Non-Revenue KPI Challenges

[claire-danaher-dept]

Scenario:

• Non-Revenue KPI - Small in magnitude compared to spend. An average value of 50 outcomes per record.

• 6 paid marketing channels ~ average spend across all channels per record of about 25K

• 4 organic marketing channels

• 1 Control Variable - Unbranded Query Volume. SEO for the company is limited and the average rank for unbranded queries is 23rd.

• ROI Priors- Paid marketing priors set to percentage of spend

  Non-revenue KPI configured meridian model with the non-revenue KPIs produced nonsensical output. As an alternative strategy, an average revenue value was multiplied by the KPI to generate a revenue value. This value was run with the same data using a revenue configured KPI meridian model and produced reasonable outputs.

Question: What type of underlying assumptions within the non-revenue KPI configured model would cause this type of behavior? It is my understanding that standard scaling is applied fro marketing channels but perhaps not for KPIs. Are there best practices for scaling that might have resolved this issue rather than adjusting the methodology to a revenue based approach?

[gousesb]

Hello @claire-danaher-dept ,

Thank you for contacting us!

The non-revenue KPI configured Meridian model likely produced nonsensical output due to a magnitude mismatch between the KPI and spend, which violates the model's core assumptions about ROI priors.

Meridian’s ROI priors are designed for a scenario where the outcome and spend are in comparable monetary units. This makes the ROI a unitless ratio. When you run a model with a non-monetary KPI (50 outcomes) and a large monetary spend ($25K), the ratio becomes an incredibly small number. This extreme difference makes the model unstable, as it can't find a meaningful relationship between the tiny KPI values and the large spend values.

Meridian handles internal scaling, so it's recommended to use priors that align with the scale of the target KPI to ensure comparable results. The best practice for using a non-revenue KPI in Meridian is to provide a revenue_per_kpi value in your input data. This explicitly converts your KPI to a monetary value and resolves the unit and magnitude mismatch. However, if there's no clear monetary unit associated with the KPI and revenue_per_kpi cannot be reasonably estimated, it's advised to use Custom priors when the outcome isn't revenue. If setting a custom prior based on your use case isn't feasible, consider using the default total paid media contribution prior.

Feel free to reach out for any further queries or feedback regarding Meridian.

Google Meridian Support Team