Learn Everywhere — the delivery model Thought Industries debuted at its annual Cognition conference — wasn’t on any roadmap. It came out of three design sprints I facilitated back-to-back, fully remote, in the three-week window before the next planning cycle locked in a year of work. Over the three years that followed, the company went from $8M ARR to roughly 2× year-over-year growth.
The Challenge
Thought Industries had spent real money on JTBD research and had the customer-effort scores to prove where the pain lived. None of it was reaching the product. The team was running feature-driven — engineering picked up whatever the loudest stakeholder asked for — and the JTBD report sat in a folder.
The risk was concrete: more user-centered competitors were starting to win deals on experience.
Three Sprints, Not One
I’d facilitated design sprints before, as a consultant. Looking at the JTBD map, no single sprint was going to get us out of feature mode — the painful steps in the customer journey clustered into three different problems, and cramming them into one sprint would have produced mush.
I proposed three back-to-back sprints, each owning one cluster:
- Pulse — read signals about what learners were actually doing, inside and outside the LMS
- Atlas — deliver learning where the learner already was, instead of forcing them back into the LMS
- A third sprint focused on authoring tools, which would ship on its own track
I co-facilitated with a PM — ex-IBM consultant who’d run sprints at this scale before. Her track record gave the three-week timeline its credibility with leadership; I owned the method and the design output.
[VISUAL: the JTBD-by-project grid — 12 steps of customer education clustered into the three projects, showing how the sprint scope came out of the research, not opinion]
The Visual That Became Learn Everywhere
We’d just wrapped Atlas, and Pulse was the next sprint that same week. By the end of Pulse — which finished early — the room could feel the size of the mountain we’d just mapped. Three sprints, three feature areas, and no single story yet for what we’d actually built.
My Chief Product Officer pulled me into a meeting that afternoon. “Show me a visual of how Pulse and Atlas work together. The simplest version. Something the whole company can get behind.”
[TODO: one sentence about what you almost drew first and why you ditched it. Shape: “My first instinct was a layered architecture diagram; I threw it out when I realized nobody outside product would read past the first box.” This sentence is the difference between “lucky shot” and “designerly decision.”]
This is what I drew:
[VISUAL: the Pulse + Atlas composite — Customer Health Score and Pulses on the left reading from Content / Support Tickets / Product Analytics; the central column showing those signals surfaced as learning content; Suggested Content with topic priority on the right]
Pulses on the left, reading signals from inside the LMS and from the tools around it — content engagement, support tickets, product analytics. Delivery on the right, surfacing the right learning to the right learner at the right moment.
That single picture was the version of the story the company could repeat. It became Learn Everywhere — the thing we announced at Cognition, and the product direction that anchored the next three years of work. The third sprint shipped its own outcome on a separate track: Merlin, a templated authoring flow. But the headline feature came from the two sprints nobody had planned to connect.
What We Cut to Make the Sprints Work Remote
The standard 5-day, in-room sprint format would have killed participation by day two. Everyone was Zoom-fatigued.
I rebuilt it. A Trello board became the source of truth for which activities had to happen live, which moved to async pre-work, and which got pushed to async after. Miro held the actual sprint artifacts. Live time got reserved for what genuinely needed the room: the expert interview, the dot voting, the decider call.
[TODO: one paragraph naming a specific activity you cut and why — or one you cut that you later wished you hadn’t. The HM’s interview question was literally “what’s an activity you cut that you later wished you hadn’t?” — answering it here pre-empts it.]
Planning a sprint used to take me a full day — eight hours of agenda design. With the new format I was getting it done in two. That 75% cut was what made running three sprints in three weeks possible at all.
[VISUAL: the Trello sprint planning board — columns for kickoff, day-by-day activities, retro, with cards tagged live vs. async]
How a Sprint Actually Ran
Each sprint opened with a 20-minute Ask the Experts session — I’d interview an internal expert while everyone else wrote How Might We statements live. From there: theme the HMWs, vote on the two themes worth solving, run lightning demos against competitors and adjacent products, sketch the user’s flow, place the HMWs onto the flow to expose hot spots, individual wireframing, dot vote, decider call.
That rhythm ran three times. The Atlas sprint was the biggest — 14 participants including three C-suite — which mattered later.
[TODO: one sentence-or-quote from a participant — an engineer, a CS person, the CPO, or even a line from a sprint retro. One outside voice keeps the piece from reading like a single POV.]
Impact
- Learn Everywhere shipped and was the marquee announcement at Cognition, the annual customer conference.
- The sprint method became the company’s default for product discovery, with planning time held at ~2 hours per sprint instead of a full day.
- Revenue trajectory: the company went from $8M ARR to roughly 2× year-over-year growth for three years following these sprints.
I won’t claim three sprints alone drove three years of growth. What I will claim: the product direction those three years were built on came out of those three weeks — and Learn Everywhere, the thing customers actually talked about at Cognition, wouldn’t have existed without running the sprints in series instead of in parallel.
What I’d Do Differently
Two things, with hindsight.
Invest in the algorithm, not the UI. A version of the underlying recommendation engine — suggestions pulled from three signal sources — was already working when we shipped. We led with a heavy UI to get to market. Pre-LLM, that was a defensible call; today I’d have spent that effort on the algorithm and kept the UI light enough to swap.
Build the goal layer. Customers never got a way to set a goal against the learning they were delivering — and that gap is still there today. Signals only mean something if you’ve defined what success looks like. We built the signal layer without the goal layer underneath it, and you can’t write a real hypothesis without one. That’s the discipline I’d add next time.
Role: Sprint facilitator & design lead · Client: Thought Industries · Timeline: November 2022, plus ~3 years of follow-on work · Tools: Miro, Figma, Maze, Trello