Research
ResearchWhat organizations ask before buying AI training
Findings from 82 conversations with 45 organizations, February to September 2026
Scholé reviewed 82 buyer conversations held between February and September 2026, and the three needs raised most often were a pilot or trial before buying (24 of 45 organizations), training for people outside the payroll (21 of 45) and training built from existing documents (19 of 45).
Published 2026-10-01. Last updated 2026-10-01. By Scholé.
What did organizations ask for?
| Need | Organizations | Share |
|---|---|---|
| A pilot or trial before buyingHow should you structure an AI training pilot? | 24 of 45 | 53% |
| Training for people outside the payrollHow do you train customers, partners or members who are not your employees? | 21 of 45 | 47% |
| Training built from existing documentsHow do you turn SOPs and handbooks into training without an instructional designer? | 19 of 45 | 42% |
| One program for very different skill levels and rolesHow do you train a workforce with very different AI skill levels? | 16 of 45 | 36% |
| A pricing model other than list price per seatWhat does AI training cost per employee, and which pricing model fits? | 15 of 45 | 33% |
| Security, privacy and compliance reviewWhat security questions should you ask an AI training vendor? | 12 of 45 | 27% |
| Languages other than EnglishHow do you deliver the same training in several languages? | 12 of 45 | 27% |
| Measuring skills and adoption, not completionsHow do you measure whether AI training changed how people work? | 11 of 45 | 24% |
| Getting staff to use an AI tool already rolled outWe rolled out Copilot or ChatGPT Enterprise and few people use it. What now? | 11 of 45 | 24% |
| Working with the existing LMS or HR systemCan you add adaptive AI training to an existing LMS without replacing it? | 10 of 45 | 22% |
| Teaching inside the real tool, during workHow do you teach people inside the tool they are using? | 10 of 45 | 22% |
| Mandatory and compliance trainingHow do you make mandatory training adaptive instead of click-through? | 8 of 45 | 18% |
| AI training for executives | 6 of 45 | 13% |
| Course libraries that go unused | 6 of 45 | 13% |
| Role-play practice for conversations | 5 of 45 | 11% |
| AI policy, guardrails and misuse | 5 of 45 | 11% |
| No learning and development team | 5 of 45 | 11% |
| Content that goes stale | 5 of 45 | 11% |
| Frontline staff on phones | 4 of 45 | 9% |
| Employee fear of AI | 3 of 45 | 7% |
| Accuracy of generated content | 3 of 45 | 7% |
Who was in the sample?
- Conversations
- 82, of which 76 were with organizations and 6 with individual learners
- Organizations
- 45: 10 customers or paid pilots, 34 prospects and 1 channel partner
- Window
- 2026-02-26 to 2026-09-29
- Unit of counting
- One organization, counted once per need
| Sector | Organizations |
|---|---|
| Government, nonprofit and associations | 8 |
| Education and training providers | 7 |
| Software and technology | 6 |
| Professional services | 5 |
| Financial services | 4 |
| Healthcare | 4 |
| Retail | 3 |
| Other industries | 8 |
Other industries covers telecom, real estate, media, transport and business services.
How was this counted?
The sample is the 82 recorded conversations that Scholé’s chief executive or a member of its growth team held with customers, prospects, a channel partner and individual learners between 2026-02-26 and 2026-09-29. Of those, 76 were with 45 organizations and 6 were with individual learners. Conversations that were not recorded are not in the sample.
Of the 45 organizations, 10 were customers or paid pilots, 34 were prospects and 1 was a channel partner. They are based in North America and Europe.
Each organization was coded against 21 needs. A need counts once per organization, however many conversations it came up in, and only when the organization raised it. A feature Scholé demonstrated without being asked does not count.
Coding was AI-assisted and worked from written meeting summaries, not full transcripts. The 6 learner conversations are in the conversation count and are left out of the organization counts.
What can this not tell you?
- These are organizations that were already talking to Scholé, so the sample leans toward buyers interested in AI training and personalized learning. It is not a survey of the market.
- A summary compresses a conversation. A need that was raised and not written down is missed, so each count is a floor.
- One coder made one pass. Nobody has re-coded the sample to measure agreement.
- A sample of 45 organizations is small. A difference of two or three organizations between two needs is not meaningful.
- Organization names are withheld because the conversations were private.
How do you cite this?
Scholé (2026). What organizations ask before buying AI training: findings from 82 conversations with 45 organizations, February to September 2026. https://schole.ai/answers/ai-training-buyer-questions-2026
The counts may be quoted with attribution to Scholé. Think a number here is wrong? Email team@schole.ai and we will check it against the coding sheet and correct it.