AI for sales works where it removes the work around the conversation, not where it tries to replace the conversation. After analysing 1,000,000 B2B cold calls in France and running an AI dialer with European teams, we rank the 12 use cases below by the impact we actually see, from "changes the funnel" to "saves time". Start with the first four.

The rule that decides whether an AI use case works

Two kinds of AI exist in sales. One tries to hold the conversation for you: an agent that emails, replies and calls on its own. The other prepares, supports and cleans up around a human conversation. In our experience the second kind pays; the first kind burns domains and brand in a market as small as France.

So the test for every use case is simple. Does it increase conversations per rep per day, or meetings per 100 conversations? If it moves neither, it is a nice tab.

Tier 1: four use cases that change the funnel

1. Signal detection in your own CRM

Website visitors, job changes, closed-lost deals, dormant leads, social intent. Reps use only about 20% of the signals already in their CRM. 50% of Sonik's own pipeline comes from signals that were already there. This is the cheapest AI use case, because the data is paid for.

2. Parallel dialing with voicemail detection

The average connect rate on B2B cold calls in France is 11.3%. To get 20 conversations a day, a rep needs about 180 dials. Parallel dialing with answering-machine detection gives 5x more conversations per hour, which turns a full day of dialing into a call block.

3. Call prep with a trigger-first opener

65% of reps open by explaining what they sell. Trigger-based openers reach a 7-10% meeting rate against 2-3% for pitch-first openers. An AI brief that puts the trigger on screen before the pickup is worth more than any script library.

4. Dormant-lead reactivation

Calling a known contact who went quiet, opening with "How have you been?", reaches about 10% success against about 1.5% for a standard cold open. AI finds those contacts in the CRM and schedules the call. See our guide on reactivating dormant leads.

Tier 2: four use cases that make reps better

5. Call scoring on the behaviours that book meetings

Discovery questions starting with "how" lift conversion by 190%. Asking "What do you mean?" before answering an objection: +197%. AI scoring that counts those, across every call, beats a manager listening to one call a week.

6. Live prompts during the call

The objection the prospect just raised, the competitor they mentioned, the last note from the CRM, on screen while the rep talks. Useful in month one and two of a rep's ramp; less needed after.

7. AI roleplay for ramp

Let new reps practise the opener and the first objection against a bot, not against your top 50 accounts. Measure the roleplay by the same scoring as real calls, so practice and reality use one rubric.

8. Automatic summary and CRM write-back

Outcome, summary, next step, logged without typing. The value is not the minutes saved. It is that the CRM becomes complete, which makes use case 1 work next quarter.

Tier 3: four use cases that save time

9. Mobile number enrichment

A waterfall across providers to find a direct line, with GDPR-compliant sourcing. No number, no conversation; everything above depends on it.

10. First-draft emails from a signal

AI drafts the email from the trigger and the CRM history; the rep reviews and sends. Review is the point. Blasting drafts is how you end up in spam.

11. Phone number health monitoring

Carrier spam labels kill connect rates silently. AI watches pickup rate per number and rotates the ones that drop. Cheap, invisible, decisive.

12. Pipeline hygiene and forecasting

Stale deals flagged, next steps enforced, forecast built from activity instead of gut feel. Real, but it is the last thing to buy, because it reads data the first eleven produce.

GDPR and the European reality

None of this is optional in Europe, and most AI-for-sales guides are written as if it were.

• Lawful basis. B2B prospecting usually rests on legitimate interest, with a right to object honoured at the first ask. Log the objection in the CRM and let the AI read it.
• Call recording. The person must be informed before the recording starts. Check your dialer's notice flow per country.
• Hosting and processors. EU hosting, a signed DPA, a public sub-processor list. Ask whether your transcripts train the vendor's models, and say no if you want to.
• Language. Test transcription and scoring in French on real calls. A model that is 95% accurate in English and 80% in French scores your reps wrong.

How Sonik approaches AI for sales

Sonik covers the Tier 1 use cases and most of Tier 2 in one screen: signals read from HubSpot or Salesforce, a parallel dialer, an AI prep card with the trigger and the opener, summary and next step written back, and coaching scored on our 1,000,000-call benchmark.

We leave email sequencing, enrichment and forecasting to the tools you already have, and feed them the call outcome. The conversation stays with the rep. That is a design choice, not a limitation.

Conclusion

AI for sales is not one product and not twelve. It is a short list of jobs around the conversation, ranked by what they do to conversations per day and meetings per conversation. This week, pick use case 1: export the contacts in your CRM with a job change or a website visit in the last 30 days, and call them at 10am and 4pm. That list is the cheapest AI project you will run this year.