CEDX Knowledge Base · Customer Service
Knowledge Base measures itself the way a sceptic would: every article carries views, avoided contacts, deflection rate and helpful votes — and the searches that found nothing are a queue of their own.
The Deflection screen prints its formula under the funnel: avoided ÷ (avoided + contacted). A number with its working shown is a number you can argue with. That is the point.
Runs on demo data. The savings line uses the screen's own $24 per avoided ticket.
What it is
The article list reads like a ledger: views, avoided contacts, deflect %, helpful %, status and author on every row. The password-reset article deflects 46% on 14K views; a data-residency article deflects 26% and is flagged for work in the same table.
Categories are managed, not just nested: 697 articles summed across 36 of them, with a 43% mean deflection and an owner on every row. Billing & plans deflects 52% on 64K views; the search category sits at 22%, visibly below the line.
918K article views in 30 days become 71K estimated contact intents, 29K avoided tickets and 42K still-contacted — and the screen says how: intent contacts are readers who would have opened a ticket without self-serve.
Product tour
Not a mockup and not a concept deck. This is what opens at /app/knowledge-base.
01 — Overview
Forty per cent deflection with the dials behind it: 229 published, 28 in review, 45 draft, 18 archived. The alerts are root-caused — one no-hit cluster has 286 misses and a suggested title; 46 published articles are not indexed, so self-serve fails before it starts.
02 — Categories
Thirty-six categories ranked by views with deflection drawn as bars: billing leads at 52%, a search category trails at 22%. The mean — 43% — is on the summary card, so every row reads as above or below it.
03 — Articles
Three hundred twenty articles, sortable by the columns that matter: an expense-codes article avoided 676 contacts in 30 days on 15K views, and its 75% helpful rate says readers needed follow-up. Low-helpful is its own filter chip.
04 — Deflection
Views to intent to avoided to still-contacted, with the formula printed under the chart and the per-article contributions table below it — views, avoided, contacted, deflect % and savings per article.
Who runs it
Roles, not references. We have no named customers yet, so nobody in these photographs is quoted, credited or claimed as one.
Works the do-this-next list: 54 published articles without owners, 42 articles to write straight off the no-hit board.
unowned · 54Reads the 71 open no-hit queries as the voice of the queue — these are the exact strings customers typed and found nothing for.
no-hit rate · 8.6%Runs the index audit. Forty-six published articles are not indexed, which means self-serve loses before the first keystroke.
not indexed · 46The shape of it
Every figure below is legible in the captures above. Nothing here is a projection of your estate — it is the state of the demo data.
How it runs
The no-hit board collects what customers typed and found nothing for — 71 open queries, one cluster with 286 misses and a suggested title.
Articles get drafted against no-hits and owned; 54 published articles currently lack an owner and the overview says so.
Deflection is computed per article — avoided over avoided-plus-contacted — and the formula sits under the funnel, not in a wiki.
Unindexed, unowned and under-deflecting work surfaces as root-caused alerts with a fix attached, not as a dip in a chart.
One record
A help centre is where the estate's knowledge goes to be read. The same articles, votes and no-hits feed the products your customers actually touch.
Desk's knowledge screen is the agent-side view of the same idea — macros and articles with uses-30d and the same stale flags.
CEDX Live ChatThe chat bot's containment — 40% on its floor — is only as good as the articles it answers from. That library is this one.
CEDX CopilotAsk AI — ⌘J — sits on every screen here. Grounded answers cite their sources; this is the kind of source they cite.
CEDX FeedbackVotes and no-hits are two spellings of the same signal. Feedback counts the first; this page counts the second.
Limits
Finding this out on the third call is worse for you than reading it here, and worse for us.
Start
Open it right now
Your articles, your no-hits
Knowledge with the rest of service
Questions
Yes. Every screenshot is a capture of the running build, open at /app/knowledge-base. It runs on demo data — the help centre name on the chrome is the demo tenant.
On the screen: intent contacts are estimated readers who would have opened a ticket without self-serve, and the deflection rate is avoided ÷ (avoided + contacted). The formula is printed under the funnel, and per-article contributions are in the table below it.
The list of searches that returned nothing useful — 71 open queries at 8.6% of 30-day searches in the demo. Each cluster can be assigned an author, and one arrives with a suggested title already drafted.
Because that is the rate printed on the demo screen, and we would rather show the screen's arithmetic than invent an industry number. Your avoided-ticket value is part of what a pilot would establish.
No certification has been issued. What we can evidence about hosting, encryption and access is on the security page.
Live build, demo data, no card. Then look up what your customers searched for this week and found nothing.