Building a knowledge base that made "ask Roy" almost obsolete

Sometimes the answer isn't an answer, it's a system

Role IT Systems Lead
Company Precision Nutrition
Stack Halp, WordPress, Slack

TL;DR

  • ~200 person remote company with no ticketing system, no knowledge base, and every question routed to a human
  • Integrated Slack-native ticketing and an internal knowledge base fed by real support interactions
  • ~20% ticket reduction in 6 months, 40+ articles in month one — a behavior design problem, not a technology one

The problem

Precision Nutrition was a fully remote company with ~200 employees. Questions arrived through Slack DMs, a general IT channel, direct emails, and a Google Group inbox. There was no ticketing, no knowledge base, no way to track what was being asked or how often.

The same questions came up repeatedly. Answers lived in Slack threads that nobody could find two weeks later, or in someone's memory. Every question, no matter how common, required a live human response. There was no self-service path.

This created two problems: it didn't scale, and it trained people to treat IT support as a person to message rather than a system to consult.

Diagnosis

The stated need was "better IT support." The actual problem was that knowledge had no persistent home. Answering questions faster wouldn't fix anything if the answers disappeared into DMs and threads. The system needed three things:

  1. A way to capture requests without forcing people to change their behavior
  2. A place for answers to live permanently
  3. A feedback loop that trained people toward self-service over time

What I built

Ticketing via Halp, inside Slack

I set up Halp with a chatbot called "roybot" (my pixel face as the avatar). People could DM roybot directly to create a ticket, with automated intake questions to capture context upfront. I also configured a custom /halp slash command so anyone could request help from anywhere in Slack without navigating to a specific channel or remembering a specific bot name.

Email still accounted for a significant share of requests. Rather than telling people to stop emailing and use Slack instead, I routed inbound email into Halp's backend so those requests entered the same queue. The person asking didn't need to change anything. I'd rather do extra integration work on my end than dictate a workflow that people will quietly ignore.

Halp is now Jira Service Management.

Knowledge base via WordPress

I built a custom WordPress install as an internal knowledge base. Designed the theme, taxonomy, article templates, and style guide from scratch. Categories mapped to tasks ("Setup Instructions") rather than departments ("IT Documentation"). Articles were written with the language people actually used in tickets and Slack, not internal jargon.

I analyzed support ticket data to determine what to write first. High-frequency questions became the first articles. The knowledge base delivered value immediately instead of launching with content nobody needed yet.

Every article followed the same structure: what you're trying to do, how to do it, what goes wrong and how to fix it, and what to read next. Consistency made the knowledge base trustworthy. If one article was well-structured and accurate, people assumed the next one would be too.

The feedback loop

Every time I answered a question that had been asked before, I wrote a knowledge base article instead of just replying. If I suspected a question would recur, or if the information was broadly useful, I wrote the article first and shared the link as my response.

This served two purposes: it built the knowledge base fast, and it trained people to expect that answers lived in the knowledge base. Over time, the implicit message shifted from "message Roy" to "check the knowledge base, and if it's not there, then message Roy."

Design decisions

Meet people where they are. The system accepted Slack and email instead of demanding new behavior. Adoption depends on reducing friction, not publishing a policy.

Reduce cognitive load. Structured intake questions, a universal slash command, and a knowledge base designed for scanning.

Build the library by doing the work. Every article answered a question someone actually asked.

Results

  • ~20% reduction in support tickets within the first 3 months
  • 40+ knowledge base articles written in the first month, growing continuously after
  • Shifted user behavior from "ask Roy" to "check the knowledge base first"
  • Created a persistent, searchable record of IT support knowledge that survived individual conversations
  • Gave the company its first visibility into what people actually needed help with, since tickets were now tracked

What this demonstrates

This wasn't a technology project. It was a behavior design problem. The tools (Halp, WordPress, Slack integrations) were implementation details. The actual work was: diagnosing why a support model wasn't scaling, designing a system that respected how people actually behave, and building a feedback loop that made the system smarter over time without requiring anyone to change their habits upfront.