
Making AI Work for Your People: Training, Coaching and Adoption
Build AI capability through actual workflows, supported practice and output evaluation. Learn what executives should expect from training and coaching.
Published · 8 minute read
AI-created illustration.

The short answer
InnovateX Solutions runs AI training for Australian business staff on work they already do, online or at your office, in ChatGPT, Claude or Microsoft Copilot. Staff practise deciding what information they may give the tool and checking its output against the source document. We agree a measure for each use, such as how often staff have to correct the AI draft, so a manager can see whether AI is making a difference. Organisations with A$10 million to A$200 million in annual revenue that want ongoing help can move on to the AI Implementation Partnership, which adds role-specific training, coaching for the staff who help colleagues use AI and hands-on implementation support.
A team member can complete an AI training session and still be unsure whether to use the tool on the next customer enquiry. They may understand how to write a prompt but not which information is permitted, how carefully to check the answer or when the existing process is the better option.
That gap needs to be addressed during implementation. Training should prepare people for the work being introduced, and coaching should help them handle what happens when they start using it.
For an executive team, this means treating internal capability as part of the investment. Budget time for people to learn, practise and help improve the process. Assess whether they can use the system responsibly before assuming a licence has become productive capacity.
Teach the workflow being introduced
Consider an illustrative procurement team preparing comparisons of supplier submissions. An assistant can help extract information from documents and produce a draft comparison. The analyst still needs to verify what each supplier offered, identify qualifications and recognise when two submissions aren’t directly comparable.
A general session can introduce AI, but the team also needs practice with this task. Give participants a suitable sample submission and ask them to prepare a comparison using the approved process. Then examine the result against the original documents.
The exercise should include an omission that matters. One supplier might exclude implementation support from its price while another includes it. A fluent summary that misses the exclusion could distort the commercial assessment. Participants need to learn how to detect that problem and what to do when the system cannot resolve it.
Use fictional or appropriately approved material for practice. Don’t make the training session the first time anyone considers whether supplier documents may be uploaded to the tool.
The purpose is to help employees complete a real process well. Prompt construction is one skill within it. Source checking, business judgement and knowing when to escalate can demand more attention than the prompt itself.
Help people understand what good checking involves
“Check the output” is too vague to be a complete instruction. Define what the reviewer is checking and how they can establish whether it is correct.
For the procurement example, a reviewer could trace each material claim back to a passage in the supplier’s submission. They would check that the comparison preserves qualifications, uses consistent units and distinguishes a stated commitment from an inference. An absent answer should remain absent rather than being filled with something plausible.
Practise with both good and flawed outputs. Ask people to explain why they accepted or rejected a statement. If a participant accepts a polished but unsupported answer, work through the evidence with them. Ask them to repeat the check on another output to see whether they can apply it themselves.
Some tasks require specialist review. An employee may be able to identify a missing commercial term without being qualified to interpret a difficult legal clause. Training needs to explain where that boundary sits and who can help.
Managers also need to understand the limits of the task. If they assess staff only on speed, employees may feel pressure to shorten checks that the implementation depends on. The performance expectations and the operating controls need to agree.
Give each role the learning it needs
Executives need enough understanding to make investment and risk decisions. They should be able to question a benefit estimate, recognise the limits of a trial and ask whether an apparently successful result depends on uncounted human effort.
Process owners need to understand how the work changes. They need to decide which steps remain, how exceptions are handled and who can approve a change. They also need a way to tell whether staff are using the process as intended.
Users need to practise their own tasks. Administrators and technology teams need the knowledge required to maintain access, manage changes and respond to failures. Giving everyone the same introductory session won’t cover those different responsibilities.
The National AI Centre’s implementation guidance recommends assessing training needs across the organisation and for people accountable for particular AI functions, with skills revisited as systems change. National AI Centre implementation guidance.
A training plan can be modest at first. Identify the people involved in the initial workflow and what each must be able to do. Expand it as the scope expands, rather than trying to teach the whole organisation every possible use before any process is ready.
Use coaching to improve the implementation
Once employees begin working with the system, problems emerge that training may not have covered. A document won’t load. A response is correct but too broad. A source is out of date. Someone finds it quicker to complete an unusual case manually.
Set aside a way to examine those cases with the team. Short working sessions can use recent examples to distinguish a learning need from a defect in the process or system. Keep a record of recurring issues and the changes made in response.
If several people make the same mistake, look beyond individual performance. The instructions may be unclear, the source material may be inconsistent or the interface may encourage a poor choice. Coaching should feed that evidence back into delivery.
Include people who question the tool. Their objections may expose a cost or risk. An experienced employee who can explain why a proposed shortcut omits a condition or exception is contributing to the design.
The fractional AI leader should remain responsible for making sure these observations lead to decisions. Some can be resolved by the delivery team. Others may require the process owner to change a rule or the executive sponsor to reconsider the scope.
Address the reasons people hesitate
Staff will reasonably want to know why AI is being introduced and what the organisation expects to change. Explain the intended outcome and avoid assurances about future staffing decisions that leadership cannot support.
Explain how usage and performance information will be interpreted. If someone reports that the tool is unreliable on a certain task, they need to know that the report will be investigated. Treating every low usage figure as resistance can hide a design problem.
Make room for learning in workloads. Asking people to attend training and experiment while maintaining unchanged output expectations can make the new process feel like additional unpaid effort within an already full day. Managers need to plan the transition, including temporary support where it is necessary.
Accessibility and different levels of confidence also deserve attention. Provide written instructions alongside demonstrations. Let people practise and ask questions in a format that works for them. Check competence through the task itself rather than rewarding whoever sounds most enthusiastic in the workshop.
Measure capability alongside adoption
Attendance and login counts tell you something about participation. They don’t establish that work is being completed well.
For a bounded rollout, assess whether employees can identify unsuitable inputs, verify a material claim and handle an exception correctly. Observe whether they know how to report a problem and continue the process when the system is unavailable.
Then examine the operating evidence. How much eligible work uses the new process? What kinds of corrections are required? Does checking take more effort than expected? Are staff escalating uncertain cases at the right point?
A high correction rate may indicate that people are checking carefully, that the system needs improvement or both. Read the measure alongside examples before deciding what it means. A low reported error rate is also difficult to interpret if nobody is looking for errors.
As capability develops, keep enough internal knowledge to maintain the process. Assign ownership of the instructions and source material. Prepare managers to onboard new team members and identify changes that need specialist help. Ongoing external support may still be appropriate, but routine operation should not depend on remembering what a consultant said in a workshop.
In an AI implementation partnership, training and coaching are part of the engagement as the work is delivered. If you’re reviewing an AI proposal, ask what your team will be able to do at the end, how that capability will be assessed and who will help them while they are learning.

Fractional Chief AI Officer · 6 parts
- 01Fractional Chief AI Officer: what it is, when SMBs need one, and where it fits next to your CIO
- 02Does Your Business Need a Fractional Chief AI Officer?
- 03What Should You Look for in a Fractional Chief AI Officer?
- 04From AI Ambition to Implementation: The First 90 Days
- 05Building Security and Governance into Your AI Strategy
- 06Making AI Work for Your People: Training, Coaching and AdoptionYou're here
Questions
How is coaching different from AI training?
How can we tell whether staff are capable of using the system?
What if employees do not use the AI tool after training?

Written by Todd Johnson
Founder of InnovateX Solutions. Todd provides fractional CIO, CTO and Chief AI Officer leadership for Australian businesses and government, and wrote the XAF Connected Architecture framework.
Todd on LinkedIn