How a Founder Fixed Their Sleep — and What Changed at Work

Meta description: A composite case study of how a founder restructured sleep habits using the sleep science literature, with specific changes to scheduling, evening behavior, and AI-assisted planning.

Tags: founder sleep, sleep optimization, productivity case study, sleep scheduling, AI planning

This is a composite case study. The character, company, and details are constructed from patterns common among early-stage founders, not drawn from a specific individual.

At eight months post-launch, Kemi Adeyemi was running a 12-person B2B fintech company and sleeping — by her own account — “whenever work allowed.” That usually meant five or six hours on weekdays, sometimes seven on Sundays if she had not spent the night before preparing for a Monday board update.

She was not dramatically impaired. She was high-functioning, as founders go. But she noticed a persistent pattern: her best strategic thinking happened in short windows, usually in the morning, and became increasingly unreliable after lunch. She was making smaller errors than she expected in financial models — not calculation errors, which her tools would catch, but judgment errors about assumptions. Her responses to difficult emails were sometimes blunter than she intended.

She read Hans Van Dongen’s 2003 study in Sleep after someone sent her a summary. The data unsettled her. Six hours per night for two weeks produces cognitive deficits equivalent to two full nights without sleep. She had been averaging about that for eight months. The part that landed was the dissociation finding: people on six hours consistently underreport their own impairment.

She decided to treat this as an experiment.

Note: Nothing in this case study constitutes medical advice. Kemi does not have a diagnosed sleep disorder. The interventions she implements are behavioral, not clinical.

The Baseline Assessment

The first step Kemi took was to establish a real baseline. For one week, she logged:

• Actual time in bed (not target time)
• Approximate sleep onset (the time she stopped looking at her phone and stopped actively thinking)
• Wake time (alarm or natural)
• Subjective energy level at three points during the day: 9 a.m., 2 p.m., and 6 p.m.
• One cognitive task that felt unusually hard that day

This was not a formal sleep study. It was directional data collection.

What she found:

• Average time in bed: 5 hours 50 minutes
• Variable wake time: ranging from 6:15 a.m. to 8:20 a.m. depending on whether she had an early meeting
• Sleep onset: frequently delayed because she was finishing Slack threads or reviewing investor materials until 11:30 p.m.
• The afternoon energy drop was consistent and significant, usually between 1:30 and 3 p.m.
• The tasks she rated as hardest were almost always ones requiring judgment under uncertainty — investor memos, hiring decisions, product prioritization discussions

The pattern was clear. Irregular timing. Late-night work eating into wind-down time. A chronic sleep onset delay driven by task spillover.

She also identified what Roenneberg would call social jetlag: her natural wake time on unscheduled mornings was around 7:30–8 a.m., but most of her week had her up by 6:30 a.m. She was, as he describes it, biologically asleep during her early morning work hours.

What She Changed and Why

Kemi’s interventions were sequenced deliberately. She had read enough implementation research — primarily through Phillippa Lally’s habit formation work and the CBT-I literature on behavioral sleep interventions — to know that changing everything simultaneously produces abandonment, not optimization.

Week 1–2: Fixing the anchor.

She set a fixed wake time of 7:00 a.m. — later than her previous alarm-driven 6:30 a.m., but aligned more closely with her natural chronotype. She held it for 14 consecutive days, including weekends. This meant that when she had late nights, she was still up at 7:00.

The first few days were difficult. Sleeping until 7:00 on a Tuesday felt like an indulgence she did not deserve. She noticed that the discomfort was almost entirely psychological — her calendar had space, meetings rarely started before 9, and the notion that 6:30 was the virtuous time was a story she had absorbed from a startup culture that treats early waking as a proxy for seriousness.

By day 10, she was falling asleep more easily. The circadian anchor was working: consistent wake time pulls the sleep onset time earlier gradually, without requiring direct intervention.

Week 3: Hard stop on evening work.

Kemi implemented a 10:00 p.m. hard stop on all work tasks. Nothing requiring cognitive effort after that time. She scheduled this as a non-negotiable calendar block labeled “close.” It ran from 10:00–10:30 p.m. and consisted of three things: reviewing tomorrow’s calendar, writing down any open loops she was worried about forgetting (this was Masicampo and Baumeister’s Zeigarnik work in practice — getting the unfinished tasks externalized to reduce rumination), and setting her phone to Do Not Disturb.

She used Beyond Time to track whether she actually completed work before the cutoff. The tool made visible what she suspected: on three nights in the first week, she had continued working until 10:45 or 11 p.m. despite the block. Seeing this pattern concretely — rather than estimating it from memory — made the specific triggers identifiable: late-day investor communication and async product reviews.

She restructured those tasks to be completed before 9 p.m. by scheduling protected response windows earlier in the day.

Week 4: Environment adjustment.

She addressed temperature. Her thermostat had been set at 71°F. She dropped it to 67°F in the bedroom, which felt cold for the first two nights and unremarkable thereafter. She also got blackout curtains — her bedroom faced east and morning light had been waking her at 6:15 on clear days, regardless of her alarm setting. The curtains removed that variable.

What Changed at Work

By week six, Kemi noticed three distinct changes:

The strategic window extended. Her morning cognitive peak — which had previously lasted about 90 minutes before she felt the quality drop — extended to approximately 2.5–3 hours. This was not a dramatic transformation. It was a measurable, usable increase in the duration of her best thinking.

The afternoon was recoverable. The energy trough between 1:30 and 3:00 p.m. was still present — this is a normal circadian phenomenon for most chronotypes — but it was less severe, and she had started scheduling lower-demand tasks there. The trough being predictable made it plannable.

Fewer second-pass corrections. She noticed that she was less frequently reviewing decisions she had made under time pressure and feeling they were off. The investor memos and product prioritization calls felt more calibrated. This is difficult to quantify objectively, but she had a concrete proxy: the number of times per week she sent a revised version of a message she had already sent decreased from five or six to one or two.

Using AI Planning to Maintain the System

Kemi started using Claude as a weekly planning partner for sleep-adjacent scheduling. Her approach was practical rather than elaborate:

On Sunday evenings (before the close block), she would review the week ahead and ask Claude to flag any patterns in her calendar that were likely to compromise her 10 p.m. hard stop — late meetings, commitments that had been scheduled without buffer time, or days where she had not protected her morning peak window.

An example of the kind of prompt she used:

“Here is my schedule for the coming week. I have a 7:00 a.m. wake time and a 10:00 p.m. hard stop on work. Flag any commitments that look likely to either start too early relative to my target wake time or push me past 10:00 p.m. Also flag any days where I have not protected a two-hour block in my morning peak window (9–11 a.m.).”

This was not a revolutionary use of AI. It was systematic: using Claude to audit a calendar for structural problems that Kemi could not always see herself because she was too close to the content.

She also used Beyond Time alongside this process to compare planned schedule to actual schedule weekly, which meant she could see when her implementation was slipping before it became a habit.

What Did Not Work

Not every change produced the expected result.

She tried a formal wind-down ritual — 20 minutes of reading followed by brief journaling — but found that the journaling activated rather than settled her thinking. She replaced it with light reading only.

She also tried going to bed at 11:00 p.m. (from an average of midnight-ish) without first anchoring the wake time. This produced no meaningful improvement in sleep quality or daytime performance. The sequencing mattered: anchor the wake time first. The bedtime shift becomes much easier after the wake time is stable.

The Relevant Principle

Kemi’s case is not remarkable because the changes were complex. They were not. A consistent wake time, a hard stop on evening work, a cooler room, and blackout curtains — none of these are sophisticated interventions.

What made them work was treating them as structural, not discretionary. She applied the same logic to sleep architecture that she applied to business processes: if it is not scheduled and enforced, it does not happen consistently. If it does not happen consistently, the benefits do not accumulate.

For founders specifically, the temptation is to treat sleep as the last variable — the thing you protect once everything else is under control. The research suggests the opposite: sleep is the substrate on which the quality of everything else depends.

Related reading: The Sleep Optimization Framework | The Complete Guide to Sleep and Productivity Science | 5 AI Prompts for Sleep Optimization