Table of Contents

Budget Optimiser

The budget optimiser uses a market-mix modeling approach. As input data it uses spend data but can also use impression data or other input data. The model builds ad-stock representing the recent investments and models decay-rates, seasonalities, country holidays and other factors that might be important.

Historical performance

Response Decomposition Waterfall by Predictor

Here it shows the impact by variables. The variables are the variables that have been used as input data, eg. sourceMedium, campaign or channelgrouping.

Actual vs. Predicted Response

The actual response here is the actual revenue or conversions. The variable that has been selected as outcome. The predicted is the modeled outcome so one can see how well the model fits the actual revenue. Especially spikes can be hard to predict etc. But the more data the model has the better it can fit the model.

Share of Spend vs. Share of Effect with total ROI

This chart shows the Share of the total spend vs. the total effect and the ROI. Where the ROI is the return on investment considering the chosen outcome. If its revenue its a simple ROI but if its conversions it has to be treated more as a relative figure.

Response Curves and Mean Spends by Channel

This graph shows the Spend on X and the Response on Y considering the diminishing returns as the spend increases. The response curves are per channel.

Geometric adstock: Fixed decay rate over time

Shows how many % the investment decays over time. The time unit is weekly by default. As an example if decay rate is 30% the effect of the investment in the channel decays 30% per week.

Fitted vs. Residual

Shows the individual datapoints with the actual response (revenue) and the fitted datapoint from the model.

Budget optimisation and scenarios

This section gives the option to see an optimised budget and experiment with different scenarios and constraints.

Shows the mean of the previous months budget. The next months budget can be inserted, it uses the average of the previous months as default.

Lower bound of % of previous spend

The budget optimiser wants to allocate more budget to the channels with high ROI and less to the channels with low ROI considering the response curves, seasonalities etc. The lower bound sets the lower bound from the previous spend. So at 80% it only allows to decrease the budget for that channel with 20%.

Upper bound of % of previous spend

Maximum allowed % of previous spend.

Modeled optimised performance

It shows the optimised budget allocation taking into account the model and the given inputs.

Total spend increase is 0 if the budget is the same as previously. Total response increase shows how much the revenue or conversions would increase or decrease with the given inputs.

Initial vs. Optimised Budget Allocation

Here is shown the initial vs. the optimised budget allocation taking into account the constraints and return curves.

Initial vs. Optimised Mean Response

Initial vs. optimised revenue per channel.

Response Curve and Mean Spend by Channel

The initial spend and optimised spend by channel on the response curves.

Making adjustments

We only recommend making small budget allocation adjustments and doing them gradually so it gives time to test and see the effect and also leaves some room for seasonalities and other external factors playing in.

Importing offline advertising data

To import offline advertising data such as TV and radio spot plans we recommend using the Google Sheets connector.

Here is a sample format: https://docs.google.com/spreadsheets/d/102zMBiT5GeFwbJHhOkGI29jAgWow_FkAb158H4UDSsc

And here is the link to the Google Sheets connector: https://onboard.windsor.ai/app/googlesheets