From first UX hire to a corporate goal with headcount.

Every piece of work in these two years was paid for by the one before it.

1. IOne designer, no users, two allies.
2. IIQuestions before the demo. Contracts grew.
3. IIIDesign came in-house. Spend fell 26%.
4. IVA tenth caught the wrong scenario.
5. VTold not to bother. 200 applied.
6. AfterWhat it changed after I left.

One designer, no users, two allies.

Avalere Health is a healthcare policy and data firm headquartered in Washington, DC, owned by Inovalon since 2015. The software team had shipped a market-analytics product and was chasing feature requests. I joined in July 2017 as its first UX designer, working from a coworking space in Philadelphia. The team was spread across four states with the developers at headquarters, so nearly all of it happened over a screen share.

Nobody on the team had ever watched a user, and the objection I met was not about money. It was why anything needed to be easy to use at all, when staff could simply learn the tool. That is a fair question if you have never seen someone fail at one, and I had no recording to show anyone.

The intranet study

The team rebuilding the company intranet had no researcher, so I offered. The users were our own staff, so recruiting was easy. A Mural workshop agreed what staff needed to get done; then we watched five colleagues attempt ten tasks each.

The readout went to HR leadership as a teaching document that showed what user research is: what we did, what broke, what to fix first. It also answered the objection. Intuitive turns out to be a measurable thing. It means a screen matches what a person already expects of it, and a recording shows the second it stops matching. Finding a role inside a department failed for three of five people, so the first recommendation was to put department names on the org chart.

That also put the argument in money. The intranet existed to save staff time, so a task that does not match what staff expect gets paid for in minutes, by everyone, every week. HR became the first ally.

By November I had written the job description for a second designer and walked it through HR and the practice lead. It was approved in December, which is how the researcher arrived the following June.

The triangle and the swarm

I was looking for language that would make design stick, and the co-founder who owned the brand had already done the hard part. The company's values were set, tied to the brand, and lived, so anything I could attach to them would mean something. In November 2017 HR gave me a slot in the company's lunch-and-learn series, and I built the session on Avalere's own imagery. The logo is a flock of birds, and the value behind it is the goose formation: whoever knows the route takes the point, and the rest share the burden behind them. Starlings work the other way. No bird leads. Each one follows three rules, alignment, separation and cohesion, and the flock finds its route together.

The session put the two shapes on Cynefin's map. A complicated problem has an answer, and the right expert can find it, so the triangle is the right shape. A complex problem has no answer until you try something and watch what happens. A new product for a market nobody had observed is complex, and the triangle is the wrong shape for it, because the bird at the point has no more idea of the route than the rest.

The three rules became the deck's three moves: find alignment through experimentation, facilitate collaboration among separate stakeholders, and tell a cohesive story about human needs. We tested the framing on analysts and project managers before running it on site. Marketing, who own the logo, came on as the second ally.

Questions before the demo. Contracts grew.

The session gave the company a way to say when an expert should lead and when a team should swarm. The product that was not selling was the first chance to use it. The market-analytics product was a year past its first release. A market study had been commissioned at the start, so the team felt the research was done. Prospects kept asking for one more feature before they would buy: the shape of a build trap.

In the deck's terms the team had picked a leader and flown in formation behind it. The market study was the expert, its answers were treated as known, and each objection from a prospect was absorbed as a feature rather than read as a signal. The problem was complex and they were working it as if it were complicated.

I joined the demos and interviewed fifteen stakeholders to surface the assumptions behind the product and its business model. The buyers were busy executives, and the company wanted a finished product in the room. I asked to try it the other way round, and to put questions before the demo. Sales let me in as the inquisitive product person. In Cynefin's terms that is probe, sense and respond. In the deck's terms it was the swarm, with the prospect inside it.

Prospects were glad to talk about their own workflows, and the way the company made products changed. Before, it paid for market research, built the product, and went out to sell it. After, it demoed a dashboard built in Looker for the prospect's own questions, worked out internally which parts kept recurring, prototyped the next product in Looker, and only then built it. Contract sizes grew, and a new group formed to find the repeatable services that could become products. Sales had let design in, and product followed, which was the work I had been hired to do in the first place.

Design came in-house. Spend fell 26%.

The sales work traveled further than I expected. In January 2018 a reorganization moved me to the Inovalon parent with a promotion. In June a design researcher joined, recruited out of the Philadelphia design scene and working from the same coworking space as me while the rest of the company stayed remote. Together we set out to understand about a dozen product teams and bring the work outsourced to a design vendor back in-house.

Managers and tech leads liked the vendor: high-fidelity mockups, fast and consistent. They also said nobody was responsible for the whole client experience, or for tying it back to the measures the company ran on. Analysts wrote requirements, the vendor drew screens, and nobody in between asked why.

With each team we ran the Innovator's Compass, a one-page canvas for framing a problem before solving it. It gave us workstreams and testable hypotheses. Findings went into an atomic research repository, and every two weeks we ran two or three experiments against them.

The transition, in ten steps

Moving the vendor's work in-house without a gap took most of 2018. The order mattered: no business case before we knew the costs, and no costs before we knew the work.

1. Access, onboarding, and compliance
2. Stakeholder interviews across the products
3. Backlog creation and prioritization
4. Transition alignment with the vendor
5. Design support costs and equipment list
6. Job descriptions and compensation analysis with HR
7. The design team business case
8. Transfer of the design assets
9. Component library contribution and bug guidelines
10. Component library 1.0 release

The component library was the vendor's, built on Aurelia, and it later moved to Angular Material. Once the assets transferred we wrote the contribution and bug guidelines and documented the components in Sketch. We pinned every release to a git tag, so a product team always knew which version it had built against and could take the next one on its own schedule. A short-term and a long-term front-end governance plan, worked out with two engineering leads, decided who could change what.

The business case

In September 2018 I took the case to leadership with the workbook attached. The design vendor's contract capped at $27,136 a sprint and ran to $687k across 2018, for a five-role team. Four in-house designers replaced it: total design spend fell 26% in the 2019 plan while the model's delivery capacity doubled. The team was funded. The vendor contract ended, and two more designer requisitions were approved for early 2019.

I built the budget model as eleven formula-linked sheets, so leadership could change an assumption and watch the totals move. This is the sheet they decided on.

A tenth caught the wrong scenario.

This ran alongside the business case rather than after it. In April 2018 Inovalon had launched a service that reads medical records and finds conditions a chart review would miss. Prospects wanted to review the findings themselves. The product team arrived with the interface already designed: not built, but written out as a requirements document the size of a book, with a budget to match. The deadline was Client Congress, the annual client conference, at the end of September.

I pushed back. A tenth of the budget would buy a high-fidelity demo instead of production code, which we could change as fast as we learned, and it would leave room for a round of research first. Inovalon had about four hundred contract clinicians doing chart reviews, so the people we were building for were already inside the company, and we went to watch them work. We captured the team's assumptions in a business model canvas and held off on writing a scenario until the sessions had run.

What the requirements got wrong

The document had the reviewer accepting the first suggestion the system made, on the strength of a confidence score. The clinicians would not accept a finding they could not check against the chart, and said so in the first session.

Eight findings came out of the sessions. We wrote the scenarios from them, a day in the life of the reviewer we had now watched, and only then drew the screens and cut the demo. I redrew the interface around the first finding: a reviewer will not act on a model's confidence score until she can see the sentence in the chart that produced it. The score is a claim. The sentence is the evidence, and every reviewer we watched went to the chart before deciding.

What a tenth bought

The prototype cost about a tenth of what the build would have, and it changed the design before the other nine tenths were committed. Over two years the practice put every build it touched through the same tenth. By its own accounting, the rework that never had to be scoped came to an estimated $2.4M.

Told not to bother. 200 applied.

One product was enough to argue with. Scale was the next problem. After the ABILITY Network acquisition in 2018, Inovalon had dozens of development teams across four business units and two designers to serve them. Hiring our way out did not pencil. The AVP who ran one of the product lines told me not to bother trying. In that view, Inovalon was not the kind of place designers would want to work, so building the capability inside was wasted effort.

The strategy became growing practitioners inside the teams instead: a Design Center of Excellence that trains, mentors, and certifies people who already do the work, and lets teams decide when to hire a designer.

One empathy map on a skeptical stakeholder set the direction. The analysts had the domain knowledge; what they lacked was method. So we would train them.

Design entered the 2019 portfolio objectives in January, and by March the chief executive had directed two of his VPs to stand the center up as a corporate initiative. We kicked off on 22 March with founding practitioners from every business unit and a design-ops canvas: what the center does, for whom, and within what limits. Then we measured where the organization stood.

What the survey found

The survey turned up something we could use. Product and development managers already spent about a fifth of their time on design, and thirteen of them spent more than half. Those thirteen scored far above the rest on maturity. Ranking them against every product's stage, revenue and receptiveness gave us the first cohort of twenty.

Blueprinting the program

Manager interviews said what the cohort needed: how to write a testable hypothesis, design an experiment, and recruit users. So I blueprinted the Design Center of Excellence like an internal startup, with the five measures a startup would track. They were written before the first cohort, so that a failure would show up in a number someone already watched.

LUMA Institute would deliver two of the three training days and the playbook. Userlytics records and redacts protected health information, which let studies run under HIPAA. Glean.ly's alpha program gave us a research repository built for the job. With the partners settled, the schedule went in the plan.

The approved headcount went out as job postings that spring. I had worked the job descriptions over with other design managers around Philadelphia first, trying drafts on them until the scope was right and the role read as one worth wanting. More than 200 qualified people applied in the first 24 hours. The AVP was wrong about that, and I had been wrong about how hard it would be. I left in July 2019 with the core team hired and the first cohort on the calendar. The pilot ran that September and the first cohort started in October, both after I had gone.

What it changed after I left.

• A new enterprise-wide discipline. The AVP of Design Strategy role I wrote was filled that November, four months after I left, and the function ran under it into 2021.
• Design got funded from a model. The budget workbook outlasted the business case. Leadership could change an assumption and watch the total move, which is a different argument from asking for a number once a year.
• My manager moved into design. The senior director I reported to took over design strategy at Inovalon in 2020 and led it until 2021.
• A colleague became a client. A director from Inovalon's innovation group left in 2018 to found an AI text-analytics company. In 2023 that company hired me to design its brand and product interface.

Colophon · Names are withheld and quotes attributed by role. Screens are reconstructions with synthetic data, not screenshots.