Documenting Support Processes Agents Actually Use
Write support documentation agents open during a live call: short pages, clear owners, simple structure, easy search and a routine that keeps it current.
Most support teams have documentation. Very few have documentation that agents trust enough to open in the middle of a call. The binder is out of date, the shared drive has four versions of the refund process and the one colleague who knows the answer is on holiday. So agents guess, ask a neighbour or improvise. Customers get inconsistent answers, and new hires take months to feel confident.

This guide is about writing and maintaining documents for the people who use them under pressure. For knowledge that also feeds automated assistants, see Building a Knowledge Base for Support and AI; for the habit of turning call insights into updates, see Finding Knowledge Gaps from Call Reviews.
Why documentation gets ignored
Common reasons agents stop opening it
- Too longPages of background before the answer. A caller cannot wait while an agent scrolls.
- Out of dateOne wrong answer teaches agents not to trust any page.
- Hard to findTitles written in internal jargon, not in the words callers use.
- No ownerNobody is responsible, so nobody fixes it.
- Written onceIt describes how the process was supposed to work, not how it works.
- Not testedNobody has tried to follow the steps from scratch.
Decide what deserves a page
You cannot document everything, and you should not try. Start with the calls that matter most, and keep the rest in people's heads until they prove they need a page. A useful ranking combines how often something happens with how costly a mistake is.
| Call type | Frequency | Cost of an error | Priority |
|---|---|---|---|
| Top ten repeated questions | High | Low to medium | Write first, keep very short |
| Refunds, exceptions and approvals | Medium | High | Write carefully with limits and approvers |
| Urgent or safety issues | Low | Very high | Write as a clear step list; test it |
| Rare edge cases | Very low | Varies | Add a short note, expand only if they recur |
| Internal tool how-tos | Medium | Low | Use the tool's own guides where they exist |
Write for a live call
Agents scan; they do not read. Put the answer first, then the steps, then the exceptions. Use plain language that matches what callers say. A good page fits on one screen and uses a repeatable pattern, so that agents learn where to look.
A template that works for most procedures
- TitleNamed after the caller's problem, not the department.
- When to use itOne sentence: the situation this page covers.
- StepsFive or fewer numbered actions, each a single instruction.
- What to sayOne or two phrases the agent can use, if wording matters.
- Exceptions and limitsWhen to stop and escalate, and to whom.
- Owner and dateWho maintains the page and when it was last checked.
Organise so people can find things
The most useful structure is the one agents guess correctly. Group pages by what the caller wants (billing, delivery, account access) rather than by internal team. Keep the hierarchy to two levels. Rely on search for the rest, and check what agents search for, because their search terms show what is missing. Link between related pages so that an agent who lands in the wrong place can move on quickly. Using Knowledge Sources in Call shows one way to keep content in a searchable place.
- Keep one page per procedure. Do not bury three processes in one document.
- Avoid duplicates; link instead of copying.
- Put the most-used pages on a short home list.
- Archive pages that are retired instead of leaving them to confuse people.
Scripts and wording
Some situations call for exact words: a recording notice, a verification question, a regulated statement. Mark these clearly and keep the rest flexible. Over-scripted calls sound robotic; under-scripted ones risk saying the wrong thing. How to Write a Sales Call Script and How to Open and Close a Support Call offer guidance on balancing the two.
Give every page an owner and a review date
Maintenance is what separates useful documentation from a graveyard. Every page needs a person responsible, a date it was last checked and a trigger to update it. Triggers include product changes, policy changes, repeated questions from agents and errors found in call reviews. Set a review cycle suited to the content: weekly for volatile pages such as promotions, quarterly for stable ones.
A light maintenance routine
- CollectAgents flag unclear or wrong pages with one click or a message.
- TriageThe owner reads flags weekly and decides what to change.
- UpdateChange the page, note the date, and tell the team if the change matters.
- VerifyCheck by following the steps as a newcomer would.
- RetireArchive pages that nobody has opened for a long time.
Mistakes that make documentation fail
- Writing it all at once in a burst, then never touching it again.
- Documenting the ideal process instead of the real one. If agents work around a step, either fix the process or document the workaround honestly.
- Using screenshots that go stale as soon as the tool changes; use words for steps and screenshots only where they help.
- Leaving old and new versions side by side without saying which is current.
- Forgetting the exceptions, which are exactly where agents need help.
A useful test is to hand a page to someone who has never done the task and watch what they do. Wherever they hesitate or ask a question, the page is unclear. Fix it and repeat.
Involve the people who use it
The strongest authors are often agents. They know which questions come up and which wording works. Let them draft pages, then have the owner review for accuracy. Recognise contributions publicly. A page written by a colleague is more trusted than one handed down from head office. Pair this with onboarding: ask each new hire to note where the documentation confused them, and fix those places. They see problems that experienced staff have learned to ignore. A 30-Day Onboarding Plan for New Agents includes this idea.
Check that it works
- Listen to a sample of calls for hesitation, which often points to a missing or unclear page.
- Track the questions agents ask supervisors, because repeated ones should have a page.
- Watch repeat contacts; inconsistent answers produce second calls. See Repeat Call Rate.
- Ask new hires how long it takes them to find an answer.
- Run a quarterly test where someone unfamiliar follows a procedure exactly.
Finally, keep the effort proportionate. A team of ten does not need a documentation department. It needs a handful of clear pages, a person who cares about them and a habit of fixing a page the moment someone notices it is wrong.
Keeping documents close to the call in Call
In Call, scripts can be attached to a phone line and shown to the agent during the call, and branching scripts can be drawn as a diagram. Knowledge sources hold the content an AI agent answers from, and message templates keep written replies consistent. The how-to guides inside the product describe each screen. See Writing and Using Call Scripts in Call, Branching Call Scripts with the Script Diagram and Message Templates in Call: Writing Once, Reusing Everywhere.
Frequently asked questions
How long should a support procedure be?
Short enough to read on one screen during a call. If it is longer, split it into separate pages or move background into a linked page.
Who should write the documentation?
A mix: agents draft because they know the real questions, and an owner reviews for accuracy and consistency.
How do we keep documents current?
Give each page an owner and a review date, make it easy for agents to flag errors and update when products or policies change.
Should we use video?
Short videos can help for visual tools, but text is easier to scan and search during a call. Use both where they add value.
How many pages do we need?
Fewer than you think. Start with the most frequent and the riskiest calls and add pages only when a gap appears in real calls.