Case study · AI, data & growth
Automating Regional Ad Targeting
I built a targeting flywheel that aligned regional ad spend with partner coverage and revenue per lead, helping paid marketing grow revenue by 28% in 2023.
The context
We advertised across dozens of verticals in several countries, but regional targeting was adjusted manually. As the number of campaigns, verticals, countries and regions grew, these adjustments did not happen often enough. That meant some spend went to regions where no active partner could accept the resulting leads.
The challenge
I set out to keep ad targeting automatically and profitably aligned with the regions where active partners wanted leads, down to postcode level.
How I approached it
I developed Automatic Region Targeting, or A.R.T., by connecting two existing sources of data: our platform knew the revenue generated by each consumer request, while our advertising platforms knew its acquisition cost. A.R.T. calculated margin by region, compared expected revenue per request with actual cost, filtered out regions below our ROAS threshold, and fed those decisions back into bidding and targeting.
I increased the update frequency so targeting could respond when partners paused activity, reached daily or weekly limits, or changed their preferred regions. Because every country used postcodes, we could replicate the approach for each new vertical and market. Later, I incorporated refund predictions to make the revenue estimate for each lead more accurate.
What it delivered
Regional changes moved from occasional manual work to frequent automated updates. Consumers were no longer directed to regions without available partners, and the margin we freed up could be reinvested in additional reach. Paid marketing delivered 28% revenue growth in 2023, almost all of which was attributable to A.R.T., with dedicated tROAS and tCPA optimisation layered on top.
Looking back
I learned that the strongest automation can come from connecting data that already exists. Combining revenue and acquisition cost at postcode level turned occasional manual adjustments into frequent decisions based on actual margin. Adding refund predictions later also showed me that an automated system is only as useful as the accuracy of the assumptions behind it.
I carry forward a simple principle: optimise growth around real operational capacity, not reach alone. Targeting worked because it responded when partners paused, reached limits, or changed their preferred regions. In future roles, I will keep looking for scalable feedback loops that align customer acquisition with availability and profitability.