Data Analysis

Anomaly Diagnosis

Funnel Analysis Misunderstanding Diagnosis

Diagnose why a user was not counted as a funnel conversion, and explain the five core funnel rules and common pitfalls.

IndustriesAll industries

Agent · Skills Live

skill · loadedindustry+100

Run this Skill on my data

Conversion funnel

stage to stage

100%

74%

52%

31%

Analysis readyevidence attached

Funnel Analysis Misunderstanding Diagnosis is a funnel-rule diagnosis capability in ThinkingAI's industry Skill library. When a user should have converted but was not counted, or funnel conversion is abnormally low, it helps data analysts investigate the root cause against the five core funnel rules one by one, not guessing at the cause but using real data to verify exactly which rule was broken. This is ThinkingAI's decade of industry experience, packaged into ready-to-use Skills.

Live walkthrough

AgentSave as AgentExport session

funnel-analysis-misunderstanding-diagnosis

Our "product view → add to cart → submit order → payment success" funnel converts at only 8.2% overall. The business side is convinced data is being lost, and they threw me a user ID, saying this person clearly paid but wasn't counted as a conversion. Diagnose whether it's a config problem or a data problem.

Get the true funnel config for a sample e-commerce platform DEMO

1-day window, associated property item_id, global filter platform = 'app'

Filter by #user_id, verify the step triggers for the user the business side provided

All four steps triggered, but the item_id for view and payment don't match

Drill into the user's full event timeline, check step order and window

Sample 500 users who "triggered all steps but weren't counted as conversions," classify by the funnel's five rules

All four causes map to the rules, not a single case of missing data

Aggregate the rule-violation distribution, generate config recommendations

The diagnosis may be the opposite of what the business side expects: both the config and the data are fine, and 8.2% is the correct result under the current rules: the user they threw at me happens to be a textbook case of the associated-property rule.

1. The current funnel and config

View-cart-order-pay conversion funnel (last 30 days)

Product view

100%

Add to cart

41.6%

Submit order

18.3%

Payment success

8.2%

This funnel has three key configs: a 1-day window, associated property `item_id` (only the same product counts as one path), and global filter `platform = 'app'` (applied at every step). Before you can make sense of 8.2%, you have to accept that all three of these rules are working correctly.

2. What happened to that "paid but not converted" user

His event timeline: 10:02 views product A001 → 10:15 adds product A002 directly from the search list → 10:23 pays for A002. All four steps triggered, but the funnel associates by item_id, and "view A001 → buy A002" does not form a conversion path for the same product, so being excluded is the rule working correctly, not lost data.

3. Sampling 500 users: four common misreads, corrected one by one

Not lost data, the associated property is working (46%): the item_id for view and payment don't match, and cross-product behavior shouldn't count as one path to begin with

Not undercounted conversions, the window is working correctly (31%): these users paid only 24 hours after viewing, past the 1-day window

Not dropped users, it's the step-order constraint (14%): a historical add-to-cart came first, then the view, so it doesn't meet "strictly in-order triggering"

Not a filter typo, the global filter applies at every step (9%): payment happened on web and was excluded at the payment step by the platform = 'app' condition

Recommendation

How to change the config depends on the business question: if you want "conversion efficiency for the same product," the current config is right and 8.2% is the true level; if you want "overall user purchase conversion," just remove the item_id associated property and build a parallel funnel, and the two answer different questions. Separately, the 31% of over-window users suggests a long decision cycle: if high-ticket categories make up a large share, try widening the window to 3 days for comparison.

On your data

That was a simulated run

Leave your work email and we will run a live walkthrough on your real business data.

The problem

Funnel analysis has five hard constraints: steps must trigger in strict order, all steps must complete within the window, every step must satisfy the global filter, the associated property must be consistent across steps, and the time zone must be unified. Many analysts don't understand these rules and assume the data is wrong when a user who clearly paid was not counted as a conversion. Traditional investigation means manually pulling per-user event details and comparing line by line, averaging 3 to 4 hours and easily missing the rule violation.

What it does

Complete output of the five core funnel rules: step order, time window, global filter, associated-property consistency, and time-zone consistency, each with a typical example

Precise user-level diagnosis flow: first filter by #user_id to verify whether the user triggered each step, then drill into the complete event timeline

Flag four common pitfalls: filtering #user_id with user_property, drilling down to a user, static timestamps for a relative time range, and missing details due to the default limit

All verification must call real MCP tools for data, with no guessing or fabrication, and data clearly marked when unavailable

When to use it

A user reports completing the conversion flow but was not counted in the report

Funnel conversion is abnormally low and you suspect a misconfigured step order, window, or filter

The associated property is inconsistent across funnel steps and you need to verify whether it caused a conversion failure

Funnel data is anomalous in a cross-time-zone project and you suspect inconsistent time zones

A newly built funnel report has lower-than-expected conversion and needs a systematic config investigation

In the field

Case

An e-commerce project · browse-add-to-cart-pay funnel investigation

The "browse, add to cart, pay" funnel had only 8% conversion. The Skill retrieved the config and found a 1-day window and an associated property of item_id, then checked the user's event details: browse A001, add to cart A002, pay A002, violating the associated-property consistency rule. The user bought item B and should not be counted in the complete conversion chain for the same item.

FAQ

Is low funnel conversion always a data problem?

Not necessarily. Step order, window, global filter, associated property, or time-zone configuration may correctly exclude the user.

Why is a user not counted even though they triggered all steps?

The property may be inconsistent, the order wrong, or the window exceeded; you need user-level diagnosis to verify item by item.

How does this differ from Retention Analysis Data Verification?

Retention verification resolves the retention-versus-event-analysis definition difference, while funnel misunderstanding diagnosis resolves rule understanding and user-level diagnosis within a single funnel model.