Recovery was harder than switching apps. So they switched.

Hired for streaming, into the one product built on it

I was brought onto Racing for a specific reason. TVG was the only place a horse racing bettor could watch the race and wager on it without leaving the platform, so a large share of the product experience was a live video problem. I had spent six years on that exact problem at NBA Digital, across sixteen apps and the web. That is the skill they were buying.

It also put me somewhere useful. Racing is a form of wagering older than the telegraph, taxed under its own structure and regulated under its own framework, and because of that it did not enforce location the way the rest of the business had to. So as I worked through the Core Product Experiences, I kept finding something I could not explain: sportsbook, casino and fantasy all carried a Location Based Issues threshold that everybody treated as ordinary. Not a defect being tolerated. A constant, with a floor nobody expected to move.

That is a historian’s position more than an analyst’s. I had a comparison case nobody else in the room had, and the comparison said the baseline was a choice rather than a law.

What actually counts as a primary source

Two things existed and did not speak to each other. The UX Research team ran a quarterly competitive benchmark scoring usability, trust and loyalty against the category, reported to executive leadership on a cycle. Customer Insights held the archive of what customers had actually said.

A primary source is the record that comes from the user. The complaint they filed in their own words. The feedback they left on a help article after it failed to help them. What they said when an error hit them during a research session about something else entirely. Not a dashboard, not a summary, not a satisfaction average. The account of a person who was there.

What I shipped in 2023 was the join. Respondents carrying verifiable user IDs were matched to their own case history for the same quarter and split three ways: no support history, unrelated support history, relevant support history. Where the detracting voices on a pillar turned out to share an experience, that experience could be scored for its measurable effect on the score, then multiplied against revenue and usage to produce a ranking. A survey number tells you a pillar is soft. The record tells you who it happened to and what they did next. Almost nobody runs them against the same person.

It interrupted a study that was not looking for it

The clearest evidence arrived uninvited. A researcher would be running an unrelated session, on some other part of the product, and the participant would hit the location error live. The session would stop being about whatever it had been about. That happened more than once, in studies designed to find something else, which is the kind of corroboration you cannot manufacture.

What the records named was not a geolocation tolerance problem. It was that the journey across a state border had never been considered necessary, so nothing had been built for it. When the token failed, the user was handed troubleshooting: move, retry, check permissions, read a help article, contact an agent who is barred by regulation from fixing it in the back end. That path is longer and less certain than opening a competitor’s app and rebuilding the betslip from scratch.

So many did bet somewhere else. They told other fans on Twitter and Reddit that they would rather use FanDuel, but that the location error meant they bet with a competitor instead. The industry as a whole had spent years over-engineering the acquisition journey to betslip and had not built a recovery journey for when something as load-bearing as a location token fails prior to conversion. Priced against handle and netted to revenue, the estimate came to $42M a year, and a significant share of it went directly to the nearest competitor.

Only six per cent of the people who hit the error ever contacted support, but around sixty per cent of those were new users. Were we training our customers not to contact us over a known bug?

A quarterly excuse for poor scores

The Login score had been on the board for four quarters and moved nothing. Not through neglect. Three reasons were offered every cycle, and each one was defensible on its own terms.

• The check is a compliance control, so the friction it causes is the cost of operating legally. Not a risk worth taking.
• Nobody owned the number, because the requirement sat above the login portal, sitting above every product, and login gets barely looked at since it measures no revenue directly.
• Support had already absorbed it into a script, which is the most dangerous of the three: friction somebody is quietly handling every shift stops looking like friction at all.

Underneath the three sat the real cause. Every state launched independently, with its own regulator, its own tax rules and its own governing body, because that is what the law required. So the organization was built one state at a time, and the journey across a state line belonged to nobody, while the people using the product average more than one home.

An elevated role to support greater access

A finding this size does not travel on its own merits. It travels on who is carrying it. As the Racing partner I could name the pattern; I could not convene the four product organizations that would have to act on it, or the compliance and regulatory people whose approval any fix depended on.

Moving to Lead Voice of the Customer Strategist changed the audience and the remit rather than the headcount. Reporting went to executives and heads of product because the problem spanned products and operations, and the scope grew to include the Voice of the Employee and the Voice of the Department alongside the customer’s.

The finding went up through my own leadership first. Then I presented it to executive leadership myself. That part of the job is not research.

Claims at this scale demand source integrity

One number from one instrument is an opinion with a decimal point. So the study was built to see whether independent sources, gathered for unrelated reasons by people who were not talking to each other, told the same story. More than eight of them did.

A dedicated analyst ran the data and I set the questions it had to answer. The qualitative half I did myself, because it is the half most easily dismissed and the hardest to hand off. Every vertical supplied a stakeholder, and their role was to be a source rather than a collaborator: what they held was the account of why a known problem had gone unshipped, which is evidence in exactly the way a customer’s complaint is evidence.

UX Research retired the benchmark campaign as it had been defined, and stayed in the study anyway, supplying archival evidence of the same problem bleeding into several of their other studies. An independent team’s unrelated work arriving at your conclusion without meaning to is the best single source in the set.

The course changed

A score with no cause is a number anyone can explain away, and for four quarters anyone did. The same score carrying a population, a behaviour, a competitor and a dollar figure stopped being a quarterly line item. Two roadmap items were accepted and prioritized, priced at $27M and $7M.

It was never a product decision alone. A location check is a compliance control, so easing it is a regulated change, and one of the two items required rewriting regulatory wording before the technical fix was permitted at all. A story that arrives as “users complain about login” does not survive that room. One that arrives with eight independent sources behind it, four product teams already in the study and a number attached does.

What a researcher controls is not delivery. It is whether the evidence is good enough that the decision stops being an opinion. On this one it was, and the organization changed course on drag it had spent four quarters agreeing to live with.