AI in sales, where it moves the pipeline and where it doesn't
The artificial intelligence applications that genuinely affect conversion and sales productivity, and the ones that sound good but change nothing.
In sales, artificial intelligence is sold as a promise of more opportunities. In practice, where it pays off most is in recovering time and improving information quality — which is what later enables better selling.
What works today
Call summaries and CRM updates. This is the application with the best effort-to-impact ratio. A rep loses between 30 and 60 minutes a day logging what happened. Automating it returns time and, more importantly, improves data quality: the CRM stops being half-empty, which affects forecasting, follow-up and every downstream analysis.
Meeting preparation. Pulling account history, recent interactions, public company news and open items before a call. What used to take twenty minutes takes two, and the conversation starts in better shape.
Portfolio prioritization. Ranking which opportunities deserve attention this week based on activity signals: replies, opens, account movements, time since last contact. It doesn't predict the future, but it stops deals going cold through neglect.
Follow-up drafts. Proposals, meeting recaps, responses to common objections. The rep edits and sends. Saves time without taking away control.
Call analysis at scale. Identifying which objections come up most, which topics correlate with won deals, where the process stalls. That information previously required listening to recordings one by one.
What doesn't work as promised
Automated mass prospecting. Generating hundreds of apparently personalized messages has a short-term effect and a long-term cost: it damages sender reputation, saturates recipients and lowers reply rates for the whole team. Real personalization requires research, and that doesn't scale infinitely.
Predictive scoring with thin data. If the company closes 40 deals a year, there isn't enough history for a model to learn something the team doesn't already know. At that volume, a simple explicit rule outperforms an opaque model.
Replacing the discovery conversation. The information that matters in a complex sale emerges when someone asks well and listens. No tool substitutes for that.
The adoption problem to anticipate
Sales tools fail for one recurring reason: reps perceive them as surveillance rather than support.
If the first application you roll out is a dashboard measuring individual activity, resistance is guaranteed. If the first application gives them back half an hour a day, adoption is voluntary.
The order matters more than the tool.
How to tell whether it's working
Three indicators, compared against an equivalent earlier period:
- Time in selling activities versus administrative time. The most direct measure of impact.
- Pipeline velocity: average days between stages. If preparation and follow-up improve, this moves.
- Stage conversion rates, paying particular attention to the stage where deals historically die.
A fourth, less obvious but revealing: CRM completeness. If the share of opportunities with current information rises, every other improvement has something solid to stand on.
A sensible rollout sequence
- Call summaries and automatic CRM entry. Immediate benefit, zero resistance.
- Meeting preparation. Uses the data the previous stage started cleaning up.
- Prioritization and alerts on opportunities with no movement.
- Aggregate call analysis to improve the whole sales process.
Each stage builds on the data quality created by the previous one. Starting at step four without doing step one means analysing incomplete information with sophisticated tools.
Frequently asked questions
Is AI useful for cold prospecting?
It's useful for research and preparation, not for sending volume. Mass-generated generic messages damage domain reputation and lower reply rates. The value is in personalizing few contacts with real information.
How reliable is AI opportunity scoring?
Useful as a priority order, not as truth. It works when there's enough history of won and lost deals; with thin data it simply repeats past biases.
What is the first application to implement in a sales team?
Automatic call summaries and CRM updates. It gives back real time every day, creates no resistance, and improves the data quality that everything else depends on.