What I learned building products in twelve weeks at Thrive Global

The first working prototype of the Thrive Global app sat on my phone on a November evening, several weeks after I had joined Arianna Huffington's new company. What I remember most clearly is the stopwatch feeling: twelve weeks from an ideation whiteboard to a live product that real users would open. The team had to figure out what a company about well-being should ship as its first object in the world.

I joined in 2016. The company launched on November 30, 2016. Our internal team was small, and the product suite spanned mobile, web and voice. I brought in Very, a development partner I had worked with before, to help us move at the speed the calendar required.

In the first three weeks, Very built a prototype for testing the flow of content from the publishing service Medium into an administration interface called Rails Admin. The team could try the workflow while we were still deciding how the broader platform should work.

Thrive's purpose was to help people deal with stress and burnout. Its initial products included articles, a marketplace, courses and a voice-guided routine for winding down before sleep. Building that range meant deciding where existing services would serve the mission and where we needed our own software.

A mission forces a product decision on day one

A company about burnout, sleep and healthier work habits cannot ship a product that itself produces more anxiety. Yet with a feature list and a launch date in front of you, the usual pull of consumer product work is to add notifications, streaks, scores and nudges. Every feature that makes an app stickier, by the standard playbook, is a feature that also makes it more extractive of the user's attention. We had to ask whether an interaction helped people live better or merely kept them interacting with us.

Deciding what engagement to build was also an editorial question. Thrive was simultaneously a media property, a content platform and a product. Arianna's point of view on attention was specific and clear, and it shaped what we said yes and no to. It was the first time I had worked on a product where the founder's position on attention itself was a design input. The mission mattered in what we left out.

Reduce scope before increasing pace

When the runway is short, reduce the scope before increasing the pace. More work in progress gives an already busy team more dependencies and decisions to carry. A focused first version lets the team test the product's assumptions while there is still time to change what it is building.

Twelve weeks is long enough to ship a well-made v1, but a week without a decision is a week the next week inherits. We set a rhythm of small decisions, made quickly, expecting to revisit some after launch and to discover that others were wrong. Accelerating before reducing scope leaves more work in flight, more people blocked and more decisions pending at the moment you can least afford them.

Working with an outside build partner

Bringing Very in was the right call for the situation. An internal team of our size would not have shipped mobile, web and voice in twelve weeks alone. Staffing up fast enough would have meant hiring before we knew what the products should be. A development partner gave us capacity to build in parallel while we kept the product voice centralized.

Very helped us prioritize the initial scope and test our assumptions. Their processes fit how we wanted to build and roll out the platform. I value that contribution as much as the capacity to write code.

They also treated the well-being constraint as real. If the partner had pushed us toward standard engagement patterns, we would have spent the runway arguing about first principles instead of shipping. Choose partners whose default instincts already align with the product's purpose. You will not have time to retrain them on the job.

Put the connections in the prototype

Medium's API, a software interface for accessing that content, gave the team a way to bring it into Thrive's platform. Very used it with Rails Admin so the Thrive team could test the content flow and get used to the workflow before launch. Very's account of the project describes this early testing as a way to identify potential problems while there was still time to address them.

A demonstration of each component on its own would tell you less about whether the combination suited the people doing the work. With the connection in place, those people could try the process instead of being asked to imagine it from a proposed design.

The choice also separated two kinds of work. Thrive needed a way to publish and use content; it did not need to create an entire publishing system to learn whether its own workflow made sense. The early custom work connected an existing capability to the product we were building.

The same distinction applied elsewhere. Shopify supplied the commerce platform, and Teachable supplied the course platform. Those choices gave the team foundations for a marketplace and a course with video, reading assignments, and progress tracking. They did not decide how the whole offering should work for a Thrive user.

I would not turn that into a rule to buy every available component. If a service forces an important part of your product into the wrong shape, connecting it may leave you with more work than building that part yourself. The useful comparison includes the work of joining the pieces and changing them later, not just the feature lists or the amount of code you can avoid writing at the start.

A voice product in 2016 was a bet

The Alexa skill was the piece of the suite that felt most like reaching. Voice in late 2016 was interesting but not yet proven as a sustained consumer surface. We built it because it fit the mission. Asking a smart speaker how to get a better night's sleep is different from opening an app about it. A guided routine could help someone wind down without putting a screen in the bedroom. Whether voice was the right distribution bet at that moment is a separate question. I am still not sure. A mission-driven company sometimes ships something that the market timing may not quite support because it is true to what the product is for.

Sam McDavid, one of Very's developers, described the engineering behind that interaction. The user needed to follow a sequence of guided steps before an audio file played. Thrive had supplied a script and wanted to be able to revise it. Sam and his development partner represented the conversation as a sequence of states: where the user was in the routine, what response to give, and where to go next.

Their design kept the wording in a configuration file, separate from the code that moved the conversation along. It also used Alexa's own response format within that configuration, reducing the need to translate a separate custom format into something Alexa understood. Sam reported that a product advisor could make minor wording changes without changing the application code. The team could test the progression through the conversation as well as individual inputs and outputs.

The team used an existing voice platform while building the behavior the product needed. They also considered how someone would revise that behavior. A small wording change and a change to the conversation's logic did not have to become the same engineering task.

What I took with me

I left Thrive Global after a short period to lead product and technology at The Wall Street Journal. The twelve weeks stayed with me in the work that came after.

Constraints surface real values faster than retreats do. A mission statement gets tested by a launch calendar. If your team cannot agree on what to cut when the clock is short, your team does not yet agree on what the product is for.

The best partners act like they work for the mission. The Very team treated the well-being constraint as a design principle. That is why we could ship.

Include the people who will maintain it

I wanted our internal team to be able to maintain the platform after launch. That requirement belongs in the decision about how to build it, alongside the first release's features and deadline.

For a team evaluating a similar project, I would ask for a demonstration of a likely change as well as a demonstration of the working product. Change an instruction. Follow a piece of content through the publishing workflow. Show how a test tells you whether the intended behavior still works. Then identify who will be responsible for that work after launch.

Those twelve weeks put our product in front of real users. We could now learn whether it helped them, and change the parts that did not. The team’s ability to test and maintain the product would determine how much we could learn from that first release.