Turn SOPs into reviewed interactive staff explainers

What shipped

Google Workspace has announced that the Gemini app can generate interactive simulations, models, tables and visual explainers from a prompt. The examples are general learning examples, such as 3D science models and interactive tables, but the more practical SME use is closer to home: showing a new starter how a company process works, step by step, using the documents the company already trusts.

For an Irish SME, the point is not the animation. The point is that approved SOPs, product guides, customer service scripts and policy documents can be turned into short explainers that staff can work through themselves. Instead of asking a manager to explain returns, booking changes, invoice approval or customer escalation for the tenth time that month, staff get a first stop that is consistent, visual and easy to revise.

This should not be treated as automatic training material. It is decision support for learning and onboarding, reviewed by a person before operational use. The named human review step should be the SME Review Gate: a subject matter expert checks the generated explainer, diagrams, tables and quiz answers before anything is shared with staff.

That review step is where the value sits. AI can draft the explainer, suggest scenarios and turn a dense procedure into a simple flow. A person still confirms that the small steps, exceptions and judgement calls are correct. If you want help designing this kind of safe, practical training workflow, it sits naturally under Artellis enablement and training.

Who should care

This is useful for HR, operations and customer service managers in companies where the same process is explained again and again, but mistakes still happen at the handover points.

Typical signs are easy to spot. New starters ask experienced staff the same questions. Managers keep making one-off Loom videos or long emails that go out of date. A procedure exists in a folder, but staff do not read it. The process has small but important steps, such as checking a delivery cut-off time, logging a refund reason, attaching the right approval note, or escalating a customer complaint before a deadline.

The companies that will benefit most are not trying to build a learning management system from scratch. They want a practical way to turn existing approved material into usable training. A small business with 20 to 200 staff might only need five or ten high-repeat explainers to remove a lot of noise from daily management.

It also matters for cross-training. If the accounts person is out, can someone else follow the invoice approval process without guessing? If a seasonal hire joins customer service, can they practise three common customer scenarios before they take calls? If a warehouse supervisor changes, can the replacement see the safety-critical steps in plain English rather than relying on memory?

The risk is that teams treat AI output as finished training. That is the wrong approach. The workflow should capture where the explainer came from, who reviewed it, what changed, when it was approved, and what staff still found confusing. Without that visible review trail, the business may simply create more content of uncertain quality.

A worked example

Take a 55-person Irish distributor with a customer service team, a small warehouse and an operations manager. One painful process is handling customer returns. The written returns SOP is eight pages long. It covers eligibility, proof of purchase, damaged goods, restocking fees, courier booking, credit notes and when to escalate to a manager.

In practice, new staff learn it by interrupting a senior colleague. Most questions are reasonable, but they slow the team down. The same errors recur: staff forget to check the return window, miss photos for damaged goods, or promise a credit note before the warehouse has inspected the item.

A useful AI-assisted workflow could look like this.

First, collect the approved source material. That might be the returns SOP, current customer email templates, product exceptions, courier cut-off rules and the credit note policy. Remove old drafts and mark the current owner of each document.

Second, use grounded retrieval and summarisation to produce a short explainer from only those approved sources. The output is not a generic lesson on returns. It is specific to the company’s own process. It might include a simple decision table: within return window, outside return window, damaged on arrival, ordered in error, suspected warranty issue.

Third, generate an interactive walkthrough. A staff member can choose a scenario, such as “customer says the item arrived cracked” or “customer wants to return after 35 days”. The explainer shows the next steps, the questions to ask, the record to update and the point where a manager must approve an exception.

Fourth, create five scenario questions. For example: “A customer has no photo of the damaged packaging. What do you do next?” The AI can draft the options and suggested answer, but the SME Review Gate is mandatory. The customer service manager checks every answer, edits language, adds local exceptions and rejects anything that could lead to a wrong promise to a customer.

Fifth, release it to staff as a self-serve first stop. It does not replace judgement or approval. It reduces repeated explanations and gives staff a safer way to practise. Each page should include a simple route for staff to ask, “This case is not covered. What should I do?” Those questions become the backlog for the next revision.

Finally, track what changes. A useful review trail might show source documents used, reviewer name, approval date, version number, staff questions received, and the next review date. That is often more valuable than the first version of the explainer itself, because it gives the business a living training asset rather than another forgotten PDF.

For a related way to think about AI inside everyday SME operations, see the Artellis insights library.

Where I'd start

Pick one high-repeat procedure. Do not begin with every policy in the business. Choose a process where managers are tired of repeating themselves and where errors have a visible cost. Good first candidates are booking changes, returns, new customer onboarding, invoice approval, stock adjustment, complaint escalation or holiday request handling.

Then set a tight scope. One approved procedure. One audience. One explainer. Five scenario questions. One named reviewer. One week of staff feedback.

The first build should answer four questions. What source documents are trusted? What must staff remember every time? Where do they need manager approval? What questions do they still ask after using the explainer?

Artellis would usually start by mapping the current training hand-off, cleaning up the source material, drafting the interactive explainer structure and putting the SME Review Gate into the workflow. The final output can be simple: a reviewed explainer, a small scenario quiz, a version log and a feedback capture step.

That is enough to prove whether the approach works. If managers get fewer repeat questions, new staff make fewer avoidable mistakes, and the reviewer can update the material without starting again, you have a pattern worth reusing across the next four or five procedures.