The real state of adoption
In March 2025, McKinsey published the results of its B2B Pulse Survey —3,942 commercial leaders surveyed across global markets— with one central question: how much of the promise of generative AI in sales has become real practice?
The answer: 19% already have it deployed in their sales process. 23% are in the process of doing so. The remaining 58% haven't taken the first step yet.
What distinguishes that first group isn't just having launched a tool. It's that they've already made the mistakes, adjusted the process, and have their own data on what works. When markets move fast, that timing difference becomes a structural advantage.
The most relevant number isn't the 19%. It's what that 19% reports having learned along the way — because that's exactly what the 81% about to launch still doesn't know.
The 7 use cases working today
The study identifies seven use cases where generative AI is already generating measurable return in B2B teams. These aren't lab experiments: they're production implementations, with documented results.
Messages tailored to the specific context of each account and each moment in the process — generated automatically without losing precision or tone.
Automated feedback on real sales conversations, running over 100% of calls without depending on manager availability.
Automatic briefings before each call: account history, deal context, previous objections, and suggested next steps.
Reduced response time in bidding processes, with proposals that already incorporate each account's context.
Data captured and structured directly from conversations — without depending on reps to remember to update the CRM after each call.
Prioritization based on real buyer behavior signals, not static criteria defined once a year.
Pattern identification in won and lost deals to update the playbook with real evidence, not team intuition.
Seven different use cases, one common pattern: all of them work better when there's structured conversation data. Without it, the model works with incomplete information — and the results look a lot like what the team already has today.
The smart coach: +7 satisfaction points, −20% training costs
The best-documented case in the study is the smart coach, implemented at a telecommunications company. Results: +7 points in customer satisfaction and −20% in training costs.
This isn't a lab result. It's what happens when feedback on how each rep sells stops depending on a manager being able to listen to a call in time — and starts running automatically over every conversation.
On a team where the manager can review 2–3 calls per week per rep, 90% of conversations never receive feedback. With an active smart coach, 100% do. Coverage is the mechanism that generates the result — not the sophistication of the model.
The reduction in training costs follows the same logic: when feedback arrives in real time and on real cases (not on simulations in a quarterly training session), the time the team needs to iterate compresses significantly.
The 5 lessons from those who already deployed it
Here's what the 19% learned — and what the 81% about to launch still doesn't know:
- Data quality is the entry condition, not the starting point. If conversation data is scattered, incomplete, or unstructured, the model has nothing to work with. The most common mistake is launching the tool before solving the capture problem.
- The cultural change is harder than the technical one. The tool deploys in weeks. The team changes how it works in months. The biggest risk isn't technical: it's reps using the tool as a formality without incorporating its outputs into real practice.
- Starting with a narrow use case accelerates return. Trying to cover all seven at once dilutes focus and makes it impossible to measure what works. Teams with the best results start with just one — the one generating the most friction today — and expand from there.
- Human oversight isn't a complement: it's part of the design. Gen AI in sales isn't autopilot. It's assistance. Teams that design it this way from day one don't run into the «the system says X but the rep does Y» problem.
- Adoption KPIs are not impact KPIs. Measuring how many reps use the tool this week says nothing about whether sales is improving. The most mature teams measure conversion rate, deal velocity, and win rate — not logins.
The common denominator of these five lessons: adoption is harder than launch. Companies that know this before day one arrive better prepared.
The gap that grows every quarter
The pattern that emerges from the study isn't one of uniform adoption advancing at a constant pace. It's one of bifurcation.
Teams that already have gen AI active are iterating on implementations that work, accumulating their own data, and improving their playbook week by week. Those who haven't started yet are paying the opportunity cost while waiting for the «right moment» to launch.
The distance between both groups isn't about technology. It's about accumulated learning. And that learning has one particular characteristic: it can't be bought or accelerated once the gap already exists — it's built use by use, conversation by conversation.
The 23% currently in process will make it in time. The rest risk arriving when the gap is already structural.
Why it all starts with visibility
The thread connecting all seven use cases isn't the technology itself — it's visibility into what happens in every sales conversation. Without structured call data, the smart coach has no material to work with, scoring runs on incomplete signals, and win/loss analysis depends on what someone remembers to write in the CRM.
Performy runs automatically over 100% of commercial conversations — capturing, structuring, and evaluating each interaction against the team's Playbook — so that when the moment comes to activate any of these seven use cases, the data is already available and structured from day one.
