AI in sales: what the evidence shows it can and can't do

Sales may be the business function where AI claims are loudest. Vendors promise "AI SDRs" that prospect around the clock, copilots that write every follow-up, and forecasts that never miss. Meanwhile, buyers are quietly doing more of their homework with chatbots before they ever speak to a rep. If you lead a sales team, work in one, or buy from one, it's hard to tell which of these changes are real.

This article sorts the evidence from the marketing. It explains the main ways AI is used in selling, walks through the strongest studies on what it actually does to revenue and seller performance, and flags the risks that the demos leave out. It ends with questions you can put to your own team or your vendors this week.

What "AI in sales" actually covers

"AI in sales" bundles together several very different tools. It helps to separate them, because they carry different evidence and different risks.

• Predictive machine learning. The older layer: models that score leads, predict which deals will close, or estimate churn. They learn patterns from historical CRM data (the customer relationship management system where sales activity is logged) and output a number or a ranking, not prose.
• Generative assistants, or "copilots." Large language models (LLMs), the text-predicting systems behind tools like ChatGPT, Claude and Gemini, used to research accounts, draft emails, summarize calls and update records. A human reviews the output before it goes anywhere.
• Customer-facing conversational AI. Chatbots and voice bots that talk directly to prospects: answering pre-sale questions on a website, qualifying inbound leads, or even making outbound calls.
• AI agents. Systems that take a sequence of actions toward a goal with limited supervision, such as finding prospects, writing and sending a personalized sequence, and booking meetings. Vendors often call these "AI SDRs," after the sales development representatives who traditionally do this work.
• AI coaching and conversation analytics. Tools that analyze recorded calls and give reps feedback on talk time, objection handling or missed questions.

A useful rule of thumb: the further a tool moves from "suggests to a human" toward "speaks to a customer on its own," the higher both the potential upside and the potential for costly mistakes.

Why sales teams are adopting it so fast

The case for AI in sales starts with a simple observation about how sellers spend their time. Salesforce's 2026 State of Sales survey of 4,050 sales professionals in 22 countries, including Canada, found that the average seller spends only about 40% of their time actually selling. The rest goes to research, data entry, internal meetings and admin³. That is vendor survey data, gathered in August–September 2025, and self-reported, so treat the exact figure with care. But the pattern it describes, where selling is a minority of a seller's week, is widely recognized.

The same survey reports that 87% of sales organizations use some form of AI and 54% of sellers say they already use AI agents, with sellers expecting AI to cut time spent on prospect research by 34% and on drafting emails and content by 36%³. Those are expected savings, not measured ones, from a company that sells AI agents to sales teams, so they're best read as a sign of intent rather than proof of results.

The more independent signal is where companies say AI is paying off. In McKinsey's 2026 global survey of 1,719 respondents across 97 countries, marketing and sales is the function where respondents most commonly report revenue gains from AI⁴. That doesn't tell you how large the gains are, and it's self-reported survey data, but it's consistent with the controlled experiments below.

In Canada, adoption is earlier-stage. Statistics Canada found that 19.2% of businesses used AI to produce goods or deliver services in the year to the second quarter of 2026, up from 6.1% two years earlier. Among those users, 28.2% reported using virtual agents or chatbots⁹. A Bank of Canada survey published in August 2026 found only 8% of firms use AI significantly in core operations, with half using it to a low or moderate degree¹⁰. For most Canadian sales teams, the realistic question isn't "should we deploy autonomous agents?" but "where do we start?"

The evidence: where AI measurably lifts sales

Surveys tell you what people believe. Randomized field experiments, where some customers or sellers get the AI and a comparable group doesn't, tell you what it causes. There are fewer of these than you might expect, but the ones that exist are informative.

Always-on pre-sale answers can move conversion. The largest recent study, by Lu Fang, Zhe Yuan, Kaifu Zhang, Dante Donati and Miklos Sarvary, ran randomized experiments across seven workflows on a large cross-border e-commerce platform in 2023–2024, covering millions of users and products⁵. The standout was a generative-AI pre-sale chatbot that answered shopper questions 24/7. Compared with a control group that saw a standard "customer service is unavailable" notice, it raised sales by 16.3%. Pairing the AI with escalation to humans produced gains of roughly 25%. Other uses were smaller: AI-refined search queries lifted sales by 2.9% and AI-written product descriptions by 2.1%, while AI-written marketing push messages and ad titles showed no statistically significant effect. Gains came mainly through higher conversion rather than bigger baskets, and return rates and ratings didn't worsen. The biggest benefits went to newer, less active and lower-spending customers.

Two caveats matter. This is a working paper, not yet peer-reviewed, and the chatbot's comparison was against no service at all outside business hours, not against a skilled human. The lesson is less "AI beats salespeople" and more "answering a buyer's question at 11 p.m. beats not answering it."

AI can match competent humans on a scripted call, until buyers find out. In a peer-reviewed field experiment with more than 6,200 customers of a financial services company, Xueming Luo and colleagues randomly assigned outbound sales calls to either an AI voice bot or human workers⁶. When the bot's identity wasn't disclosed, it was as effective as proficient workers and four times more effective than inexperienced ones at generating purchases. But disclosing that the caller was a bot before the conversation cut purchase rates by more than 79.7%. Customers who knew they were talking to a machine were curt and saw it as less knowledgeable and less empathetic, even though its objective performance was the same. Disclosing later in the call, and customers' prior experience with AI, softened the effect. The study predates today's LLMs and used highly structured calls, so it's a ceiling on a narrow task rather than a verdict on modern agents. Its central tension, between effectiveness and honesty, is still live.

Assistance helps the middle and the newcomers most. A second experiment from Luo's group tested AI coaches that analyzed sales agents' calls and gave feedback⁷. The benefit followed an inverted U: middle-ranked agents improved the most. Bottom-ranked agents were overwhelmed by the volume of feedback, and top performers resisted taking advice from a machine. Trimming the amount of feedback helped the weakest agents, and a combination of AI coach plus human manager beat either alone.

This mirrors the best-known study of generative AI at work. Erik Brynjolfsson, Danielle Li and Lindsey Raymond followed 5,172 customer-support agents given an AI assistant, and found it raised issues resolved per hour by 14% on average and by 34% for novice and lower-skilled workers, with minimal gains for the most experienced⁸. That study is in customer support rather than sales, but the tasks (live conversations, product questions, calming frustrated customers) overlap heavily with inside sales. Across these studies, a consistent picture emerges: AI tends to spread the habits of top performers to everyone else, and it helps least the people who are already best.

What buyers are doing with AI

The other half of the story is on the buyer's side of the table. In a Gartner survey of 646 business-to-business (B2B) buyers conducted in August–September 2025, 67% said they prefer a buying experience without a sales rep, and 45% said they had used AI during a recent purchase¹. That sounds like bad news for sellers. But a follow-up release from the same survey found that 69% of buyers turn to sales reps to validate insights they got from generative AI². Gartner's analyst put it directly: buyer preference for self-service should not be read "as a signal that sellers matter less"².

Read together, these numbers suggest a shift in the seller's job, not its disappearance. Buyers arrive better informed, or at least more confidently informed, since AI summaries can be wrong. The rep's value moves from supplying information toward checking it, tailoring it to the buyer's situation, and helping a buying group reach a decision. Gartner's survey is vendor research and covers B2B only, so the pattern may differ for consumer sales.

Where it goes wrong

The same features that make AI useful in sales create specific risks.

Confident errors become your company's promises. Language models can produce fluent, plausible statements that are false. In sales, a wrong answer about price, terms or capability isn't just embarrassing. It can be binding. In Moffatt v. Air Canada (2024), British Columbia's Civil Resolution Tribunal held the airline responsible after its website chatbot wrongly told a customer he could apply for a bereavement discount retroactively. The tribunal rejected the argument that the chatbot was a separate entity, finding it was part of Air Canada's website and the airline was responsible for all of its information¹². The award was small, about C$650, but the principle applies to any customer-facing sales bot.

Disclosure is a real trade-off, and hiding it isn't a safe answer. The Luo study shows that telling buyers they're talking to AI can hurt conversion⁶. The tempting conclusion, don't tell them, runs into trust, brand and legal risk, especially as transparency rules for AI interactions spread. The more durable response is to deploy AI where it's genuinely good (fast, accurate answers) and make human handoff easy, which is also what produced the largest gains in the e-commerce study⁵.

Volume isn't the same as effectiveness. AI makes it nearly free to send a thousand "personalized" emails. It does not make buyers want to read them. The e-commerce experiment found no significant lift from AI-written push messages⁵, and there is not yet, to our knowledge, rigorous independent evidence on fully autonomous outbound "AI SDRs." Most published figures come from the vendors selling them. In Canada, every commercial email or text still falls under Canada's Anti-Spam Legislation (CASL), which requires consent, sender identification and a working unsubscribe mechanism, however the message was written¹¹.

Uneven benefits can become uneven problems. If AI help lifts newer reps most while overloading the weakest and being ignored by the best⁷⁸, a single company-wide rollout can leave some of your team worse off. Top performers' resistance matters too: they are often the source of the best practices the AI is spreading.

Adoption outpaces measurement. Most of the headline figures in this space are surveys of expectations or self-reported gains. Very few companies run the kind of controlled comparison that would show whether their AI tool is adding revenue or just activity.

What this means for you

A few practical steps, whether you manage a team, carry a quota or buy software:

1. Start where buyers are waiting. The strongest evidence is for answering buyers' questions quickly, especially outside working hours. Map the moments where prospects currently get silence, and test AI there first, with a clear path to a human.
2. Use AI to lift the middle, not replace the top. Aim assistants and coaching at newer and mid-level reps, keep feedback short and focused, and pair AI insight with a manager's judgment.
3. Prepare for the buyer who has already asked a chatbot. Ask reps to open discovery calls with "What have you already found, and what are you still unsure about?" Then be the person who checks and tailors, not the one who reads the brochure.
4. Lock down what a bot is allowed to promise. Give customer-facing AI an approved source for prices, policies and product claims, log its conversations, and review a sample weekly. Assume anything it says, you said.
5. Measure with a control group. Before you scale any AI sales tool, run it for some reps or segments and not others for a few weeks, and compare conversion, deal size and customer ratings, not just emails sent or hours "saved."

Sources

1. Gartner. (2026, March 9). Gartner sales survey finds 67% of B2B buyers prefer a rep-free experience (press release, survey of 646 B2B buyers, Aug–Sep 2025). gartner.com/en/newsroom/press-releases/2026-03-09-gartner-sales-survey-finds-67-percent-of-b2b-buyers-prefer-a-rep-free-experience
2. Gartner. (2026, May 20). Gartner survey finds 69% of B2B buyers turn to sales reps to validate AI-generated insights (press release, survey of 645 B2B buyers, Aug–Sep 2025). gartner.com/en/newsroom/press-releases/2026-05-20-gartner-survey-finds-sixty-nine-percent-of-b-two-b-buyers-turn-to-sales-reps-to-validate-ai-generated-insights
3. Salesforce. (2026, February 3). State of Sales report, 2026 (survey of 4,050 sales professionals in 22 countries, Aug–Sep 2025). salesforce.com/news/stories/state-of-sales-report-announcement-2026
4. McKinsey & Company. (2026, August 25). The state of AI in 2026 (survey of 1,719 respondents in 97 countries, May–June 2026). mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai
5. Fang, L., Yuan, Z., Zhang, K., Donati, D., & Sarvary, M. (2026). Generative AI and sales productivity: Field experiments in online retail (arXiv:2510.12049, v6, June 29, 2026). arxiv.org/abs/2510.12049
6. Luo, X., Tong, S., Fang, Z., & Qu, Z. (2019). Frontiers: Machines vs. humans: The impact of artificial intelligence chatbot disclosure on customer purchases. Marketing Science, 38(6), 937–947. doi.org/10.1287/mksc.2019.1192
7. Luo, X., Qin, M. S., Fang, Z., & Qu, Z. (2021). Artificial intelligence coaches for sales agents: Caveats and solutions. Journal of Marketing, 85(2), 14–32. doi.org/10.1177/0022242920956676
8. Brynjolfsson, E., Li, D., & Raymond, L. (2025). Generative AI at work. The Quarterly Journal of Economics, 140(2), 889–942. academic.oup.com/qje/article/140/2/889/7990658 (working paper: nber.org/papers/w31161)
9. Statistics Canada. (2026, June 11). Analysis on artificial intelligence use by businesses in Canada, second quarter of 2026. www150.statcan.gc.ca/n1/pub/11-621-m/11-621-m2026010-eng.htm
10. Chawla, C., & Arnburg, C. (2026, August). Canadian businesses' use of AI: What the evidence shows. Bank of Canada, Sparks at the Bank. bankofcanada.ca/2026/08/sparks-at-bank-article-2026-20
11. Canadian Radio-television and Telecommunications Commission (CRTC). Frequently asked questions about Canada's Anti-Spam Legislation (accessed September 30, 2026). crtc.gc.ca/eng/com500/faq500.htm
12. Moffatt v. Air Canada, 2024 BCCRT 149 (February 14, 2024), as summarized in Lifshitz, L. R., & Hung, R. (2024, February). BC tribunal confirms companies remain liable for information provided by AI chatbot. Business Law Today, American Bar Association. americanbar.org/groups/business_law/resources/business-law-today/2024-february/bc-tribunal-confirms-companies-remain-liable-information-provided-ai-chatbot