What did Gartner predict about AI agents and sellers?
Gartner predicts that by 2028, AI agents will outnumber sellers by 10 times, yet fewer than 40% of sellers will say AI agents have improved productivity. Gartner made the prediction in a press release dated 28 July 2026, under the subheading that sales leaders risk AI agent sprawl unless they overhaul data, automation and user experience.
Dan Gottlieb, VP Analyst in the Gartner Sales practice, summed it up: "more agents will not automatically mean more productivity." The release makes a second prediction too: by 2028, chief sales officers (CSOs) who overhaul data, automation and user experience will be five times more likely to gain ROI from AI than those choosing quick fixes.
The release cites one survey, and it measures something different. Of 210 CSOs and senior sales executives surveyed from January through February 2026, 60% of CSOs said their revenue number is largely driven by elements outside their control. The release does not explain how the 10-to-1 figure was reached. Gartner clients can read more in its report Using AI to Drive Breakthrough Sales Productivity.
What does Gartner tell sales leaders to do?
Gartner names three actions to avoid agent sprawl, and the third one is about measurement. Here they are, paraphrased from the release, with what each means for a team of 1 to 5 people:
| Gartner's action | Gartner's description, paraphrased | For a small team (our reading) |
|---|---|---|
| Own AI-forward sales infrastructure | Build a central context layer that connects company data, systems and seller judgement, so agents produce more relevant output | Your CRM and lead list are your context layer. Clean them before an agent reads them. |
| Orchestrate winning sales behaviours | Use AI to cut admin, spread top performers' expertise and give managers data for coaching | Write down what your best outreach looks like, so an agent has a model to follow. |
| Harvest AI's impact on sales capacity | Track more than time savings: measure how AI expands seller capacity, improves effectiveness and supports commercial outcomes | Record hours, reply rate, meetings per 100 people contacted and pipeline before you start. |
Gottlieb's other warning explains why the groundwork matters: "If those systems are fragmented, the agents will scale the fragmentation." For a founder, fragmented systems can mean a lead list in one spreadsheet, replies in two inboxes and no record of what booked a meeting. An agent added on top would send more, and you still couldn't tell whether it worked.
Which numbers should you record before adding an AI SDR?
Four: lead data quality, reply rate (with positive replies on their own line), meetings per 100 people contacted, and hours spent prospecting, plus two email guardrails. Copy this table, fill in the baseline column this week, and fill in the last column 30 days after an agent starts.
| Metric | How to measure it | Baseline (dates: ____) | After 30 days |
|---|---|---|---|
| Lead data quality | Share of a 50-record sample that passes all five checks in the next section | ____ % | ____ % |
| Reply rate, by channel | People who replied ÷ people contacted, over the last 8 weeks | Email ____ % / LinkedIn ____ % | Email ____ % / LinkedIn ____ % |
| Positive reply rate | People who replied with interest ÷ people contacted | ____ % | ____ % |
| Meetings per 100 people contacted | Meetings booked ÷ people contacted × 100 | ____ | ____ |
| Hours on prospecting a week | Logged time on finding leads, research, writing, follow-ups and CRM updates, averaged over 2 weeks | ____ h | ____ h |
| Email bounce rate (guardrail) | Bounced emails ÷ emails sent | ____ % | ____ % |
| Gmail spam rate (guardrail) | From Google Postmaster Tools. Google's sender guidelines say to keep it below 0.10% and avoid ever reaching 0.30%. | ____ % | ____ % |
Write the date range next to every baseline number. A figure without dates can't be compared with anything later.
How do you check lead data quality in an hour?
Score a random sample of 50 records against five checks, then fix whatever fails most. The steps:
Take a random sample
Export your lead list or CRM view and take every nth record, for example every 10th from a list of 500, so you don't only check the names you remember.
Run five checks on each record
The company is right and still trading. The person still holds the role (open their LinkedIn profile). The email is deliverable (an email verifier, or no past bounce). The role matches your buyer. The company fits your ICP on size, industry and location.
Score strictly
A record passes only if all five checks pass. Pass rate is passes ÷ 50. Note which check failed most often.
Fix the source
If stale roles dominate, the list is old, so refresh it first. If fit fails most, tighten your targeting rules; our lead scoring framework helps you write them.
Our rule of thumb, which is advice and not a published standard: if fewer than 40 of the 50 records pass, fix the data before you add an agent. An agent will contact the failing records too, at whatever volume you set.
How do you set a reply-rate and meetings baseline?
Use people contacted as the denominator for both numbers, and keep each channel separate. Reply rate is people who replied ÷ people contacted. Meetings per 100 is meetings booked ÷ people contacted × 100. Counting people instead of messages stops follow-ups from inflating the base.
A worked example: in the last 8 weeks you emailed 240 people, 14 replied, 8 of them with interest, and you booked 6 meetings. Reply rate is 14 ÷ 240 = 5.8%, positive reply rate is 8 ÷ 240 = 3.3%, and meetings per 100 is 6 ÷ 240 × 100 = 2.5.
Small samples swing. At 100 people contacted, a single reply moves the rate by a full percentage point. Use at least 8 weeks of data, and a longer window if you contacted fewer than 200 people.
For a rough outside reference, AiSDR gives a rule of thumb on its pricing page of a 1 to 3% conversion from cold prospect to booked meeting, or 1 to 3 meetings per 100 contacts. AiSDR says this is based on the historical performance of teams using it and varies with market fit and offer; the page gives no date range. Treat a vendor's figure as context and your own baseline as the test. Our look at whether AI SDRs work covers how to test one fairly.
How do you measure time and capacity?
Log hours in five buckets for two weeks before an agent starts, then log what the freed hours turned into. That follows Gartner's third action, which asks for measures of capacity, effectiveness and outcomes beyond time saved.
| Task | Hours a week, before | Hours a week, with an agent |
|---|---|---|
| Finding and listing leads | ____ | ____ |
| Researching each person | ____ | ____ |
| Writing first messages | ____ | ____ |
| Follow-ups and replies | ____ | ____ |
| CRM updates and admin | ____ | ____ |
| Sales calls and demos held | ____ | ____ |
The last row is the capacity number. If the first five rows fall and the last one doesn't rise, the agent saved time without adding selling. Reviewing what an agent sends also takes time, so log it under follow-ups and replies.
If you're weighing an agent against a hire, the same log shows how much work there is to hand over; our AI SDR vs human SDR comparison goes through it task by task.
How do you run a fair 30-day test?
Change one thing, the agent, and compare the result with your baseline table. Our suggested setup:
- Keep the rest fixed. Same ICP, same offer and same channels as the baseline period.
- Check leads before anything sends. Run the five checks on 50 leads the agent found. Siela's AI SDR, for example, has a Find leads only mode that saves leads to the CRM with the reason each was picked and sends nothing.
- Keep a comparison. Go on sending your own outreach to part of the list, so you have a same-month result to compare with.
- Decide on day 30. Fill in the last column. Keep the agent if meetings per 100 people contacted held or rose and your prospecting hours fell. If replies dropped, check data quality and targeting before blaming the writing.
Gartner's release is aimed at chief sales officers, but we think Gottlieb's point holds at any size: an agent scales what is already there. The baseline you record this week tells you whether it multiplied the right things.
Frequently asked questions
What did Gartner predict about AI agents outnumbering sellers?
In a press release dated 28 July 2026, Gartner predicted that by 2028 AI agents will outnumber sellers by 10 times, yet fewer than 40% of sellers will say AI agents have improved productivity. It also predicted that CSOs who overhaul data, automation and user experience will be five times more likely to gain ROI from AI than those choosing quick fixes.
What does Gartner mean by AI agent sprawl?
In Gartner's 28 July 2026 release, Dan Gottlieb, VP Analyst in the Gartner Sales practice, says that without the right data foundation, workflow integration and seller experience, CSOs risk creating agent sprawl: more digital activity with little improvement in seller impact. The release says sales leaders risk agent sprawl unless they overhaul data, automation and user experience.
How do you calculate meetings per 100 leads contacted?
Divide meetings booked by the number of people contacted in the same period, then multiply by 100. If you contacted 240 people over 8 weeks and booked 6 meetings, that is 6 ÷ 240 × 100 = 2.5 meetings per 100. Count each person once, however many follow-ups they received, and keep each channel separate.
What reply rate should I record before trying an AI SDR?
Record your own reply rate for the last 8 weeks: people who replied divided by people contacted, separately for email and LinkedIn. Record positive replies on their own line. Before comparing with any published benchmark, check how it defines reply rate and what it counts as a reply. Your own number is the one an agent has to beat.
How many leads should I sample to check data quality?
We suggest a sample of 50 records, which takes about an hour to check. Take every nth record from your list, check company, current role, email deliverability, role fit and ICP fit, and count a record as passing only if all five checks pass. Our rule of thumb is to fix the data if fewer than 40 pass.
Sources
- Gartner press release: Gartner Predicts AI Agents Will Outnumber Sellers 10 to 1 by 2028, Yet Fewer Than 40% of Sellers Will Say Agents Improved Productivity (published 28 Jul 2026)
- Google Gmail Help: Email sender guidelines (checked 20 Oct 2026)
- AiSDR: pricing page (FAQ on meetings per 100 contacts) (checked 20 Oct 2026)