The work that earned the opinions.
NS Studio is a young studio. We publish only the work we can write about with full context, what we did, what we measured, what we killed, and what we'd do differently. New case studies added each quarter.
A pre-launch education company came to NS Studio with a mission and a domain name. They left with a brand system, a shipped product, and an AI-driven learning workflow validated against a real pilot cohort of 82 families.
<role>
Act as the founding team of a
pre-launch education company.
You have a mission, a domain,
and a budget. You need a brand,
a product, and a learning loop
that survives a beta cohort.
</role>
<task>
Ship the smallest version that
proves families come back for
week 4, not the demo.
</task>
<output>
82 families · 68% completion ·
3 features killed with data
</output>The prompt that ran the engagement.
The Dreamers Education was a pre-launch venture aimed at students underserved by mainstream EdTech, kids who don't fit the standard curriculum but have outsized capacity. The founder had a name, a hunch, and a small budget. What they didn't have: a brand, a product, a positioning, or any clear sense of how AI fit into what they were building.
They needed a partner who could do all four, and tell them honestly which corners to cut.
Five components. We deliberately resisted the urge to over-engineer. The discipline of saying "no, that's not in scope" was worth more than any model choice.
We ran a 10-week pilot with 82 beta families recruited through a parent waitlist. These are the four charts we presented to the founder at the end of the pilot, verbatim.
Conversion through the pilot funnel. The two numbers we cared about: did families show up, and did they come back the second week?
28% week-10 retention is strong for early-stage EdTech (industry median ≈ 12–18%). Source: Dreamers internal product analytics, pilot cohort.
The invisible competitor wasn't another EdTech app. It was worksheets and ChatGPT. Self-reported parent behavior, week 0 vs. week 8.
Parent-reported "primary tool used for at-home practice." Sample: 55 active families at week 8.
Parent-reported student confidence on a 1–10 scale, taken at intake and at week 8. Distribution shifted right; the long tail compressed.
74% of parents reported a 2+ point lift on the 10-point scale. Self-reported data, directional, not clinical.
Two-week A/B with 40 students, split evenly. Gamified version had streaks, points, and avatar rewards. Non-gamified had none.
Result: we shipped the non-gamified version. Engagement-as-clicks was the wrong North Star. n=40, two-week test.
We built the visual system in two weeks: name lockup, logo system, color (a warm, dreamy palette anchored in earned-not-given rust amber and serif typography that signals seriousness without feeling academic), tone-of-voice guidelines, and a launch-ready brand book. The constraint was clarity, students, parents, and teachers all needed to see themselves in the brand.
Core flows designed end-to-end: onboarding, learning module navigation, practice loops, and progress feedback. We made one opinionated call early, no gamification gimmicks, no streaks, no points, and then we proved it with the A/B test above. Trust was the differentiator.
We positioned The Dreamers around a single contrast: most EdTech optimizes for engagement, we optimize for outcomes parents actually pay for. That sentence drove the homepage, the pricing page, the email sequence, and every paid acquisition test. It also told us which channels to ignore (social spectacle) and which to invest in (parent communities, school partnerships).
The heart of the engagement. Five components (above), one non-negotiable: every model decision had to be explainable in plain English to a parent who'd never used AI before. The rule-based diagnostic layer exists because of that constraint, not in spite of it.
The hardest part of an engagement like this isn't the work. It's the ruthless prioritization. Every week, three things ask to be built. We ship one and tell the founder why we killed the other two.
Three features made it into the prototype and didn't make it to launch. All three had champions inside the team. None of them survived contact with usage data.
An earlier version of this case study said "two weeks to launch-ready." That was true for the brand system, not the product. Here's the real arc.
We over-invested in the brand book in week one. The founder needed less documentation and more momentum. Future end-to-end engagements start with the AI workflow, not the brand system, even when the client thinks they want the opposite.
We also waited too long to run the gamification A/B. We had a hunch in week three. We ran the test in week eight. Five weeks of avoidable arguing.
That lesson became part of the standard NS Studio engagement shape. It's why every engagement now starts with the assumption test, not the deliverable inventory.
We publish two to three case studies per year, with full context and client permission. Subscribe to be notified when new work goes live.
How we returned 4 hours per CSM per week without buying a single new tool. Coming soon.
Two pre-registered empirical studies on what AI is doing to executive prioritization at F500 scale. Open call for partner companies and research collaborators.
Drafting an AI use policy that legal, ops, and the union could all sign. Coming soon.
We work with a small number of clients at any one time, by referral or after a fit call. If the conversation is mutually right, we move fast.
Book a 30-min callMETHODOLOGY · Pilot ran Q1 2026 with 82 beta families recruited from a 143-person waitlist. Engagement metrics from internal product analytics. Confidence scores and behavior change are parent-reported via weekly surveys (n=55 active families at week 8). Gamification A/B was a two-week test with 40 students randomly assigned. Numbers rounded for legibility. Figures are illustrative of pilot performance, not a forecast. Published with founder permission.