Apple's Sports App Is Barely a Data Product. That Is the Point

The sports technology market is sold as a data problem. Recent headlines talk about performance tracking, ticketing analytics and AI in sports, all of it pointing at bigger pipelines and smarter models. One of the most used sports products of the past two years goes the other way.

Slate reviewed Apple's Sports app in June under the headline that it "does almost nothing." The piece describes an app with no news headlines, no streaming guides and no design features meant to keep users inside it. What it has is team names and logos, records, game times and scores of games in progress or completed, plus a standings page. That is the product.

The cost sits in the query, not the index

The contrast with how analytics vendors frame the problem is sharp. A post published in August by Cassis, a data tooling company, lays out a simple framework for where the human cost of answering a data question lands. It splits that cost into indexing, the upfront work of structuring information so it can be found later, and querying, the effort at the moment someone actually needs an answer. The post borrows a wardrobe analogy from Hrishi Olickel: schema-on-write is putting clothes straight into a wardrobe, sorted by function, while schema-on-read is sorting through a messy pile when you need something.

Apple's app is an extreme indexing bet. Someone had to decide which leagues matter, how standings are structured, how box scores are laid out, and how fast a score propagates to a lock screen. The user pays almost nothing at query time. Slate's reviewer notes the app was repeatedly surfacing Pittsburgh Penguins developments during the Stanley Cup Playoffs before the streaming broadcast caught up, enough that automatic updates had to be turned off.

"We created Apple Sports to give sports fans what they want, an app that delivers incredibly fast access to scores and stats."

That quote from an Apple executive, cited in the Slate piece, is the entire pitch. No claims about agentic anything. The reviewer lists the app's coverage as World Cup matches plus MLS, the NWSL and roughly 30 other leagues, with Ecuadorian Serie A and second-division Bundesliga included, alongside men's and women's tennis and the LPGA Tour.

Agents move the cost, they do not delete it

Most of the current sports analytics push assumes the opposite trade-off: accept a heavier product and get more back. That assumption is under pressure across data tooling generally. The Cassis post argues that AI agents do not remove the human cost of making data useful, they redistribute it, and that the redistribution is easy to miss. A senior data manager quoted in the piece described a weekly rotation in which one person is dedicated to ad-hoc questions, consuming about 20% of a team of five or six people, permanently.

Starburst published its own analysis of agentic data analysis in September, and it lands in a similar place from a different direction. The company describes an option it deliberately calls "option 0": extracting raw data from source systems and sending it to an agent as files or a direct pipe. Starburst's objection is cost, noting that sending terabytes of data to an agent can get prohibitively expensive if the agent charges per item processed. Its preferred option is to give the agent credentials to the database and let it write its own SQL, so the database engine does the work it was built for.

Starburst also flags the operational risk in that approach. Agents, the post says, can be far more demanding than humans and can overwhelm a database with speculative queries as they iterate. The company frames this as a workload problem the database industry is actively working on rather than a solved one.

Small tools, small surfaces

The same instinct is visible in smaller open source releases. Adaca Analytics, a self-hosted dashboard project for Google Analytics 4 published on GitHub in September, runs on Cloudflare Workers with daily rollups from the GA4 Data API or a BigQuery export landing in a D1 database the operator owns. Realtime stays on Google. The repo advertises six dashboards out of the box, a four-step builder, and precomputed detail pages so that clicking a number returns in one round trip. It is MIT licensed.

DataZen, shown on Hacker News in late August, takes a comparable line on scope for database work: a desktop client under 15 MB, GPLv3, no account, with credentials AES-256-GCM encrypted in the OS keychain and AI requests going only to the provider the user configures. Plainoldanalytics, posted in September, is a Go middleware that mounts a dashboard on an existing HTTP router and flushes traffic to disk roughly once a second.

None of these products promise to answer the CFO's question about which enterprise customers renewed. They promise that the answer, once someone defines it, is cheap to retrieve. Cassis puts the unbudgeted part plainly: getting four teams to agree on what "active customer" means is indexing work, and it does not go away because an agent is now writing the SQL.

For sports products the analogy holds with one difference. Fans already agree on what a score is. The definitional argument that eats analytics budgets does not exist, which may be exactly why doing almost nothing works.