From One Week to One Hour: How AI Is Speeding Up Salesforce Documentation

By Alex Quirarte, Senior Salesforce Administrator, UCPath

Our Salesforce team now uses AI to write the first draft of two documents every system enhancement needs: a technical document and a user testing guide. Each first draft used to take about a week to write by hand. With AI, it takes less than an hour. People still review, correct, and approve every document before it is used.

Why documentation takes so long

Every Salesforce enhancement needs two documents before it can go live: a technical document that explains how it was built, and a testing guide that walks testers through proving it works.

Even one feature has many moving parts. Our Request Supervisor Callback feature has 19 separate pieces of setup, including data tables, automated workflows, custom code, access rules and screen layouts.

Writing that up by hand meant opening each piece, working out what it does, and describing it in plain English. Then the same information had to be written again as step-by-step test instructions. It was slow, repetitive work that took time away from building, and late changes could easily go undocumented.

How AI helps

Salesforce stores every part of a feature’s setup as files in our project. Instead of describing the feature from memory, we give those files to an AI assistant (ChatGPT) and ask it to explain what it finds.

Read the setup. The AI reviews the files that define the feature: its data, workflows, code, access rules, and screens.

Draft the technical document. It writes the same sections every time: purpose, a plain-language overview, a list of every component, field details, access rules, and how each automated step works.

Flag what looks off. Because the AI reads every detail literally, it catches things people skim past. In our first document, it spotted that the internal name of an access setting was misspelled permisson, something testers and admins need to know.

Draft the testing guide. Every rule the system enforces is something a tester should confirm. The AI turns each one into a test case with numbered steps and the expected result.

People review. A developer checks the technical document against the build and adds the business reasons behind it. A business user walks through the testing guide, removes cases that don’t apply, adds real-world scenarios, and confirms results in a test environment before user acceptance testing.

Publish. The final documents are saved with the project and shared with stakeholders as Word files.

Accessibility is built in from the start: every document includes accessibility notes, and every testing guide includes an accessibility check.

The AI works only from the feature’s setup files, not from employee or production data, and the team follows UC’s guidelines for using AI tools.

The results

We first used this approach on the Request Supervisor Callback feature.

Measure	Before (manual)	With AI-assisted drafting
First draft of technical document	About 1 week	Less than 1 hour
First draft of user testing guide	About 1 week	Less than 1 hour
Document structure	Varied by author	Same sections every time

Times are for the first draft only. Review and approval by developers and business users happen afterward and are not included.

The technical document covers all 19 components, and the testing guide includes 12 test cases, from access checks to an accessibility check. Because every release follows the same template and every error message is quoted exactly, handoffs to testers, support staff and future developers are easier.

Keeping people at the center

AI writes the first draft; people stay accountable for the result. A few lessons from our first run:

Start from the source. The AI is most accurate when it reads the actual setup files, not a summary of the feature.

Give it a template. Asking for the same sections every time makes drafts predictable and quick to review.

Keep people in the loop. The setup files show what was built, not why. Developers add the business reasons, and testers confirm every expected result.

Build accessibility in. Including it in the template means it is never an afterthought.

What’s next

Next, we plan to use this approach for all upcoming Salesforce monthly releases and make it a standard step in our release process. The same pattern can help any team that writes documentation from structured information: give AI the source material and a clear template, and have people review everything it drafts.