The first wave of AI in CRM produced predictions on a dashboard that nobody opened. The current wave writes into the record itself: it summarises the call, drafts the follow-up, updates the stage and suggests the next action. That is a different proposition, because it changes what a salesperson does in an hour.
What genuinely works today
- Call and meeting summarisation, which removes the worst admin task in most sales teams
- First-draft follow-up emails grounded in the actual conversation
- Next-best-action prompts based on stage, last contact and deal age
- Automatic field population from email and messaging threads
- Lead scoring that accounts for engagement recency rather than a static form score
What still needs a human
- Anything with pricing, discounting or contractual language
- Relationship repair after a complaint or a missed commitment
- Discovery questions, where the value is in reading the room rather than generating text
- Final judgement on whether a deal is real, which is where over-trusted scores cause damage
The data conditions AI adoption depends on
AI features inherit the quality of your CRM data. If stages are used inconsistently, if notes are empty and if half the deals have no close date, every suggestion will be plausible and wrong. Fix the record discipline before switching the features on, in that order.
A safe 90-day adoption path
- Start with summarisation only, and measure whether notes are actually getting logged
- Add drafting, with a mandatory human send step, and track reply rates as the test
- Add scoring last, once the underlying stages are trustworthy
- Review a sample of AI-drafted messages monthly for tone and factual accuracy
- Publish an internal rule on which fields AI may write to without review
Cost and licensing questions to ask vendors
- Is the AI capability included in the tier you are buying, or priced per seat above it
- What are the usage limits, and what happens when you exceed them
- How is your customer data used, retained and isolated
- Can the feature be scoped to specific teams rather than switched on for everyone
Where this connects to the rest of your stack
An AI-capable CRM is only useful if leads arrive reliably and outcomes are measured. That means clean capture from your ads and forms, and honest reporting out of GA4 guide. For the earlier question of which platform to sit on, see our CRM comparison for Indian small businesses.
The realistic expectation
Expect meaningful time savings on admin and faster first responses. Do not expect the tool to fix a pipeline that has no qualification process, and do not expect it to replace the judgement that closes a deal.