The structure of a funnel analysis

A funnel is the sequence of events a user has to pass through to reach a target action. For each transition you calculate the drop-off rate: the share of users who did not make it to the next step.

A basic e-commerce funnel:

Site visit             100,000 users   100%
       ↓  (−70%)
PDP view                30,000 users    30%
       ↓  (−60%)
Add to cart             12,000 users    12%
       ↓  (−75%)
Begin checkout           3,000 users     3%
       ↓  (−40%)
Payment                  1,800 users     1.8%

The 75% drop-off between cart and checkout is the point with the largest absolute potential. Lifting the conversion of that step by 10 percentage points raises the overall rate by 0.36 pp — more than the same improvement anywhere else in the funnel.

Segmenting the funnel: where the gap hides

An aggregated funnel hides the differences between segments. Breaking it down along the key dimensions often reveals non-obvious priorities:

Segment CR cart → payment
Desktop 35%
Mobile 18%
New users 12%
Repeat buyers 45%
iOS 22%
Android 16%

The mobile-to-desktop gap — 18% against 35% — is a concrete hypothesis about mobile checkout, not a vague statement that “our conversion is low”.

Tools and technical requirements

Funnel analysis requires event tracking with enough context at every step:

event: "add_to_cart"
properties:
  product_id: "SKU123"
  category: "electronics"
  price: 99.90
  source: "PDP"  // where it was added from
  session_id: "sess_abc"
  user_id: "u_456"  // or anonymous_id

Without session_id and user_id the events cannot be stitched into a funnel. Without source you cannot tell an add from the product page apart from an add out of a recommendation widget.

Tip: always attach a source or placement attribute to funnel events. It lets you compare the conversion of different entry points and attribute the contribution of personalization blocks.

Common mistakes

  • The wrong time window: a one-hour funnel for products with a decision cycle of several days
  • Ignoring multi-session journeys: a user who returned three days later is not counted in a narrow window
  • Analysis without segmentation: “conversion is 1.8%” means nothing without a breakdown by device, channel and user type
  • A funnel with no exit event: knowing where the user went after the drop-off — to search, or to a competitor — matters more than knowing only that they left