Skip to content
home

Services

Know what your marketing really delivers, and where the next pound should go

We build Bayesian mix models on your data, show where the evidence is strong and where it isn't, and turn the result into a budget decision finance will accept.

The problem

Attribution and platform reports tell you where conversions were recorded, not what caused them. A mix model looks at all your marketing, and everything else that moves sales, together.

What you get

  • Readiness assessment

    Whether your data can support the model, and what to fix first.

  • Model specification

    Channels, controls, seasonality and priors, documented and agreed.

  • Two engines

    PyMC-Marketing and Meridian on the same data, disagreement shown.

  • Budget scenarios

    Reallocation at the same budget, with the expected uplift and its range.

  • Calibration plan

    Which channel to test next, and how.

  • Handover

    A board-ready readout, and the model handed over in your tools, or in CausalPanda once it opens.

Where revenue came from

Separate marketing's contribution from the sales you'd have had anyway.

Where revenue came from, last 52 weeks
  • Baseline
  • Media channels
  • Total
Illustrative: total outcome decomposed into baseline and channels, with credible-interval whiskers on each channel.£0£10m£20m£30m£40m£33.4mBaseline£4.2mSearch£3.5mTV£1.8mSocial£1.2mVideo£400kRadio£44.5mTotal

Where the next pound goes

Reallocation at the same budget, with the expected uplift and its range.

Recommended allocation, same £9.8m budget
  • Current
  • Recommended
Illustrative: optimiser output as a dumbbell: current spend (steel) to recommended spend (forest) per channel, with the percentage change.£0£1m£2m£3m£4m£5mPaid search+35%TV-15%Paid social-13%Online video+82%Radio-11%

How an engagement runs

  1. Discovery call

    The decision and the constraints.

  2. Diagnostic

    A fixed-scope look at where you are.

  3. Delivery

    Agreed scope, regular check-ins.

  4. Handover

    Documentation and training.

FAQ

Questions

Ideally two years or more of weekly data, with spend that varies. We'll check before you commit.

Let's talk about the decision you need to make

Thirty minutes, no slides.