We looked for the original document behind the twelve AI sales figures that circulate the most. Only four say what they are credited with, and even those come with fine print; five are cited outdated, misattributed or backwards; one is a vendor's data about its own product, and for two we could not find the original.
Why do the same AI sales figures show up everywhere?
Almost any article about AI in sales repeats the same numbers: that responding within five minutes multiplies your odds, that the sales rep barely sells, that agents are already in every company. It's the problem we covered in the AI industry lies to you with numbers and in the WhatsApp open rate: almost nobody links to the original document.
The usual path goes like this: a vendor publishes a list of statistics, cites another blog that cited a report, and at every hop something gets lost. First the year, then the country and the sample. In the end, a figure about calls becomes a figure about sales, and a Salesforce survey ends up attributed to Gartner.
That's why we looked for and read the original of each figure, and put it in one of three groups: with a current primary source that says what it is credited with; outdated, misattributed or saying something else; or a vendor's data about its own product. When we couldn't find the original, we say so and don't treat the figure as true or false.
The twelve figures at a glance
The table summarizes what we found; the detail on each one is below.
Figure in circulation | Who cites it | Actual origin | Verdict |
|---|---|---|---|
"100 times more likely to contact a lead if you respond within 5 minutes" | Listicles, presented as if it came from Harvard or MIT | A study by a software vendor with an MIT professor, covering six companies: it measures calls that reach a contact, not sales (InsideSales.com and MIT (2007)) | Outdated and misattributed |
"Companies take 47 hours to respond" | Listicles, with no source | We couldn't find the document; the closest is an average of 42 hours across 2,241 U.S. companies (Harvard Business Review (2011)) | No original |
"79% of marketing leads never convert into a sale" | Listicles, attributed to MarketingSherpa | We couldn't find the document; the MarketingSherpa figure with that percentage measures something else (MarketingSherpa (2012)) | No original |
"The sales rep spends 60% of the week on tasks that aren't selling" | Listicles; 28% and 30% of time spent selling also circulate | Survey of 4,050 sales reps in 22 countries (Salesforce (2026)); the 28% is from 2022 and the 30% is from 2024 | Holds up; the older versions don't |
"67% of sales reps will miss their quota" | Listicles, attributed to Gartner | 67% did not expect to meet it that year (Salesforce (2024)) | Misattributed |
"81% of teams that use AI sell more" | Listicles, attributed to Salesforce | 81% were using or trying AI; revenue grew at 83% of teams with AI and at 66% of teams without it (Salesforce (2024)) | Says something else |
"Teams that use AI are 3.7 times more likely to meet their quota" | Listicles, attributed to Gartner | It refers to sales reps who partner effectively with AI tools, in a survey of 1,026 (Gartner (2024)) | Says something narrower |
"89% expected benefits from agents and only 45% received them" | Listicles, attributed to Gartner | 45% say their vendors' agents don't deliver what was promised (Gartner (2025)) | Cited backwards |
"By 2028, a third of enterprise software will have agentic AI" | Listicles, with no link to the original | Prediction published alongside another: more than 40% of agentic AI projects will be canceled (Gartner (2025)) | Holds up, as a prediction |
"Only 10% use agents and 82% will add them within three years" | Listicles, with no link to the original | Companies with more than USD 1,000 million in revenue in 14 countries, none of them in Latin America (Capgemini (2024)) | Holds up, within its universe |
"The market is worth 15,010 million dollars", or 32,200 million | Listicles that put both together | Paid estimates for 2030 (MarketsandMarkets (2025)) and for 2033 (Grand View Research (2025)) | Holds up, as an estimate |
"A prospecting agent costs 83% less than a sales rep" | Listicles, linking to a vendor's blog | The 83% isn't in that blog: it has a cost per lead that is 85% lower, based on the vendor's own analysis and U.S. salaries (a vendor's blog (2025)) | Vendor data |
Response speed: studies from 2007 and 2011 cited as new
The 100x figure comes from a study that InsideSales.com, a software vendor for responding to leads, presented in 2007 with James Oldroyd, then at MIT (InsideSales.com and MIT (2007)). It used data from six companies that worked with its system and more than 15,000 leads from web forms. It measured when a call managed to reach the person and did not study sales closing: calling within 5 minutes instead of 30 multiplied the probability of contact by 100 and the probability of qualifying the lead by 21.
Along the way the figure changed owners and meaning: today it is attributed to Harvard or MIT and used as if it measured sales. When presenting its survey, a website chat vendor (2017) already wrote that after five minutes the chances of qualifying drop tenfold, a sentence that is not in the study.
What Harvard Business Review (2011) did publish is a different piece of work: in an audit of 2,241 U.S. companies, those that responded to a test lead took 42 hours on average, and whoever attempted contact within the first hour was almost 7 times more likely to qualify it than whoever did so an hour later. The "47 hours" that circulates is not something we found in any original document.
We also did not find the original of the "79% of leads that never buy," which lists attribute to MarketingSherpa without a link. The only similar thing on its site is a survey by MarketingSherpa (2012) of 1,745 marketing professionals, which uses the same percentage for something else: those who did not apply lead scoring.
The practical conclusion does not change: responding fast matters. What does not hold up is using these multipliers as a sales promise, and even less for a lead who now writes through WhatsApp, a channel that none of these studies measured.
Selling time, quota and revenue: the same report, different years
Several figures come from the State of Sales, the survey of sales reps published by Salesforce, which we partner with. The current edition (Salesforce (2026)) surveyed 4,050 sales professionals in 22 countries, including Mexico and Brazil, and says that 60% of the week goes to tasks that are not selling. It is the one we use in how much of their week a sales rep loses.
Lists keep citing as current the 28% of time spent selling from the Salesforce (2022) edition or the 30% that can be derived from the 2024 one. All three are estimates by the sales reps themselves, not measurements of their calendars.
Two more figures that circulate incorrectly come from the Salesforce (2024) edition. The one about 67% of sales reps who will not reach their quota is attributed to Gartner, but it comes from that survey: 67% did not expect to meet it that year. And the one about 81% of teams with AI selling more mixes two numbers from the same report: 81% were using or testing AI, and revenue grew at 83% of teams with AI versus 66% of those that did not use it. It is a correlation, not proof of cause.
The 3.7 times figure is indeed from Gartner (2024), but it is narrower. In a survey of 1,026 B2B sales reps, those who partner effectively with AI tools were 3.7 times more likely to meet quota: it describes a skill of the sales rep, not teams that use AI. In the same survey, half of the sales reps felt overwhelmed by the amount of technology they had to use.
AI agents: a survey read backwards and half-told predictions
The worst cited is the Gartner (2025) survey of 413 marketing technology leaders. It is repeated that 89% expected a major benefit from agents and only 45% received it. The press release says that 89% of those with agent initiatives expect major benefits and that 45% say their vendors' agents fall short of what was promised. It is almost the opposite, and it also measures marketing, not sales.
The prediction that by 2028 a third of enterprise software will include agentic AI is real (Gartner (2025)), but it travels without the other half of the press release: Gartner anticipates that more than 40% of agentic AI projects will be canceled before the end of 2027. We cover what usually happens to those pilots in your AI pilot is going to die quietly.
Something similar happens with the 10% who already use agents and the 82% who will add them. That comes from Capgemini (2024), based on companies with more than USD 1,000 million in revenue in 14 countries, none of them in Latin America. The following year, the same institute (Capgemini (2025)) found that 14% had deployed agents, barely 2% at full scale, and that 23% were in pilot. A plan is not adoption.
Market size and cost versus a sales rep: estimates and vendor data
The market figures are estimates from paid reports on agents that do the work of a sales development representative. MarketsandMarkets (2025) estimates 4,120 million dollars in 2025 and 15,010 million in 2030; Grand View Research (2025), 3,800 million in 2024 and 32,200 million in 2033. They are often placed side by side as if they contradicted each other, but they refer to different years and each one defines the market in its own way.
The cost comparison is the typical case of vendor data. You hear that a prospecting agent costs 83% less than a sales rep, with a link to a vendor's blog. In the article on that blog that we read (prospecting agent vendor (2025)) the 83% is not there: what's there is a cost per lead that is 85% lower, with no published methodology and measured against a U.S. sales rep with a base salary of USD 45,000 a year.
Why don't U.S. dollar costs work for comparisons in LATAM?
That salary is the underlying problem. Comparisons between an agent and a person are built with U.S. salaries, commissions and hourly costs, and in Mexico, Colombia or Argentina every term changes: the salary, the payroll costs the employer pays, the commission, turnover and the exchange rate at which the software is paid for. The savings percentage does not carry over.
The work changes too. Those comparisons assume cold prospecting by email, LinkedIn or phone. In Latin America a large share of leads arrive through WhatsApp, often from an ad that opens the conversation, and the task is to reply to someone who already wrote to you, not to go looking for them. That channel adds a cost that U.S. calculations don't have: Meta charges for each delivered message according to its category and the recipient's country (Meta (2026)).
And then there are the samples: the 2007 study and the 2011 audit were done with U.S. companies, and the Capgemini survey covered 14 countries, none of them Latin American. The newest edition of the State of Sales does include Mexico and Brazil, mixed in with 20 other countries.
What to do instead: measure your own baseline
The figure that helps you decide is the one from your own operation. Before comparing an agent with your team, measure for a few weeks:
First response: the real time, including outside business hours, not the time your team thinks it is.
Volume: conversations per month and the minutes it takes to handle each one.
Cost per person: salary, payroll costs, commissions and tools, in your currency.
What gets lost: unanswered leads and quotes with no follow-up.
With that data, the sales reps vs. AI agent calculator sizes how many people you need to respond on time and at what cost an agent makes sense, in Mexican, Colombian or Argentine pesos. The WhatsApp lost sales calculator estimates what responding late costs you, and the KPIs for measuring an AI agent tell you what to track afterward.
The third-party figures we use on the site are listed in data transparency, with their source, their year and whether they are an independent benchmark or vendor data. If you want to build the calculation with your own numbers, book a demo and we'll review it together.
