When your ROAS drops 20%, the first question is "why?" AI-powered metric decomposition automatically breaks changes into contributing factors - revealing whether it was CPM inflation, conversion rate decline, audience mix shift, or something else entirely.
What Is Metric Decomposition?
Metric decomposition breaks an aggregate change into its component drivers. If blended ROAS fell from 4.0 to 3.2, decomposition might reveal: CPM increase contributed -0.4, conversion rate decline contributed -0.3, and average order value increase offset +0.1.
Traditional decomposition requires analysts to manually build waterfall charts. machine learning analytics automate this process, running decomposition across all your metrics continuously.
How AI Enhances Decomposition
- Automatic driver identification: AI tests hundreds of potential contributing factors simultaneously, finding drivers humans might miss
- Confidence scoring: Each identified driver comes with a confidence score indicating how certain the attribution is
- Cross-channel correlation: AI identifies when changes in one channel drive changes in another (e.g., pausing paid search affects organic search volume)
- Anomaly contextualization: When AI anomaly detection, decomposition automatically explains what changed and why
Practical Examples
E-commerce ROAS Decline
ROAS dropped 25% week-over-week. AI decomposition reveals: Meta CPMs increased 15% (seasonal auction pressure), Google Ads conversion rate dropped 10% (landing page speed regression), but email contribution held steady. Actionable insight: fix the landing page speed issue first (highest controllable impact).
SaaS Lead Quality Drop
MQL-to-SQL conversion fell from 35% to 22%. Decomposition shows: LinkedIn targeting expansion added lower-intent audiences (contributing 60% of the decline), while Google Search lead quality remained stable. Action: tighten LinkedIn targeting criteria.
Implementing Decomposition
- Define your key metrics: Start with 3-5 top-level metrics that matter most to your business
- Map the driver tree: For each metric, identify the underlying components (e.g., Revenue = Traffic × Conversion Rate × AOV)
- Set up automated tracking: Ensure all component metrics are tracked at the same granularity and frequency
- Configure alerts: Set thresholds that trigger automatic decomposition when key metrics change significantly
How Xplainr Helps
Xplainr's AI automatically decomposes metric changes into contributing factors with confidence scores, turning 'what changed' into 'why it changed' across all your connected platforms.
See AI Diagnosis in Action
2-minute setup • No credit card required
Related Resources
- Understanding Changes
Use decomposition to explain why metrics changed.
- AI Confidence Scores
Interpret confidence levels in AI decomposition results.