fact about an assessment. Keep it to what was actually observed.
It should state: the person's name, the level reached, the date assessed, the specific tools and versions the assessment covered, who assessed it, and the expiry date six months later. If a level was partly met, say which parts.
It should not claim: any vendor's endorsement or certification (it is yours, not OpenAI's, Anthropic's or HubSpot's), that the person is "AI trained" in general, an equivalence to a regulated qualification, or anything about tools the assessment did not cover. And it should not be presented to a client or a regulator as evidence of a control unless the assessment records behind it are available.
The honest measure is in the CRM, not the completion rate
Course completion is the easiest number to produce and the least informative. If the programme worked, it shows up in the pipeline data: response times on inbound enquiries, how many quotes go quiet past 30 days, the share of deals with a next step recorded, and the correction rate on AI-assisted output before it goes out. Agree those measures first, take the baseline, and read them at 6 and 12 months against it. If the CRM numbers do not move, the training did not work, whatever the certificates say.
Frequently asked questions
How long does it take to find out whether our sales team is AI competent?
The baseline audit and the first assessment take about three weeks for a team of ten to forty people: a usage audit, a scenario test on safe use and data handling, and a competence survey. That is enough to tell you where people sit across the four levels and which problems are training rather than governance.
Can you certify our team against an official AI qualification?
No. There is no vendor or regulator qualification for general business AI use that a consultancy can award on someone else's behalf. What you can have is your own internal certification, assessed against a published rubric, valid for a stated period and honest about what it covered. That is what most buyers actually want when they ask for certification.
Why re-certify every six months?
Because the ground moves. OpenAI's Assistants API was sunset on 26 Aug 2026 and is no longer available. Anthropic's pricing documentation states that the 1 Sep 2026 increase for Sonnet 5 will not occur. Connector scopes, retention terms and model behaviour all change on similar timescales, so an assessment more than six months old is describing tools that have since changed.
Is this the same as AI training?
No. Training is the delivery. Competence assessment is the measurement, and it is the part that gives the business something it can act on: who can be trusted with which tasks, where the gaps sit, and whether the money spent changed anything. A programme that only delivers sessions cannot answer any of those questions.
What if our data is too messy for AI to be useful anyway?
That is a common and fair objection, and it is worth testing first. Inconsistent contact and company records undermine anything built on top of them, and no amount of team competence compensates. A short diagnostic will tell you whether data work needs to come before the training programme or alongside it.
Who should be assessed at level 4?
A small group, usually the sales lead and two or three people who already own a process end to end. Level 4 is about designing an agent's goal, guardrails and exit condition and then owning the result, which is a management responsibility as much as a technical one. Assessing everybody at level 4 produces designs nobody is accountable for.
Where to start
Start with the baseline: what is in use today, on which accounts, with which data. Our approach to AI for sales covers the agent side of the same picture, and the method behind the sessions themselves is set out in our guide to training a team on HubSpot. For a structured look at readiness before committing to a programme, the AI and Data Readiness Assessment sits on our AI implementation page.
SpotDev is an AI and digital transformation consultancy that builds real software. If you want this scoped, request a quote.
Stay Updated with Our Latest Insights
Get expert HubSpot tips and integration strategies delivered to your inbox.
