cx-self-service-funnel

Use to analyse the help-centre funnel from search through article view to contact, find actionable drop-off, and align KB content to contact drivers. Trigger for "self-service funnel", "contact after

SKILL.md

GitHub

Self-service funnel analysis

Teams report article views and ticket volume as if one explains the other. They do not. A customer can read three articles and still open a ticket — and that sequence is usually a self-service failure, not success. The funnel you need is:

search (or land) → article view → [optional: in-article action] → contact (or exit)

Without instrumented steps and a contact-after-view flag, you are optimising page traffic while contacts stay flat.

Define the funnel steps

Step Event Minimum fields
Enter Search submit or direct article land session_id, timestamp, query or article_id
View Article body rendered article_id, session_id
Engage Scroll depth, time on page, clicked CTA optional but useful
Resolve signal "Helpful" vote, flow completion, no contact in window session_id, customer_id
Contact Ticket/chat/call opened customer_id, timestamp, channel

Contact-after-view (CAV) — contact within a defined window (often 24–48h) after an article view in the same session or for the same customer_id:

CAV rate = contacts after view / article views   (not "deflection")

High CAV on an article means the article is associated with failure — wrong content, stale steps, missing prerequisite, or it describes a problem you cannot self-resolve. That is the prioritised fix list, not "low views."

Instrumentation requirements

If any of these are missing, state what you cannot conclude:

Gap Consequence
No search logging Cannot diagnose findability vs content
Anonymous views only Cannot link to tickets
Ticket opens not tied to identity CAV is understated
Bot and KB on different sessions Double-count or miss paths
No article version on view event Stale fixes look ineffective

Join keys: customer_id when logged in; session_id with careful TTL for anonymous; never rely on email subject matching.

Diagnose drop-off by stage

Drop-off pattern Likely cause Action
Search, no click Findability, titles, synonyms Query clustering, title rewrites
Click, immediate bounce Wrong result ranked first Snippet quality, intent routing
Long read, then CAV Content wrong or incomplete Rewrite against resolved tickets
High views, low CAV, high contacts elsewhere Article fine; process blocks self-serve Fix product or policy path
No search, direct land, CAV Links from product/errors In-app copy vs KB parity
Views falling, contacts flat Traffic shifted to bot or phone Cross-channel funnel

Match KB to contact drivers

Export top contact drivers (taxonomy or tags) and map each to:

  1. Primary article — exists? accurate? last updated?
  2. CAV rate and view volume
  3. Contained bot sessions on same intent, if applicable
Driver tier KB priority
High volume + high CAV Rewrite immediately
High volume + no article Write net-new
Low volume + high CAV Fix or delete misleading article
High views + low CAV Candidate for promotion / in-app link
Stale + high views Harmful — update before new content

An article that ranks well in search but drives CAV suppresses deflection — customers waste time before contacting anyway. That is worse than no article.

Success metrics that are not vanity

Metric Use Misuse
CAV rate by article Prioritise fixes Sole deflection KPI
Search success rate Findability Ignoring content quality
Contacts per 1k MAU Holistic self-service Blaming support for product bugs
Issue-level self-serve rate Requires identity + driver Article views alone

True deflection requires knowing whether the contact would have happened without self-service — historical funnels only describe association. Holdouts and matched pre/post at driver level are stronger when you need a causal claim.

Traps

"Helpful" votes. Satisfied readers vote; angry readers contact. Votes skew positive; CAV catches the rest.

Publishing more articles. Coverage without CAV analysis increases findability of wrong answers.

SEO traffic. External hits inflate views; segment logged-in and in-app paths.

Contact button on every page. Makes CAV the rational choice; measure button placement experiments.

Present results to the user

  1. Funnel counts — search → view → contact with conversion rates at each step; define the attribution window used.
  2. Top CAV articles — volume, driver mapping, staleness, suggested fix type (rewrite, merge, deprecate, in-app only).
  3. Search gap list — queries with no click or high click-then-CAV.
  4. Driver coverage table — top N drivers vs KB asset and CAV rate.
  5. Instrumentation gaps — what biases the funnel (identity, session, channel).
  6. Prioritised actions — ordered by expected contact reduction, not page views.