AI training: only 36% say they were trained enough (BCG)

Only 36% of workers say they received sufficient AI training, according to the fourth edition of BCG’s AI at Work survey, published in June 2026 with 11,749 respondents in 14 markets, including 1,000 in Brazil. The figure is identical to 2025, and over the same period regular AI use among frontline employees rose from 51% to 74% (BCG, 2026). The survey covers leaders, managers and white-collar employees without managerial duties, and the percentage measures each person’s view of their own training. Read next to data from the OECD, the World Economic Forum, Microsoft and LinkedIn, it shows companies that spread the tool faster than they trained people, and those who received more hours of training, with practice and follow-up, use AI far more.

At Iasbeck&Co we implement CRM and AI agents in B2B operations, and I have trained more than 10,000 professionals. I wrote this for the board, for sales, marketing and CS leaders, and for HR, who decide how much to invest in training and how to tell whether it worked.

How many employees have received AI training?

The answer depends on the question each survey asked, and BCG’s three most recent editions show the trend with a similar method. In 2024, BCG asked who had already been trained on how AI will change their own job: 50% of leaders, 30% of managers and 28% of frontline employees said yes, up from 44%, 20% and 14% the year before (BCG, 2024). The sample was 13,102 people in 15 countries and regions. In 2025 and 2026 the question shifted to whether the training was enough, and the answer “yes, my training was enough” stayed at 36% in both editions (BCG, 2025).

Of all respondents, 88% believe they will need major upskilling in the next five years, and 72% say AI has already changed the skills expected in their role (BCG, 2026). Regular use, defined as daily or several times a week, reached 74% on the frontline, 88% among managers and 93% among leaders. Microsoft and LinkedIn found a similar share of AI users trained by their company in 2024, and I discuss that figure in AI training for companies.

BCG AI at Work	Frontline regular use	Training question	Result
2024 (13,102 respondents)	52%	Have you been trained on how AI will change your job?	28% of frontline, 30% of managers, 50% of leaders
2025 (10,635 respondents)	51%	Was your training enough?	36% say yes
2026 (11,749 respondents)	74%	Did you receive sufficient training?	36% say yes, unchanged

In the 2026 edition, among frontline employees, sales and business development shows 68% regular use, below the 74% average and ahead only of operations and production, at 61%; IT and marketing lead with 88% and 85%. By country, Brazil shows 82% frontline regular use, the fourth highest of the 14 markets. The survey does not break down by country the share who say they were trained enough.

Which surveys measure AI training?

BCG and the OECD bring the data closest to training itself, the World Economic Forum surveys employers, and Microsoft and LinkedIn are surveys by companies that sell AI tools and courses.

Source	Year	Sample and method	Main finding
BCG, AI at Work	2024 to 2026	Questionnaire with leaders, managers and white-collar employees; 11,749 people in 14 markets in 2026	36% say they received sufficient training in 2025 and 2026; regular use grows with hours of training
OECD, survey of employers and workers	2023 (fieldwork in 2022)	5,334 workers and 2,053 firms in finance and manufacturing in seven countries	58% of AI users in finance and 53% in manufacturing received training provided or funded by their company
OECD, Bridging the AI skills gap	2025	Analysis of course catalogues in Australia, Germany, Singapore and the US	Between 0.3% and 5.5% of the courses analysed have AI content, most of it advanced
World Economic Forum, Future of Jobs	2025	Over 1,000 employers, 14 million workers, 55 economies	77% plan to reskill their workforce to work alongside AI
Microsoft, Work Trend Index (vendor survey)	2025 and 2026	31,000 knowledge workers in 31 markets (2025); 20,000 AI users in 10 markets (2026), online, by Edelman Data x Intelligence	Organizational factors such as culture and manager support weigh more than twice as much as individual factors in reported impact
LinkedIn, Workplace Learning Report (vendor survey)	2025	937 HR and L&D professionals and 679 learners, Brazil included	49% of L&D professionals say their executives worry employees lack the skills to execute the strategy

How many hours of AI training make a difference?

The most useful data point for program design is in a chart from BCG’s 2025 edition, with 10,635 respondents. People who received no training are regular users in 18% of cases. With one to five hours, the share rises to 63%; with five to ten hours, to 82%; with more than ten hours, to 89%. The same chart shows 79% regular use among those who had in-person training, against 67% among those who did not, and 84% among those with access to a coach, against 70% among those without (BCG, 2025).

Training condition (BCG 2025)	Regular AI users
No training	18%
1 to 5 hours	63%
5 to 10 hours	82%
More than 10 hours	89%
No in-person training / in-person training	67% / 79%
No coach / access to a coach	70% / 84%

The biggest jump happens between zero and five hours, and the second between the one-to-five band and the five-to-ten band. A one-hour talk leaves a person in the bottom band. The chart’s reading note has an inconsistency: it assigns 79% versus 67% to the comparison of more and less than five hours, figures that in the bars belong to in-person training. I used the bars.

The OECD reached a similar relationship by another route. In its 2022 survey of finance and manufacturing workers in seven European and North American countries, more than half of AI users said their company had provided or funded training so they could work with the technology: 58% in finance and 53% in manufacturing. Trained workers more often reported positive effects of AI on performance and working conditions, with gaps of up to 39 percentage points in mental health and well-being in finance (OECD, Lane, Williams and Broecke, 2023). The authors warn that causality cannot be established and note that trained workers were also more worried about losing their jobs.

What companies say they will do on AI upskilling

In the Future of Jobs 2025, 77% of employers surveyed by the World Economic Forum plan to reskill and upskill their current workforce to work better alongside AI, the most cited strategy in response to the technology, ahead of hiring people to design AI tools (69%). The same employers name skills gaps in the labor market as the main barrier to business transformation, cited by 63%, and report that 50% of the workforce has completed training as part of company learning strategies, up from 41% in the 2023 edition (World Economic Forum, 2025). The report’s figures on how many workers will need reskilling by 2030 are in AI and jobs.

In Microsoft’s 2025 survey of 31,000 knowledge workers in 31 markets, 47% of leaders list upskilling existing employees as a top workforce strategy for the next 12 to 18 months. Among managers, 51% say AI training or upskilling will become a key responsibility for their teams within five years, and 35% are considering hiring AI trainers to guide adoption (Microsoft, 2025). In LinkedIn’s 2025 report, which surveys L&D professionals, 49% agreed that their executives worry employees do not have the skills to execute the business strategy, and 71% said they are already exploring or using AI in their own L&D work (LinkedIn, 2025).

AI literacy: the course supply still targets the specialist

The OECD looked at the other side of the counter, the supply of courses. In a study published in April 2025, it analysed training catalogues in Australia, Germany, Singapore and the United States and found AI content in between 0.3% and 5.5% of courses. Most of these programs cover advanced skills, such as building and maintaining models, while only about 1% of jobs highly exposed to AI require that kind of complex skill; the vast majority of exposed workers need AI literacy, the general understanding needed to use and check the technology (OECD, 2025). The OECD admits the figures may be underestimates, since they miss learning inside companies, and even so the open course market offers little of what a sales rep or a support agent needs.

Why does the manager matter more than the course?

In the 2026 Work Trend Index, Microsoft tested individual, organizational and demographic factors against the impact that 20,000 AI users in 10 markets, Brazil included, say they perceive, and concluded that culture, manager support and talent practices account for more than twice the reported impact of individual factors such as mindset and behavior (67% versus 32%). In a separate study of 1,800 workers cited in the same report, when managers visibly modeled AI use, employees reported a 17-point lift in perceived value (Microsoft, 2026). The most advanced professionals in the sample more often say their manager uses AI openly (85% versus 64%) and sets quality standards for AI work (83% versus 57%).

BCG came close to this on the frontline. In 2025, among 3,537 employees without managerial duties, only 25% said they had received enough support from leadership on how and when to use AI, and among those who had that support regular use reached 82%, against 41% among those who did not (BCG, 2025). In the 2026 edition, 66% of frontline regular users say they get limited or no guidance on what to do with the time AI saves (BCG, 2026). The sales rep saves an hour on call summaries, nobody says whether it goes to prospecting or to the account base, and it disappears into the routine.

What these data do not say

Nearly all the numbers are self-reported. “I feel properly trained” measures perception, and someone who took ten hours of courses may answer no because the bar moved; someone else, with one hour, may say yes because they use little AI. The questions also change from one survey to another: BCG in 2024 asked about training on how work will change, in 2025 and 2026 about whether it was enough, and the OECD about training provided or funded by the company. These percentages neither add up nor form a series.

The link between training hours and use is a correlation. People who already use AI often seek more training, and companies that train more tend to be the ones that gave access to the tool first; BCG’s chart does not separate these effects, and the OECD warns that causality cannot be established. The samples are also limited: BCG surveys only white-collar workers, leaving out factory floors and stores, the OECD covered only finance and manufacturing in seven rich countries, with fieldwork in 2022, before generative AI tools became widespread, and Microsoft’s 2026 survey covered only people who already use AI at work.

Microsoft and LinkedIn sell AI tools and courses, and BCG sells transformation projects. None of the sources measures Brazil separately on the training indicator, and none follows the same group of people before and after a training program to measure the effect on their work. Only your company can produce that data.

What an AI training program needs, according to the data

The surveys do not design a curriculum, and they point to conditions associated with higher use. I crossed those conditions with what I see in the sales operations we work with.

What the data show	Source	What it asks of the program
Regular use rises from 63% to 82% when training passes five hours	BCG 2025	More than five hours per person, spread over weeks, with practice between sessions
In-person and coaching show 12 and 14 more points of regular use	BCG 2025	Group practice on real cases from the function and someone to handle questions after the session
Leadership support doubles frontline regular use	BCG 2025, Microsoft 2026	Managers trained before their teams, using AI visibly and setting quality standards
66% get no guidance on the time saved	BCG 2026	A written leadership decision on where freed-up time goes in each role
Most people need literacy, and the open supply focuses on advanced skills	OECD 2025	Literacy for everyone and practice by function, with specialization only for those who configure agents

In practice, that means separating the sales rep from the marketing analyst and the support agent and bringing the week’s deals, campaigns and tickets into the room, with the rules taught alongside the technique. In 2025, 54% of BCG respondents said they would use unauthorized AI tools if the company did not provide the right ones, and training that ignores the AI use policy pushes that use to personal accounts. The design by level and role, with the deliverables of each cohort, is in AI training for companies.

How to measure the results of AI training

BCG’s questions work for HR as an internal survey, applied before the first session and repeated 30 and 90 days later, to compare the company with the global 36% and with itself. The perception question alone misleads, so it sits next to process-use indicators that the CRM or the tool already records, such as call summaries approved without edits or tickets classified without correction.

Cross the answers by function and level. If the sales team shows high use and “no” on question 3, training is behind the tool; if it has training and “no” on question 4, the bottleneck is leadership. Financial return comes later, with the logic of AI ROI.

What leadership does with these numbers

1. Measure the gap in your company. Run the five questions by function and level before hiring any training, and record the results with a date.
2. Train managers first. Microsoft and BCG data tie team use to manager behavior, and a manager who does not use AI does not know what to ask for in the pipeline review.
3. Plan more than five hours per person. Spread the load across short sessions with practice on the week’s work, instead of a full day of class.
4. Include follow-up after the session. Decide who answers questions in the following weeks, whether an advanced user in the function or an external coach.
5. Decide where the time saved goes. Write down by role what changes in targets or accounts when AI frees up hours, and tell the team.
6. Teach the rules with the tool. Bring the use policy into the exercises, with the data that can and cannot go into each tool.
7. Repeat the measurement at 30 and 90 days. Compare perception and process use with the baseline and adjust the program before expanding to other functions.

For HR and L&D

The numbers give HR an argument to move AI training out of the “courses” budget and into the operating plan: the board understands the comparison between 18% regular use without training and 82% with five to ten hours, with the caveat that it is a correlation. The request that usually reaches L&D is “an AI course for everyone”, and the answer the data support is a program by function, with managers first, enough hours, follow-up and measurement before and after.

The gap between use and training shows up in every survey, and each company closes it with people trained on their own processes, with their managers alongside. That is the work of the in-company applied AI training, with leadership, sales, marketing and CS cohorts built on the company’s real cases, which delivers a prompt library by function and an agent architecture, covering what each agent reads, what it executes on its own and who approves the rest.

Frequently asked questions

How many employees receive AI training?

In BCG’s AI at Work survey, only 36% of the 11,749 respondents in 2026 said they had received sufficient AI training, the same share as in 2025. In 2024, the same survey showed that 28% of frontline employees had been trained on how AI would change their job, against 50% of leaders.

How many hours of AI training are needed?

BCG’s 2025 data show regular use rising from 63% among those with one to five hours to 82% among those with five to ten, and 89% above ten hours. The relationship is a correlation, and even so five hours per person, spread out with practice, works as a reasonable floor for a program.

What is AI literacy?

It is the general understanding needed to use AI at work: what the tool does, when it gets things wrong, how to ask and how to check the result. The OECD estimates that most workers exposed to AI need this literacy, and only a small minority need advanced technical skills.

How do I prepare the sales team to use AI?

Start with managers, who need to use AI visibly and set the quality standard. Then train sales reps, marketing and support in separate cohorts, on the week’s real deals, campaigns and tickets, with follow-up in the following weeks and a clear decision on how the time saved will be used.

Does AI training increase tool use?

BCG and OECD surveys show that people who received training use AI more and report more positive effects. Neither proves causality, because people who already use AI also seek more training, so it pays to measure use in your company before and after the sessions.

How does HR measure AI training?

Run a short survey before the first session, covering frequency of use, hours received, whether training was enough and manager support, and repeat it at 30 and 90 days. Add process-use indicators, such as drafts approved without edits and corrected classifications, recorded in the CRM or the tool.