Live Demand

Active content had grown to 42 items, but new releases still represented only 7% of sales. That result told me that adding supply alone would not give customers a stronger reason to buy.

Weekly repeat purchases per buyer were 2.13 for a top product and 3.39 for a real-time product. I saw a signal in that gap: repeat demand appeared alongside the need to check what was happening right now, not just the content itself.

I reframed the revenue decline as a gap between a customer’s current concern and a relevant product they could buy—not a shortage of content.

I believed we needed to connect the moment a reason to buy appeared with the right product, rather than simply make more content.

One option was to avoid any overlap with other product shelves. I chose to prioritize products that fit the customer’s current concern instead of the operator’s taxonomy.

I replaced the open text field with four visual concern cards. The goal was to help customers recognize their situation and move forward without first deciding what to type.

GA purchase events recorded within five minutes of an interaction · before and after the contextual-fit change, Weeks 7→8

93 → 199

FORCE attributed within five minutes

70,410 → 180,600

In Phase 1, the team and I turned the first post into a guided concern-selection entry point. I designed the recommendation structure around four concerns, three depth-and-price tiers, and seven products per tier.

In Phase 2, the team used the concern a customer selected in the post to show relevant products across 12 event pages and in-app messages.

A key placement was missed during Weeks 3 through 7. Placement, assortment, messaging, and format also changed from week to week, so I could not compare that period on product performance alone.

That miss gave me a new rule: verify exposure before interpreting performance. It became a principle for my later decisions, but not a company-wide standard.

Across the adjacent weeks around the card-format change, viewers increased by 10.6% and interacting users grew from 3,130 to 15,879. During the same period, GA purchase events within five minutes of an interaction rose from 199 to 660, and attributed FORCE rose from 180,600 to 532,910.

Both the number of people who engaged with the post and the number of purchase events increased. However, purchase events as a share of post interactions fell from 2.70% to 1.73%. The post drew interest from more people, but the share that went on to purchase did not improve.

In Week 9, higher-priced, more in-depth products represented 21.5% of attributed purchase events but 40.4% of the FORCE used in those purchases. FORCE per purchase event was also 3.40× higher than for lower-priced products. This product group generated fewer purchase events, but each purchase used more FORCE, so I saw value in testing offers for customers seeking more depth alongside lower-priced options.

User interaction rate

5.16% → 23.69%

GA purchase events within five minutes

199 → 660

Purchases per interaction event

2.70% → 1.73%

FORCE attributed within five minutes

180,600 → 532,910

Here, an attributed purchase means a GA purchase event recorded within five minutes of a post interaction. It does not directly measure a product-detail visit or an order-level conversion.

This operational data has no control group, and post IDs, copy, assortment, and exposure conditions also varied. Without user IDs, order IDs, or cancellation and refund definitions, I do not use it to claim incremental revenue, unique payers, or repeat-purchase rate.