Data Analysis
Anomaly DiagnosisAE Analysis Model Selection
Not sure which AE analysis model to use? Intelligently recognize analysis intent, recommend the best-fit model based on real tracking metadata, and output a precise event and property configuration.
Segment breakdown
by groupAE Analysis Model Recommendation and Configuration is a data-engineering capability in ThinkingAI's industry Skill library. When an analyst is unsure which model to use, it intelligently recognizes analysis intent, recommends the best-fit analysis model based on real tracking metadata in AE, and outputs a precise event and property mapping, avoiding skewed results from a wrong model choice or misconfiguration. This is ThinkingAI's decade of industry experience, packaged into ready-to-use Skills.
Live walkthrough
pvp_battle, register, payment, guide_step are all in the tracking planguide_step lacks a step-identifier property; provided both an add-tracking and a SQL optionAll four requests mapped to a model: one each for distribution, interval, event, and funnel, with every config validated against real tracking metadata. Request (1) I already configured and ran end to end.
1. Model mapping for the four requests
| Business question | Recommended model | Why not another | Key config |
|---|---|---|---|
| (1) PvP-count difference, paying vs. non-paying | Distribution analysis | Event analysis only shows totals and averages; comparing "distribution shape" needs distribution | Event pvp_battle, grouped by tag "payment status" |
| (2) Interval from registration to first payment | Interval analysis | A funnel only shows conversion rate, not "how long it took" | Initial event register, subsequent event payment |
| (3) Change in core-gameplay participation after version | Event analysis | A trend-comparison scenario, no distribution or cohort logic involved | pvp_battle triggering users, grouped by #app_version |
| (4) Drop-off at each tutorial step | Funnel analysis | Retention looks at return visits, not step sequences | guide_step ordered by step, 1-day window |
2. Request (1) already produced a result
Distribution analysis result: paying users have a median of 5 daily pvp_battle counts, non-paying users 2; by shape, 61.3% of paying users fall in the 4–8 range, while 57.8% of non-paying users cluster in 0–2. Core-gameplay engagement depth is strongly correlated with payment, so the operational hypothesis of "drive PvP participation → drive payment" has data support.
3. Request (4) has a tracking gap
The guide_step event has no step-identifier property, so a funnel can't strictly order by step. Two options: add a step_id property (recommended, solves it once and for all); or temporarily approximate step order via SQL sorted by #event_time, which produces numbers, but the step order will be distorted after any tutorial-flow redesign, so it's only fit as a stopgap.
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
AE offers 10+ analysis models, yet more than 60% of analysts spend over 30 minutes agonizing over model choice for a specific need. Worse is the hidden bias from a wrong choice: the same business question yields completely different definitions under different models, and in about 40% of wrong-model cases the analysis conclusion is off by more than 30%. Configuration is just as subtle: the right model with the wrong event or property still distorts the result.
What it does
When to use it
The analysis need is clear but you don't know which AE model to use
You know which model to use but aren't sure how to configure events and properties
Multiple models could meet the need and you need to compare and choose the best
Existing tracking fields don't meet the analysis need and you need completion suggestions
Guiding a new analyst to get up to speed on AE model selection and configuration
In the field
FAQ
Does this Skill recommend the model or configure it?
Both: it recommends the best-fit model first, then outputs the configuration (event names, property names, grouping dimensions, filters) based on real metadata.
What if a tracking field is missing?
The Skill clearly flags the missing field and provides a SQL-query alternative or a tracking-completion suggestion.
How do you choose when several models apply?
The Skill lists each model's strengths and limitations and recommends the best option based on the analysis-goal priority.
Related Skills
Equip your Agent with AE Analysis Model Selection
Book a demo and see how it works in your own business.
