Skip to content
Idea Atlas
Sign in

HelpDeck

As described:StopNarrowed:Narrow

No as described. Weak Go, only as a Forge help-center app for Confluence, built as a second Forge app.

software teams · ~$50 a month per customer · last reviewed (next review by 29 Dec 2026) · 25 cited sources

Save
Share

← All research

5. Pre-Mortem: It's September 2028 and HelpDeck Is Dead

Three most likely causes of death, ranked by probability. They compound: #1 makes #3 fatal.

Cause #1: Nobody switched (probability ~45% of failure cases)

What happened. The Forge app shipped in month 3. It worked. The pilots stayed. But the Marketplace showed Scroll Sites (111 reviews) and Refined Sites (63 reviews) above it for every help-center search. Teams on Zendesk Guide or Intercom liked "write once", but moving 300 articles and fixing redirects was a project nobody had time for. Teams already on Scroll or Refined had no reason to move. By month 18 the app had 45 customers and $2.2k MRR, and new installs had flattened at 3 a month.

Early warning signs

  • Fewer than 5 paying pilots by day 45.
  • Discovery calls end with "interesting, maybe next quarter".
  • More than half of trials come from teams already on Scroll or Refined, and they don't convert.
  • Trial-to-paid below 20% after the listing goes live.
  • You're adding themes to win demos.

Fixable? Partly. Doing the move for them (import plus redirects) is the only lever that lowers the switching cost. If teams won't switch even when the move is free, the pain isn't there. Stop.

Cause #2: The platform shipped it (probability ~30%)

What happened. One of three things happened:

  • (a) Atlassian's public-pages feature for Confluence (unverified today) turned out to be native, then gained search and custom domains.
  • (b) Rovo started answering customer questions straight from Confluence inside Jira Service Management, so "a public help center" mattered less.
  • (c) For the Notion version: Notion added visitor search and password protection to Sites. Every Notion help-center tool lost its entry tier overnight.

Early warning signs

  • Atlassian or Notion changelog items mentioning public pages, site search or external agents.
  • Atlassian community posts asking "why pay for an app now?".
  • Your cheapest tier's churn rises while the Advanced edition holds.

Fixable? Partly. You can't stop the platform. You can sell what it has little reason to build: the doc-owner loop (turning visitor failures into assigned work) and the move-in tools from other help centers. Keep those central, and keep the plain site cheap.

Cause #3: AI changed how customers look for help (probability ~25%)

What happened. By 2028, more customers asked an AI assistant or the vendor's in-app agent than opened a help center. Help-center page views fell. Customers saw fewer visits in their monthly report and asked why they were paying for a website. The doc-owner loop still worked, but it had less data to work with.

Early warning signs

  • Median help-center views per customer fall quarter on quarter.
  • Customers ask for "an API or MCP server for our docs" more than for themes.
  • AI-crawler traffic (llms.txt, bot hits) outgrows human traffic.

Fixable? Yes, if it's caught early. Treat the help center as the source of truth that AI reads. Serve clean, cited content to AI tools (llms.txt, an MCP endpoint), and count "answers served from your docs" alongside page views. HelpKit already ships an MCP server, so this is table stakes, not a moat.

Honourable mentions (less likely, still real)

  • Trust event. A restricted Confluence page is published by mistake. One leak of internal docs ends the app. Treat page restrictions as a hard block, and test them before every release.
  • Two products at once. At ~15 hrs/week, running this alongside StatusLens or another idea leaves both half-built.
  • Forge limits. Serving fast public pages may need an external host, which adds cost and may cost you the "Runs on Atlassian" badge. Check this in week 1.
  • Name clash. "HelpDeck" is almost the generic "help desk". Search and trademark will be hard.