In early 2018, I joined a sales-focused email marketing company as the first data person. I stayed for nearly four years. International markets were doing $20–30K in monthly revenue. Campaign operations were manual: each sendout took 4–5 minutes across six separate systems. Performance data was only available at the aggregate level, with no way to see deliverability split by email provider (Gmail vs. Outlook vs. Yahoo). The entire sending infrastructure was rented from third parties, which limited both control and margin.

Three bottlenecks feeding each other. Manual operations capped the volume of campaigns we could run. No provider-level visibility meant targeting decisions were guesswork. And dependency on external sending infrastructure meant we couldn’t automate even if we solved the first two.

Mercury: a campaign management platform that integrated six operational systems. Per-sendout time dropped from 5 minutes to 15 seconds, reducing campaign manager workload by 68%, and it produced provider-level deliverability data for the first time — solving the visibility problem as a side effect of solving the operations problem. I grew the team from 1 to 12 and designed the architecture they built and operated. An analytics warehouse made demographic-level segmentation (male boomers vs. female Gen Z) structurally possible for the first time. And I migrated sending from external providers to self-hosted infrastructure on our own IP range, saving €300,000 in the first year (~40% of COGS) with full control of the delivery pipeline.

The next step was full personalization: AI-generated subject lines, multi-armed bandit A/B testing, individualized newsletters from each recipient’s interaction history. I initiated development in early 2021. After a leadership change we disagreed on strategic direction (buy vs. build); I left rather than watch a direction I disagreed with play out. Mercury was still in use years later.