AI Knowledge Assistant

How I designed the documentation layer that lets 1,300+ clients find answers without opening a ticket

Figma

Figjam

Confluence

Pendo

Google meet

ChatGPT

AI Knowledge Assistant

How I designed the documentation layer that lets 1,300+ clients find answers without opening a ticket

Figma

Figjam

Confluence

Pendo

Google meet

ChatGPT

  1. Problem statement

Documentation lived in two separate systems — Confluence for internal teams, a Wiki for customers. Neither talked to the other. Neither was easy to search.

The result: everyone defaulted to opening a Support ticket, even when the answer already existed. Support became a search engine. CSMs spent time answering questions instead of managing clients.

  1. discovery & Research

"I know the answer is in Confluence — but it's faster to ask a colleague." — Internal team member"

10

Interviews

Process

6 internal, 4 customers

~2

Weeks

With CMS, Enforcement, Onboarding, SMBs and Enterprise

Process

Before the redesign, retrieving a document meant navigating a maze built from uncertainty. Agents didn't just search for files, they searched for where to even begin. Wrong turn? Ask a colleague. Wait half an hour. Try again. Find multiple results? Spend another 20 minutes reviewing each one manually. Find nothing? Loop back, try different keywords, escalate, restart.

And even when the right document finally surfaced, it still had to be validated, and if it failed, everything started over.

No momentum. No recovery. Just loops.

  1. Design

The new way of finding anything

Now agents just ask. They type a query into the chatbot, "Client patents in China", and within seconds the AI searches both Confluence and the Knowledge Center simultaneously, retrieves the relevant documentation, and surfaces it directly in the platform.

No colleagues to interrupt. No dead ends. No restarting from scratch.

What used to take up to half an hour of navigating, waiting, and guessing now takes a single prompt.

The new way of finding anything

Now agents just ask. They type a query into the chatbot, "Client patents in China", and within seconds the AI searches both Confluence and the Knowledge Center simultaneously, retrieves the relevant documentation, and surfaces it directly in the platform.

No colleagues to interrupt. No dead ends. No restarting from scratch.

What used to take up to half an hour of navigating, waiting, and guessing now takes a single prompt.

Design assets

Results

Launched early 2025, in active use by customers and internal teams.

Fewer repetitive Support tickets in how-to, status, and integration categories — tracked through Pendo and the ticketing system. Higher self-service confidence captured via in-assistant post-session surveys. Faster new-hire onboarding — the assistant became an unofficial tool for new agents, reducing senior colleague dependency. Documentation framework extended to 3 additional product areas by the Knowledge Management team.

Impact data still maturing. Metrics updated as Q2 2025 data becomes available.

The end of the page, the start of something good. Get in touch:

2026 ©️ Juan Daza

The end of the page, the start of something good. Get in touch:

2026 ©️ Juan Daza