Let's Connect
Workflow adoption

Workflow-First AI Training: Fixing Adoption After the Workshop

Diagnose why AI training has not changed daily work, then design relevant practice, manager support and office hours around the actual bottleneck.

7 min read Updated

An employee can enjoy an AI workshop and return to the same way of working the next morning. They may understand the tool, remember a useful prompt and still have no clear reason to use it on the task waiting in their inbox.

When that happens, start by watching the work. Ask the employee to walk through a recent task, including the parts between the obvious steps: finding the right document, clarifying a request, checking an answer and waiting for approval. The next training session should address the difficulty that observation reveals.

At Ajaia, discovery regularly changes the work a program is designed around. A team may ask us to make one piece faster, then show us a larger source of effort elsewhere in the process. That is a reason to adjust the curriculum before teaching it.

The first request may describe only part of the problem

Imagine a team asks for help drafting weekly updates. This is a fictional example. During discovery, the facilitator learns that writing takes ten minutes; collecting reliable information from several project owners takes much longer. A writing workshop might produce smoother updates while leaving the main delay untouched.

Walk through the latest completed update with its owner. Where did the information come from? Which dates were tentative? Which details required clarification? What happened when two sources disagreed? Record the answers before deciding which step AI should support.

The resulting exercise might teach the team to organize approved notes, identify missing information and draft questions for project owners. The final update would come later. If the information is inaccessible because of permissions or a missing integration, assign that dependency to its owner. A more elaborate prompt cannot fix access.

Use a short workflow record: the trigger, source inputs, steps, output, reviewer and recurring difficulty. The AI workflow mapping guide can help a program owner investigate the full process.

Give employees a reason to care about the task

The sponsor may want higher utilization of an enterprise subscription. The employee wants to finish a difficult piece of work, avoid a mistake or feel better prepared for a meeting. Explain the benefit in those terms.

For example: "This exercise will help you find unanswered questions in project notes before you send an update." That is a benefit someone can test. "AI will transform your productivity" gives them little guidance for tomorrow morning.

Ask what concerns them, too. An employee may worry that using AI looks like cutting corners, that an incorrect answer will damage their credibility, or that they will ask a basic question in front of a confident colleague. Those concerns may remain unspoken in a large session. Private questions and smaller practice groups make it easier to find out what is getting in the way.

Avoid promising that AI will make every task easier. Show the employee a bounded task, what they remain responsible for and how to check whether the method helped. A poor result can lead to a useful decision to change the method or keep that task manual.

Make the learner do the work during the session

A polished demonstration shows what a facilitator can do. To assess learning, the participant needs to produce and review an output.

In Ajaia's live learning format, participants log into an online portal, access the material, answer questions and practice alongside the demos. The facilitator can use those questions to slow down, clarify a step or work through a difficulty. Agree on the portal features and follow-up access included in the program before scheduling a cohort.

For the fictional project-update task, give learners a short set of approved notes. Ask them to separate confirmed actions from unresolved decisions, identify missing owners and draft a brief update. Include a tentative date so the review requires judgment.

If the AI turns that date into a commitment, pause. Have the learner locate the source, explain the discrepancy and correct the output. Then repeat the task with a different input. A prompt copied from the screen is useful only if the participant understands what needs to change when the source material changes.

Asking the model to review its own answer can help surface problems. It does not establish that the answer is correct. The learner still needs to compare the output with the source. NIST's Generative AI Profile describes the risk of confidently stated false content; fluent writing is not a reliability test.

Diagnose low adoption before prescribing another course

Use a conversation about a recent attempt rather than a general question about whether someone "likes AI."

  • "I couldn't log in." Check licenses, account access and the approved environment with IT.
  • "I didn't know what I could upload." Have the policy owner clarify the permitted data and give the learner a usable example.
  • "The answer looked good, but checking it took too long." Observe the review, narrow the task and compare total completion time.
  • "I didn't have a task worth using it for." Revisit discovery with the employee and their manager.
  • "It worked in class, but I got stuck on my own." Provide a place to ask the specific question and repeat the task with guidance.
  • "The task needs information from several systems." Investigate the process and technical dependencies before commissioning more training.

These are different problems. Record the evidence, assign an owner and agree on a next action. If everyone with low usage receives the same refresher course, the underlying problems can survive another round of training.

Office hours should resolve questions from real attempts

The question that matters often appears after the workshop, when an employee tries the technique on their own work. Schedule support before the first session and explain how to use it.

Ask participants to bring the task, a permitted input or anonymized description, the output they received and the point where they became stuck. The facilitator can then help them inspect a source, improve an instruction or decide that the task needs a different approach. Sensitive material should stay within the organization's approved environment.

Keep a record of recurring questions. If many people cannot tell which policy version to use, improving the policy library may help more than repeating a prompting lesson. If one role repeatedly struggles with review, build a short exercise around that difficulty. The employee office-hours guide explains how to structure that support.

Managers also need a practical role. Give them review criteria and time to discuss an attempt, including one that did not work. A request to "use more AI" without permission, examples or support can make experimentation feel like extra work.

Recognize when training reveals an implementation need

Ari describes an HR team that completed training and then wanted an agent that could answer HR questions. Ajaia's implementation team built a new front door for those questions. This anonymous example illustrates how learning can reveal a different way to deliver a recurring service; it does not establish a measured reduction in HR workload.

If your team identifies a similar need, scope it separately. Decide which sources the agent can use, who owns updates, which questions it must escalate and how it will be tested. The training program can help people recognize the opportunity. Engineering and ongoing ownership determine whether the system is ready to use.

Decide what to reinforce using evidence from the next attempt

At the first follow-up, inspect three things: whether the employee feels able to try the task, whether the output meets the agreed standard and whether the method is being used again. Keep those observations separate. Confidence can rise before quality improves, and usage can rise while review becomes more expensive.

Choose a fresh task sample and compare it with the baseline, including corrections. Ask the employee what helped and what remains difficult. Use that evidence to decide whether to reinforce the skill, change the workflow or resolve a dependency.

The AI education program guide includes a discovery worksheet and a practice plan. Start with one task employees want to improve, and make the next attempt easier to complete and review.

Research and documentation

References

Get started

Build AI training around the work your teams actually do.

Ajaia helps organizations turn AI literacy into role-specific workflows, responsible-use habits, champions, office hours, and measurable adoption.

Talk to Ajaia