A clerk before a co-pilot

Key findings

1. 01

   Selling is still the smaller part of a seller's week. In the latest survey of 4,050 sales professionals in 22 countries, sellers spend 40% of their time selling¹.

2. 02

   The first thing sellers ask AI for is record keeping. Among 1,000 sales and marketing professionals who use a CRM, automated data entry is the single most requested AI capability, named by 15%².

3. 03

   Few would let an agent act unseen. 16% of 220 B2B sales professionals would let an agent take a sales action without reviewing it first³. Three wider surveys put the share who accept AI working unchecked between 10% and 24%⁴⁵⁶.

4. 04

   Trust thins as the work gets closer to a decision. Sellers use AI most for customer research (43%) and least for forecasting and pipeline analysis (28%)⁷.

5. 05

   Leaders trust AI more than the people doing the work. 21% of sales leaders would skip the review, against 12% of individual contributors³.

6. 06

   Trust grows with use, and stays watchful. With one coding agent, about 20% of new users' sessions and over 40% of experienced users' sessions ran without asking for approval, and the experienced users stopped the agent more often⁸.

7. 07

   The measured gains are in conversations with customers. A chatbot answering shoppers before purchase raised an online retailer's sales by 16.3%⁹, and an assistant raised support agents' issues resolved per hour by 15%¹⁰. We found no controlled study of B2B sellers working in a CRM.

8. 08

   Smaller firms start further back. 24.7% of small EU enterprises use CRM software, against 65.4% of large ones¹¹, and 17% use AI, against 55%¹².

Where does a seller's week go?

Most of it does not go to selling. In the 2026 edition of the largest recurring survey of sales professionals, sellers put the share of their time spent selling at 40%¹. The 2024 edition reported 70% of time on non-selling tasks¹³, and the 2022 edition 28% of the week spent selling¹⁴.

The three editions asked different people in different countries, so the rise is a direction, not a trend line. In the same years the share of sales teams using AI passed four in five¹³¹, and more than half of the week still goes elsewhere.

Much of the rest is keeping records. In a survey of 1,000 sales and marketing professionals who use a CRM, 38% said they spend most of their time logging calls, emails and meetings, and 11% counted closing deals among their most frequent weekly activities². Among 220 B2B sales professionals, 80% said they lose at least a tenth of their selling time to finding, cleaning or updating data, and 63% use three to five separate tools to research and reach a single prospect³.

What do sellers ask AI for first?

The paperwork. Automated data entry was the single most requested AI capability among 1,000 sales and marketing professionals who use a CRM, named by 15%². A top answer of 15% also says the wishes are spread across many capabilities, and the release does not print the full list.

• Research and draftingSellers expect agents, once fully in place, to cut the time they spend on prospect research by 34% and on drafting emails by 36%¹. These are expectations, not measurements.
• One place to look79% say they must switch between several tools to get a full view of a customer account².
• AI they can find14% of those whose CRM includes AI never use it: 8% because they do not know the features exist, and 6% by choice².

What sellers already use AI for is reading up and writing. Of 500 sales people asked in another survey, 81% use AI for some part of the job: 43% for customer research, 39% for content, 35% for summaries of calls and meetings, and 28% for forecasting and pipeline analysis, the lowest of the activities listed⁷.

Work in general shows the same order. Among US workers who had used a chatbot for work, the most common uses were research (57%), editing written content (52%) and drafting it (47%)¹⁵. In work conversations with one AI assistant, sampled between April and June 2026, the most common result was a document or a report (20%), ahead of an explanation (9%) and an email draft (7%)¹⁶.

Firms point to the same place. Among US firms that used AI between November 2025 and January 2026, sales and marketing was the most common business function, at 52%¹⁷. Among EU enterprises that used AI in 2025, marketing or sales was the most common purpose, at 34.7%¹².

Where does trust stop?

At the point where an agent acts alone. In a survey of 220 B2B sales professionals, 16% said they would let an agent take a sales action without reviewing it first³. Half said they would personally be blamed if an agent acted on bad data, and 14% said an agent could act on their current sales data as it is, without a cleanup first.

Other surveys, asking other people other questions, land in the same narrow band. Among nearly 6,000 knowledge workers in nine countries, 10% said they trust AI to operate autonomously today⁴. Among 942 owners of US businesses with 2 to 249 employees, 22% were completely confident that AI can do low-level business tasks without human supervision⁵. Among 2,950 decision-makers in three regions, 24% were comfortable with agents working in the background without people knowing⁶.

The same people are not against agents. Three in four of those decision-makers were comfortable with an agent working alongside them, and 30% with an agent managing them⁶. Of the knowledge workers, 77% said they will come to trust AI to operate autonomously, 54% trust people and AI to do most work together, and 63% said more human involvement would build their trust⁴.

Wider studies of work draw the line in the same place. Across 844 tasks in 104 occupations, 1,500 workers were positive about automating 46.1% of the tasks, most often to free time for higher-value work, the reason chosen in 69.4% of those cases¹⁸. The level of involvement they chose most often, in 47 of the 104 occupations, was an equal partnership with the AI. Of the 28% who voiced concerns, 45% named a lack of trust in the system's accuracy or reliability. In a survey of 5,273 employed adults in the United States, 52% said they were worried about how AI may be used at work, and 36% said they were hopeful¹⁵.

An older experiment says why the look matters. In three forecasting studies with money at stake, people were considerably more likely to use an imperfect algorithm when they could modify its forecasts, and they did better for it. How much they were allowed to change made little difference: what they wanted was some control over the result, even a slight amount¹⁹.

What would people hand over, and what would they keep?

No survey we found asks sellers this task by task. The closest evidence is about knowledge workers in general, and about what sellers do with AI today.

Asked which work they trust AI to do with no person involved, knowledge workers most often named writing code (15%), uncovering insights in data (13%), written communications (12%) and acting as a personal assistant (12%). Asked which work they trust people alone to do, they named being inclusive (47%), onboarding and training employees (46%) and keeping data safe (40%)⁴. Taken together, they said they already trust AI with about 43% of their work tasks.

Decision-makers in another survey ranked their trust in agents by kind of work: highest for IT support and skills development, lowest for hiring, finance and legal matters⁶. Sellers' own use follows the same slope: most for research and content, least for the forecast and the pipeline⁷.

Our reading: the line follows what a mistake costs and whether it can be taken back. A wrong summary is corrected in a minute. A wrong forecast, a wrong price or a message a customer has already read is not. None of these surveys tested that explanation; it is ours.

What agents do today fits it. In one AI developer's study of the agents built on its models, published in February 2026, 73% of the agents' actions appeared to have a person in the loop in some way, and 0.8% appeared to be irreversible. Software engineering made up nearly half of those actions; sales was one of several fields at no more than a few percent each⁸.

A look only helps when it is used. A 47-country study of more than 48,000 people reports that 66% rely on AI output without evaluating its accuracy and that 56% have made mistakes in their work because of AI; its summary does not say of which group²⁰. Asking for a review does not mean a review happens.

Who trusts it more?

By role: leaders more than the people doing the work. Among the 220 B2B sales professionals, 21% of leaders and 12% of individual contributors would let an agent act without review³. Among knowledge workers, leaders said they trust AI to do 51% of their work, and rank-and-file workers 40%⁴. Among US employees who use AI, 21% of leaders and 13% of individual contributors said it has had an extremely positive effect on their productivity²¹.

One reading is that caution sits with whoever cleans up. In the same sales survey, 61% of individual contributors said their tools are still held together by manual workarounds, against 43% of leaders³.

By size of company: the question has not been put to sellers. We found no survey that compares sellers in small and large companies on what they would let an agent do unreviewed. The nearest figure is about owners, not sellers. Among owners of US businesses with 2 to 249 employees, 22% were completely confident in unsupervised AI for low-level tasks, and 31% named distrust of AI's accuracy as a barrier to using it more⁵. We found no matching figure for large companies, so the evidence does not say which end trusts more.

By experience: the line moves, and attention moves with it. Among decision-makers still exploring agents, 36% trust their organization to use them responsibly; among those further along, 95%⁶. Usage data shows the same shift. With one coding agent, about 20% of new users' sessions ran without asking for approval, and over 40% of experienced users' sessions. The experienced users also stopped the agent more often: in about 9% of its turns, against 5% for new users⁸. They had moved from approving each step to watching and stepping in.

Where has AI in sales work been measured?

Where AI talks with customers, and where it works beside an agent. Four studies with a comparison group come close to a seller's work.

• Customer supportAmong 5,172 support agents at one company, an AI assistant raised the number of issues resolved per hour by about 15% on average. Less experienced agents gained in both speed and quality. The most experienced gained little speed and lost a little quality¹⁰.
• Selling by telephoneAt a telemarketing company selling credit cards, an AI caller made the first contact and passed interested customers to 40 sales agents. Across 3,144 customers, the agents who worked after the AI were 2.33 times as successful at answering questions outside their training, and their customers were almost twice as likely to buy. The gain was much larger for the top-ranked agents than for the bottom-ranked²².
• Online retailOn one cross-border retail platform, seven uses of generative AI were tested on millions of shoppers in 2023 and 2024. A chatbot answering questions before purchase raised sales by 16.3%. Refined search queries and AI-written product descriptions raised them by 2.9% and 2.1%. AI-written push messages and advertising titles had no detectable effect⁹.
• Coaching sellersIn a series of field experiments with sales agents, an AI coach helped middle-ranked agents most. Bottom-ranked agents were overloaded by its feedback, and top-ranked agents resisted it. An AI coach working with a human manager did better than either alone²³.

Office work shows time saved, and little else so far. In a six-month experiment with 7,137 knowledge workers at 66 firms, the 80% who used an AI tool built into their email, meetings and writing spent two fewer hours a week on email, and the mix of their tasks did not change²⁴. In an online experiment, 453 professionals given a chatbot finished writing tasks in 40% less time, with output rated 18% higher²⁵. At a consumer goods company, among 791 professionals working on new product ideas, one person with AI matched a two-person team without it²⁶. And in Denmark, a study that linked surveys of 25,000 workers in 11 occupations, marketing and customer support among them, to official records found no effect on earnings or recorded hours two years after chatbots arrived, ruling out effects larger than 2%²⁷.

For a B2B seller working in a CRM, we found no controlled study. What is published is of three kinds. One is a software company's account of its own sales force: 9.4% more revenue per seller and 20% more won deals for a group of sellers using its assistant, with no control group described²⁸. Another is a peer-reviewed survey of sales representatives at one health-care company, where those who used generative AI more also reported a more effective sales process, more efficient administration and better sales results; the answers are self-reported and from one firm²⁹. The third is what sales leaders say: 89% of 100 leaders whose teams already use agents report a positive effect on sales growth³⁰, and 83% of sales teams with AI report revenue growth against 66% of teams without¹³. None of the three separates what AI did from which teams chose to use it.

The controlled studies agree on one more thing: the gain is uneven. It went to the newer agents in support¹⁰, to the strongest in telephone sales²² and to the middle in coaching²³. Who gains depends on which part of the work the AI takes over.

Where do companies start, by size?

Smaller firms start further back. The surveys above asked people who already work in a sales team and already use a CRM. Official statistics show how many firms that leaves out. In the European Union in 2025, 24.7% of small enterprises used CRM software against 65.4% of large ones, and 41.1% against 88.7% used ERP software¹¹. In the same year 17% of small enterprises and 55% of large ones used an AI technology¹².

In the United States, 18% of firms used AI in at least one business function between November 2025 and January 2026. Weighted by employment the share was 32%, which suggests that larger firms use it more¹⁷.

Generative AI has shortened the distance. In a survey of over 5,000 small and medium-sized enterprises in seven countries in late 2024, 31% used it, and 65% of those said it had improved their employees' performance. Among those not using it, 57% said it did not suit their work, 54% named copyright, legal or regulatory concerns, and 50% a lack of skills³¹. Among EU enterprises that had considered AI and not taken it up, the most common reason was a lack of expertise, named by 70.9%¹².

For a company without a CRM, the first question may not be which AI its CRM has, but whether its customers and deals are in one place at all.

So what do sellers need?

Read together, the evidence gives five answers, in the order a seller would meet them.

• The records kept for themLogging calls, emails and meetings is where 38% say most of their time goes, and it is the first thing they ask AI for².
• The research done and a first draft writtenIt is what sellers already use AI for⁷, and where the measured time savings are²⁴²⁵.
• A look before anything is done in their nameAbout one in six would go without it³. A little control is enough to make people use an imperfect system¹⁹.
• A line they can moveTrust differs by task, by role and with experience⁴⁸. One switch for a whole team fits none of the three.
• Customers and deals in one place first14% say an agent could act on their sales data as it is³, and one small EU enterprise in four uses a CRM at all¹¹.

What this does not show

• A third of the sources are vendor surveys. Ten of the 31 were run or paid for by companies that sell to the people they asked. We sell business software too. Read their numbers, and our reading of them, with that in mind.
• People described their own time. Every figure on how a week is spent is self-reported. Nobody measured it.
• Editions are not a trend. The 2022, 2024 and 2026 surveys asked different people in different countries.
• The trust figures answer different questions. The band of 10% to 24% comes from four surveys that asked four groups four things. It shows where the answers fall, not a single measure.
• Task by task, sellers were not asked. The split between what AI may do alone and what stays with people is from knowledge workers in general, and the usage data is mostly software engineering.
• The controlled evidence is near selling, not on it. Support, consumer telephone sales and online retail were tested against a comparison group. We found no such study of B2B sellers working in a CRM.
• Two samples are small. The question closest to ours, whether a seller would let an agent act unreviewed, was put to 220 people, and the source gives no country and no dates. The survey of sales leaders asked 100.
• By size of company, there is a gap. No source compares sellers in small and large companies on what they would hand over.
• The smallest firms are missing. The EU figures cover enterprises with 10 or more persons employed. Companies of one to nine are not in them.
• Two studies are older than today's AI. The experiment on control over an algorithm is from 2018 and the one on an AI coach from 2021. They are here for what they show about people, not about current models.