The Salesforce State of Sales report finds that 83% of sales teams using AI saw revenue growth, against 66% of teams not using AI. That is a 17-point gap, in the metric that matters most to a sales organisation, in a single survey cycle. Across professional categories surveyed by anyone, sales is the function showing the cleanest line between AI adoption and commercial performance.
Other parts of the data fill in the picture. Across all sales orgs, 87% are using AI in at least one task. 64% of reps say they save 1 to 5 hours a week through AI automation. The high performers in the data are 1.7x more likely than the underperformers to be using AI agents specifically for prospecting, which suggests the gap will widen as agents move from beta to production through 2026.
What sales reps are actually using AI for
Sales work breaks into a few well-defined activities, and AI has landed in each of them differently.
Prospecting and outreach is the largest pull. Drafting personalised outreach at scale, researching accounts before a call, and generating first-pass call lists have all been transformed by general-purpose models like ChatGPT and Claude, and by sales-specific tools like Salesforce’s own AI features, Apollo, and Gong. The work that used to fill the morning of a rep with research before a single call now takes 20 minutes.
Call recording and analysis is the second. Conversation intelligence tools transcribing, summarising, and scoring sales calls against a methodology have been mainstream for several years, and the generative AI layer has materially improved the quality of the post-call notes that go into the CRM. Reps who hated CRM hygiene now have a system that does most of it for them.
Forecasting and pipeline analysis is the third. AI-assisted forecasting tools surface deals that are sliding, accounts that have gone quiet, and opportunities the rep has not flagged but the data suggests they should. Sales leaders are getting earlier warning signs on weak quarters and clearer evidence on which reps are running disciplined pipelines.
Proposal and follow-up drafting is the fourth, and probably the least glamorous. Most reps now generate first-draft proposals, follow-up emails, and contract redlines through AI rather than from scratch. The savings here are unevenly distributed, with reps in long-cycle complex deals saving more than reps doing transactional work where the proposal is templated to begin with.
The agent layer is pulling ahead of the rest
Salesforce’s data point that high-performing reps are 1.7x more likely to be using AI agents for prospecting is the leading indicator worth watching. An agent in this context means software that can take an instruction (research these 50 accounts, draft personalised outreach for the top 10, schedule send for tomorrow morning) and execute the steps without the rep babysitting each one. That is the next layer above current generative AI usage, and the McKinsey State of AI 2025 data shows 23% of organisations already scaling agentic AI somewhere in their operations.
For sales specifically, the work that lends itself to agents is the prospecting and outreach work that traditionally consumed 30 to 50% of a rep’s week. The reps using agents are not getting that week back. They are using it to call more accounts, with better preparation, more often.
What’s not getting automated
Three parts of the sales role are stubbornly human.
Discovery is the first. The conversation in which a rep figures out what the customer actually wants, what they are willing to pay for, and what is going to kill the deal, is the conversation AI cannot run. Models can prepare the rep, score the call afterwards, and surface what was learned. They cannot replace the rep in the chair. The discovery skill is the most valuable thing in sales, and AI has not changed that, possibly the opposite.
Negotiation is the second. The actual back-and-forth on price, terms, and timing requires reading the room and reading the silence in ways AI cannot currently do, and that the customer would object to even if it could.
Account ownership is the third. The trust required for a customer to call you when they have a problem they have not figured out yet, and the implicit promise that you will respond as a person rather than a vendor, has not been touched by AI. Long sales cycles in enterprise are won on relationships, not on outreach quality.
The productivity gain, sourced
The Salesforce figure of 64% of reps saving 1 to 5 hours per week through AI automation is the most credible specific number in the sales-AI literature, since it comes from actual reps reporting on actual usage rather than vendor projections. That brackets to something like 5 to 15% of a 40-hour week saved on average, with bigger savings for high-volume transactional sales and smaller savings for complex strategic accounts.
Setting that next to the Harvard Business School and BCG study showing roughly 25% time reduction on consulting tasks and the GitHub Copilot RCT showing 55.8% speedup on benchmark coding, the bracket on AI-good tasks holds. The realistic productivity gain on the slices of sales work AI is good at sits in the 20 to 40% range. That gain is real but partial, since most sales work is the conversation work AI does not touch.
What the next 12 months look like
Three movements are likely to be visible by the end of 2026. First, the revenue-growth gap between AI-using and non-AI-using teams will widen further as AI matures and agentic capabilities move into production. The Salesforce 17-point gap is not the steady state. It will get bigger before it gets smaller.
Second, sales tooling consolidation will accelerate. The current proliferation of point AI tools (one for outreach, one for transcription, one for forecasting, one for research) will give way to integrated platforms. Salesforce, HubSpot, and the AI-native challengers will compete on which one has the most coherent end-to-end agent layer.
Third, the productivity divide between high-performing and low-performing reps will widen further. AI does not lift floors equally. The reps who already had good discovery, good qualification discipline, and good account knowledge will get more out of the tools than reps who hoped AI would compensate for weaker fundamentals. That is consistent with most data on AI productivity in expert work: it accelerates good practice and exposes bad practice rather than equalising the two.
If you sell for a living, the question is not whether AI is part of your role. It already is. The question is whether your team is the one widening the revenue gap or the one watching it widen on the other side.