How to Turn AI Policies and SOPs Into Role-Based Training at Scale
One approved policy, four different jobs. Golpo drafts a role-specific explanation from that source and a short role brief. You review the obligations, lock the wording, and reuse one visual identity when the policy changes.
One policy, different responsibilities. Golpo turns the same approved source into explanations that show each role what to do. You bring the policy and a short brief for the audience. Golpo drafts the narration, you check the obligations, and then you render the wording you approved.
Role-based training changes the example and the action. It keeps the rule that is in the source. Writing four scripts by hand before Golpo starts splits the policy into four unofficial copies. Draft first, review the exact obligations, then lock the script.
Who it serves
Policy owners and learning leads preparing the next rollout. The artifact is a set of short videos that share one source version and one visual identity, each showing a different job. Sales, support, engineering, and the manager do not need the same story with the nouns swapped.
Keep one source and a role matrix
The approved policy or SOP stays authoritative. Each video records its source version, audience, owner, reviewer, approval date, and what replaces it. Departments contribute realistic cases. They do not fork the rule.
| Role | Must know | Must do | Must escalate |
|---|---|---|---|
| Employee | Allowed tools and data classes | Verify the output before it is used | Unexpected sensitive content |
| Manager | Which team workflows are approved | Review quality and the response when someone reports a mistake | Repeated misuse |
| Process owner | The control, the exception, and the current examples | Retire an example when the source changes | A material edit to the policy |
| Technical owner | Access and logging | Operate the controls | An incident or an outage |
Write the objective as a decision. “Recognize restricted customer data and use the escalation channel” can be reviewed. “Understand responsible AI” cannot.
A hypothetical policy, four briefs, two videos
The policy in this walkthrough is hypothetical. Use the same shape with the policy you are about to roll out.
Source, version 1.2. Restricted customer data stays out of unapproved AI tools. Restricted data includes payment numbers, government IDs, and health details. If restricted data appears in a draft, stop and use the escalation channel.
| Role brief | What Golpo should draft |
|---|---|
| Sales | A call note contains a full card number. Show how to record the need without the number, and when to escalate. |
| Support | A ticket contains a full account number. Show what may be summarized into the assistant, and what stays in the ticket. |
| Engineering | A log line contains an email address and an access token. Show redaction before the line goes into the assistant. The escalation owner is security. |
| Manager | Someone reports a paste. Show the response: stop the use, preserve the example, notify the policy owner. Leave the card-number demo to the sales video. |
Prompt Mode generates four draft scripts from the same policy plus one brief at a time. The sales draft and the engineering draft are the two to review first, because their examples are genuinely different: a customer conversation versus a production log. The support and manager drafts follow the same settings after those two are solid.
From draft script to approved video
- Source. Register the policy: owner, version 1.2, approval, next review. Upload it with one role brief. In Prompt Mode, ask for a two-minute video that uses only obligations from the policy and the example in that brief.
- Script review. The policy owner checks the rule. The role owner checks the example. Confirm “unapproved AI tools” and “restricted customer data” survived. Exact approved sentences then go through Script Mode, with drawing notes in Video Instructions, not in the spoken words. About 1,050 characters of narration fit in one selected minute. Turn on on-screen text from script only so labels stay inside that vocabulary. Golpo may show a subset of the script on screen.
- Settings. Save one Custom Style from an approved reference and select it for every role. The style is shared. The example is not.
- Result. The sales video walks through the call note and the escalation. The engineering video walks through the log line, the redaction, and the security owner. Same rule, same look, different work.
- One correction. If the engineering narration says “redact the email and the token” and the picture says “delete the log,” correct the label with frame editing. Frame editing changes the picture for that selected duration and leaves the audio as it is. If the spoken rule dropped the word “unapproved,” change the script and render again.
- One reuse. Version 1.3 of the policy says the last four digits of a card are still restricted. The sales example that treated “Visa ending 4412” as safe is now wrong. Update the sales script from the new source, render that role again, and point the old asset at the replacement. Support and engineering stay on 1.2 until their examples are affected.
What each control changes
| Control | What it does for this rollout | Plan |
|---|---|---|
| Prompt Mode | Drafts a role script from the policy and the brief, so four roles do not require four hand-written scripts before production starts. Generated wording still needs the obligation check. Uploading the policy does not certify the draft. | All plans. Length and controls vary. Prompt and Script Mode |
| Script Mode and on-screen text | Renders the approved narration. On-screen text from script only limits labels to that narration. It is not a caption track, and it does not prove the source is correct. | On-screen text: Canvas and Sketch. On-screen text |
| Custom Style | One named style, saved in My Styles, applied to every role video so the library is recognizable. It follows the reference and does not copy exact typography. | Business and higher, and Pay As You Go. API use also needs API access. Custom Style |
| Frame editing | The targeted visual correction after review. Charged on the selected duration of that version. | Business and higher, and Pay As You Go. Editing |
| Language variant | If a role also needs another language, that variant is a separate render. Review terms and examples in the new language. Record it as its own row in the manifest. | Creator and higher. Starter has a language add-on. Pricing |
| API job plus your manifest | The API generates a video and exposes job status. Your system stores source id, source version, role, language, job id, asset URL, reviewer, and approval state around that job. Watching the policy repo, routing approval, and publishing to an LMS are your orchestration. | Scale includes API access. Business has a paid API add-on. API Only is separate metering and does not include dashboard generation. API docs |
Plan, duration, and credits
Each role video in this walkthrough is 2 selected minutes. Standard generation is 1 credit per selected minute: 2 credits for sales, 2 credits for engineering. On a subscription, that frame-edit version is another 2 credits at the selected duration. On Pay As You Go, the first edit of a video is free, and a later finalized edit is charged at the selected duration. The version-1.3 sales render is another 2 credits. Support and manager, at the same length, are 2 credits each when you produce them. Custom Style and frame editing require Business or higher, or Pay As You Go. On Pay As You Go, edit-script-before-video bills 10% after the script and 90% after the video. See pricing. The Free plan is not a downloadable publishing workflow.
Lineage you keep outside Golpo
Store a manifest row for every job:
- Source id and version (1.2 or 1.3).
- Role and language.
- Golpo job id and the asset URL.
- Reviewer, approval state, destination, and expiry.
When the approved-tool list changes, the manifest tells you which rows to replace. Golpo will render the new job. It will not detect the policy edit, approve the replacement, or withdraw the old file from your LMS.
What to compare after the rollout
- Coverage: eligible people have the current version for their role.
- Decisions: people choose the allowed action on a role-specific case.
- Update work: hours from the source approval to the replaced video, and how many assets the manifest marked stale.
Treat incident counts as context, not as a score the video caused. For how to choose which SOPs enter the library, see Build a Training Video Library From SOPs. For the moment-of-use videos after the policy is taught, see employee adoption.
Review checklist
- One source version, named owner, next review date.
- Each role brief changes the example and the action, and preserves the rule.
- Obligations checked before render; Script Mode used once the words are locked.
- On-screen labels stay inside the approved wording.
- One saved style across the set.
- Manifest row for source, role, language, job, reviewer, and destination.
- A source change produces a new render of the affected role and a recorded replacement.
Frequently asked questions
What is role-based AI policy training?
Training that translates one approved policy into the decisions, examples, actions, and escalation rules relevant to a specific role. The examples change with the work. The rule stays the one in the source.
Should each department edit its own policy copy?
Keep one authoritative source and derive each audience video from it, with the source version, role, language, and asset id recorded in your own manifest.
Can Golpo automatically publish training to our LMS?
Golpo generates the videos and, on the API, returns job status. LMS publishing, approval routing, and watching the policy for edits stay in your systems unless a documented integration covers that destination.
How often should policy videos be updated?
Review on the cadence you set, and again when a material source change affects that role’s action or example. Update the affected video and record the new source version.
What should reviewers check?
Policy fidelity, role realism, narration and on-screen wording, language quality, accessibility needs, source version, and expiry. A polished video can make a drifted rule look official, so check the obligations before you approve it.
Do we write four scripts before using Golpo?
No. Prompt Mode drafts each role script from the approved policy and a short role brief. After reviewers lock the obligations, Script Mode renders that approved wording. Visual directions go in Video Instructions.
Will Golpo notice when the policy changes?
No. Your team monitors the source, updates the version in your manifest, and sends a new generation job for each affected role. Golpo does not synchronize policy systems or publish the replacement on its own.
The variant to make before the next rollout
Pick the role that currently gets the policy wrong most often. Create that video from the approved source and the role brief, then compare two things you can actually count: correct decisions on that role’s cases, and the work required to replace the video when the source changes.
Create the highest-friction role video · See plan and API options
Tags
