Contacts Upload

Analytics showed drop-off at every stage, but the most damaging issue was a false signal. After selecting a file, a green checkmark appeared on screen. To users, that looked like confirmation the upload was complete. Many closed the browser and moved on, never knowing they still had mapping and consent steps ahead. The flow had an invisible exit built into it.

Beyond the false completion signal, the mapping interface was visually noisy and hard to parse. Help content was scattered rather than surfaced where it was needed. The consent step appeared as a blocking interruption that broke forward momentum. Error states were silent. Each issue was individually minor. Together, they made completing the flow feel like work.

The redesign stayed within the existing architecture. I removed the false completion signal and replaced it with progressive disclosure that made the next step explicit. Primary actions after file selection became more visually prominent. The field mapping interface was simplified for clearer column relationships. Help documentation moved from scattered placements to contextual inline guidance. Consent became an inline modal instead of a blocking step. Error states were rewritten with clear explanations and recovery paths.

None of these were large surface changes. The bet was that the cumulative effect of removing small friction at every step would meaningfully improve completion, and that the false completion signal alone was doing enough damage to move the primary metric.

The experiment ran as an A/B test through Statsig from December 16, 2025 to January 27, 2026. Both the control (2,201 users) and test group (2,261 users) were users who reached the file upload step, so baseline rates are higher than the general trial population.

The redesigned flow shifted users toward deeper contact uploads. 3+ contacts improved 3.0%, and downstream sending for power users (3+ emails) improved 2.2%. Those upstream gains cascaded into a 3.0% lift in trial-to-paid conversion. The clearest causal chain of all three experiments: better uploads led to more sends, which led to more conversions.

False completion signals are worth naming as a specific category. They don't generate support tickets. Users don't report them as bugs. They just leave, convinced they finished something they didn't. You can't see the exit in the data because it looks like a successful session.

This experiment produced the most balanced improvement of all three. The causal chain was visible at every step: deeper uploads led to more sends, which led to more conversions. Each step improved because the one before it did. That kind of cascading effect is uncommon in growth experiments and validates the principle that reducing friction at a high-intent moment unlocks downstream behavior.

The approach proved sound. The team continues to apply this friction-reduction framework across other complex workflows in the product.