Almost every article about AI in sales is about what you can automate. This one is the opposite, because in B2B the damage from over-automating exceeds the benefit of under-automating.
The context
McKinsey finds 80% of respondents perceive individual productivity gains, while only 37% of organisations detect EBIT impact. In sales that gap has a specific explanation: activity goes up, conversion doesn’t.
Sending more emails is not selling more. In some cases it is selling less.
What you should not automate
The personalised first touch. AI writes emails that look personalised. Your prospects receive a dozen a day and recognise them instantly. What used to signal that someone had researched your company is now noise, and the sender pays the reputational cost.
Qualification of large opportunities. A model can score leads on firmographic data. It cannot detect that your contact has said “I’d have to run it past my boss” three times, which is the information that actually determines whether this closes.
Responding to an objection. The moment a client voices a real concern is the moment of maximum value in the relationship. Delegating it to a system throws away the one part of the process that justifies having a salesperson.
Follow-up on an at-risk account. When a customer is unhappy, the automated email makes things measurably worse.
Commercial terms. Discounts, timelines, scope. Anything that commits the company goes through a person with authority. It is common sense, and it is also the correct practice under the EU AI Act where a decision has effects on third parties.
What you should, heavily
AI in sales genuinely delivers on everything that happens around the conversation:
- Preparing the meeting. Account summary, history, recent news, who’s who. Fifteen minutes of work that is now thirty seconds.
- Keeping the CRM current. The rep dictates two sentences after the call and the system fills the fields. This one moves the needle, because a badly maintained CRM is the silent problem of nearly every team.
- Spotting pipeline signals. Opportunities with no activity, overdue dates, contacts who stopped replying.
- Summarising calls and extracting commitments. Who said they would do what, and by when.
- Drafting proposals from templates and account history, for the rep to adjust.
The pattern is clear: automate the preparation and the record-keeping, not the conversation.
The thirty-second test
Before automating anything customer-facing:
If your prospect knew a system generated this, would they mind?
If the answer is yes, don’t automate it — because they will find out. If it is no, or they wouldn’t care, go ahead.
A B2B sales team’s asset is the trust accumulated across a few relationships. It is slow to build, fast to lose, and no productivity gain compensates for having burned it.