How to choose corporate AI training
When generative AI first became widely available, I was in college. I was amazed by what the tools could do. Later, a 2025 report from Project NANDA put a sharper question to it. If companies were spending tens of billions on generative AI, why were so few pilots showing up in the financial results?
Over the past year, we’ve worked with more than 25 companies. Three lessons from that work shape how we help teams choose what to train on, give people time to practice, and measure what changes.
Three numbers that changed how we think about training
- About 5%
- of integrated AI pilots in Project NANDA’s preliminary 2025 research were reported to generate substantial value. The authors cited $30–40 billion in enterprise generative AI investment and reported little measurable profit-and-loss impact across most pilots they studied. Project NANDA, The GenAI Divide
- 28% lower
- median learning and development spending per full-time employee in SHRM’s 2026 benchmark compared with 2025. Median training time held at eight hours per employee. SHRM’s 2026 benchmark
- 32%
- of organizations in the same SHRM benchmark offered AI skills training beyond basic compliance. SHRM’s 2026 benchmark
Start with work the team already does
Most people can learn what an AI tool does. The harder part is finding where it belongs in a real workflow. We sit with the people doing the work and ask them to walk through a recent example, including where the source material lives, what slows them down, and who signs off on the result. From there we choose one task frequent enough to measure and small enough to test.
For a team writing client updates, the pilot might turn approved meeting notes into a draft with decisions, owners, and dates. The reviewer checks each item against the notes before the draft goes out. The before-and-after comparison includes both drafting and review.
Give people time to practice with their own examples
Separate participants by the work they do. A proposal team might practice using approved service descriptions to draft an answer to an RFP. An operations team might turn a handoff into a checklist. In both sessions, agree on what information can go into the tool and give people time to revise the output with a colleague.
At Converge Strategies, Ruby worked alongside employees in cohorts using Claude and Gemini. The projects included a proposal builder that drew on Converge’s service descriptions, a writing checker tied to the company’s checklist, and a way to turn approved reports into drafts for new audiences. Read the Converge case study
Measure what happens after the session
Choose a workflow owner and record a few normal examples before training. For each example, track time to a usable result, corrections needed, and whether the output passed review. At 30, 60, and 90 days, check how often staff still use the workflow, which approved tool they use, and how the new results compare.
For example, if a task falls from 45 to 30 minutes including review, 40 uses in a month free up about 10 hours. That’s an illustration of capacity. To estimate financial return, account for training, licenses, and support, then document how the business used the freed time or avoided an expense. Ruby’s guide to measuring training results
What to ask before you hire a provider
- What task will our team practice, and what should they leave with?
- How will you tailor examples to each role and handle our data rules?
- What do preparation, delivery, follow-up, and additional cohorts cost?
- Who will own the workflow afterward, and what will we measure before expanding?