Work

During the early stages of the creator discovery launch, search quality was assessed anecdotally, relying on client feedback and demo success rather than rigorous data. With an index of over six million creators, the relevance of the top twenty results for any query went unmeasured. This was a deliberate trade-off for a seed-stage team with limited product resources.

After the acquisition, Creator Search was the primary creator-sourcing engine inside an enterprise platform, which meant its quality needed to be measurable: a golden set of queries with agreed-correct results, an accuracy metric the team could interrogate, and an observability program to catch regressions before clients did.

The first measurement put demographic model accuracy at 72%. That baseline turned an open question into a prioritized backlog, with every failure case in the golden set becoming an item of work. The program closed at 93% accuracy, with latency down 68%. The same period shipped a creator brand-safety score on the GARM framework, reading image, video, and caption to flag risk at the point of discovery.

A proportional score treats every flagged post as equal and rewards a large, mostly clean feed. The question a brand is actually asking is closer to a gate: is there anything here that disqualifies this creator. One serious violation can be disqualifying on its own, however small its share of the feed, so the model needed severity to override volume rather than average against it.

The rebuild scored GARM harm categories weighted by severity, recency, and frequency, reading image, video, caption, and comments so that risk carried in a video or a comment thread was not missed by a caption-only pass. Brands set their own tolerances against the GARM categories, so the same creator can read safe for one brand and risky for another. An on-demand deep scan let a reviewer pull a closer read on a specific creator, and human review handled the contested cases the model flagged but could not settle.

The three tools operated on separate roadmaps with independent owners and conflicting client commitments. The investment case argued two things: that the portfolio's growth targets did not work with the offerings sold separately, and that fixing this meant stopping some funded work to make room. The second argument consumed most of the alignment effort, because a priority reset means walking back commitments already made by other owners, which had to be negotiated stakeholder by stakeholder.

We shipped a unified product that integrated our 9.5M-member sampling community with ambassador and paid creator workflows. Success relied on maintaining a consistent product definition across Sales, Finance, and four engineering teams. This required rigorous coordination to ensure that how we sold, priced, and built the product remained aligned throughout every planning cycle.

The build-versus-buy framework assessed long-term strategy rather than immediate needs. Payments wasn't just a one-off checkbox; within 18 months, we would need to handle sampling payouts, ambassador incentives, and affiliate commissions and a new vendor integration would have to be redone for each of them.

There was reasonable pushback. Finance preferred a specific vendor, Legal preferred the smallest possible compliance surface, and a credible vendor already existed for this exact use case, so taking on money-movement risk voluntarily required strong justification. The framework resolved it on a narrow point: the vendor route solved the immediate deal and left the roadmap use cases of payouts, incentives, and commissions, each requiring its own re-integration.

The build scope covered admin-led managed transfers, full transaction history, and a new compliance path defined with Legal and FinOps. After launch it addressed 28% of previously lost Affable deals and reopened 24 stalled commercial conversations, and the platform now serves as the payment rail under Integrated Sampling.

This transition was executed as a series of deliberate, incremental adjustments. First, I secured essential LLM tooling. By embedding AI prototyping into our scoping process, feasibility questions were answered before commitment. An agent prototype was built ahead of engineering capacity, accepted as an unresourced bet so the team would have a working reference point when capacity opened up.

The effectiveness of this model is best demonstrated by our work on Intelligent Product Tagging. Previously, accuracy stalled at 10% with no framework to diagnose failures. Under the new operating model, we built a 'golden set' of queries, analyzed failure cases, and treated every prompt revision as a testable hypothesis. This methodology drove accuracy to 60% within a year across 25 enterprise pilot clients, establishing a new default standard for the team's AI development.

When cinemas closed in March 2020, the company's primary revenue stream vanished overnight. We had to build and launch a transactional streaming product on web, mobile, and TV in a matter of weeks. The typical quarterly planning cycle was condensed into days, requiring us to re-evaluate our product priorities weekly as the market evolved. Navigating this volatility meant continuously refining our discovery and checkout experiences to maintain momentum.

The platform ultimately generated approximately 80% of the company's revenue, sustaining a 30% month-over-month growth rate while theatres remained closed. Furthermore, I optimized the customer experience by reworking the Help section and introducing contextual chat; this reduced support voice volume by 30%, effectively scaling our service capacity without additional headcount.