AI Sports Offer Automation

Every week, sports operations teams were hand-crafting ticket and merchandise offers — comparing last year's games, checking weather forecasts, guessing at what fans would respond to. The data existed to do this scientifically. Nobody had built the tool yet.

Setting the Scene

FEVO is a sports commerce platform connecting teams with fans for ticket and merchandise sales. The core workflow was manual and inefficient: operations teams of 10-20 people spent days creating and optimizing offers for each game. There was no systematic way to factor in all the relevant variables — opponent strength, season timing, weather, inventory, day of week — so teams fell back on intuition and precedent. The revenue they were leaving on the table was significant.

The Problem

Offer creation was subjective by necessity. The same game situation generated wildly different offers across teams — and even across seasons for the same team. Data existed but wasn't being used. Teams wanted guidance but distrusted black-box algorithms. The design challenge was building an AI system that was accurate enough to outperform human intuition and transparent enough that teams would actually use it.

What the Research Revealed

• →Manual offer creation was highly subjective — identical game conditions generated vastly different offers from different teams
• →Data existed everywhere but teams weren't using it — they wanted AI guidance but not at the cost of their professional judgment
• →Offer type mattered: perceived value (% off) consistently outperformed absolute savings ($ off) in conversion
• →Context variables were enormous: same offer worked differently based on opponent strength, day of week, weather, and rivalry status
• →Teams wanted to understand the 'why' behind every recommendation before committing it to production

The Approach

I built an AI offer assistant that analyzed historical data, comparable games, and current inventory to recommend specific offers. Crucially, every recommendation came with its reasoning — not just 'here's the offer,' but 'here's why: games with this opponent strength and this inventory level historically converted 23% better with this structure.' Teams could override anything. But they rarely needed to.

Proving It Out

Two-month pilot with 5 teams using mixed AI vs. manual offers. Historical backtesting of AI recommendations against actual past games. Conversion rate tracking across every AI-recommended offer. And close monitoring of how often teams actually accepted or overrode the recommendations — the adoption rate was a proxy for trust.

• →AI offers achieved 3x improvement in conversion rate vs. manually created offers — not marginal, transformative
• →Teams using AI recommendations saw 12% average revenue increase per game
• →Offer creation time dropped from 3-4 hours to 15 minutes — the time savings alone justified the investment
• →Adoption jumped from 45% to 85% once we added reasoning to recommendations — that single change was the product

Beyond the Numbers

• →Teams reported genuine confidence in offer decisions for the first time — data-backed instead of gut-feel
• →Operations teams freed from manual offer creation had time to focus on strategic work instead
• →Created competitive advantage for early-adopting teams vs. those still working manually
• →Became the standard offer creation tool across professional sports organizations using FEVO

What I Took Away

Explainability is the most critical design decision in AI adoption for conservative industries. Sports teams wouldn't touch a black-box recommendation — but they'd trust a transparent one with comparative reasoning. Sacrificing 5% model accuracy for 90% user trust was the right tradeoff, and it wasn't even close.

• →Comparative reasoning is more convincing than statistical accuracy — 'similar situation → this offer → this result' lands better than 'our model predicts X'
• →Domain expertise matters for AI credibility — understanding sports (opponent strength, rivalry, home/away) made recommendations feel specific to their world, not generic
• →A/B testing results drive adoption. Showing teams real performance data from their own AI recommendations created a flywheel of trust
• →Human override capability isn't a fallback — it's a trust mechanism. Teams adopted more, not less, because they could override

What's Coming Next