Digital Sleepwalking for Market Researchers: How “Agency Under Fatigue” Warps Signals—and What To Do About It

From Matt Gullett at Between Silicon and Soul

The midnight purchase that broke my crosstab

A client once asked why our concept test “spiked with Gen Z after 11 p.m.” The brand team was thrilled. The ops lead was suspicious. The next morning, returns on the same SKU jumped 18%. Same respondents, same product—two very different stories.

What happened? Digital sleepwalking: people acting online while tired enough that memory encoding and executive control dip. They still tap, like, comment, and buy—but in a state of agency under fatigue. Your instruments register behavior. Your follow-ups often capture… shrugs.

This post is the market-researcher’s playbook for naming it, measuring it, and designing around it.

Why it matters (beyond one weird night in your data)

Recall bias on steroids
If respondents can’t remember a late-night decision, post-hoc “why” data collapses. Your beautifully worded “drivers of choice” question is suddenly modeling confabulation.
False product/creative positives
After-hours clicks can look like preference. In the morning they behave like noise (returns, unlikes, unsubscribe).
Algorithm drift you can’t see
Sleepwalk actions still train recommendation engines. Your “culture read” from social listening inherits those twilight artifacts.
Brand trust risk
If consumers feel they were nudged while half-awake, you’ve traded short-term lift for long-term suspicion.

The BetweenSiliconAndSoul lens: a meta-skill for researchers

Guarding agency under fatigue is a research meta-skill. It asks:

When was the choice made?
In what cognitive state (fatigued proxy)?
What’s the right method for that state?

Your job shifts from “measure behavior” to “measure behavior in context.”

Field implications you can act on now

1) Instrument time and state—don’t treat all clicks equally

Tag daypart (local time, not GMT). Create AM/PM cohorts and a “late-night window” (e.g., 11:00 p.m.–4:00 a.m.).
Add simple fatigue proxies (self-rated alertness 1–5, last sleep hours, session length) at intercept or diary entry.
Sessionize: Was this a first-session purchase or the end of a multi-day consideration path?

What changes: your crosstabs graduate from “who liked it” to “who liked it when awake.”

2) Triangulate “said” with “did”

Pair stated recall with behavioral trace (transaction logs, event pings, cart open/close).
Use cued recall (screenshots, cart contents, creative stills) to refresh memory before asking “why.”
Add a “next-morning check”: recontact a small sample that converted at night and ask intent again.

What changes: you stop asking tired brains to reverse-engineer last night’s motives from scratch.

3) Rethink success metrics for after-hours activity

Report “conscious conversion rate” (midday) vs “twilight conversion rate” (late-night).
Track regret markers: 7-day returns, “accidental purchase” codes, delete/undo rates, next-day unlikes.
Create a Sleepwalk Index: (Late-night conversions that reverse ÷ total late-night conversions). Trend by cohort.

What changes: your KPIs stop rewarding behavior that won’t stick.

4) Design studies that meet respondents in situ

In-the-moment mobile prompts (lightweight, <30s) triggered by cart close, like/share, or add-to-list.
Digital diaries with time-locked entries (no backfilling) and a “morning reflection” question.
Micro-RCTs: For big-ticket choices at night, test a “sleep on it” confirmation vs. immediate checkout and measure long-term satisfaction.

What changes: you capture cognition before it fades, and you test interventions brands can adopt.

5) Ethics that build trust (and better data)

Identify, don’t exploit: no dark patterns in the twilight window.
Offer “confirm in the morning” or “send me a reminder” options for major actions.
Report daypart effects to clients candidly. Short-term lift that erodes trust isn’t a win.

What changes: respondents feel respected; cooperation and data quality improve.

Platform literacy (the stuff that ruins great decks if you miss it)

Timestamp truthing: “Purchase time” ≠ “decision time.” Pull cart first-seen and first exposure where possible.
Timezone hygiene: Always convert to local wall time; mixed-time analyses quietly break patterns.
Session decay: If a “late-night” order began at 6 p.m., treat it as daytime deliberation + nighttime execution.
Auto-plays & infinite scroll: Distinguish active vs passive exposures; fatigue inflates passive counts.

Excel/PowerPoint → Interactive deliverables (so clients use this)

Add three controls to your dashboards:

Daypart slicer (Morning/Afternoon/Evening/Twilight)
Fatigue flag (self-rated ≤2, long session, short sleep)
Stickiness trio (return/cancel, next-day like, 7-day usage)

And a one-click “Conscious-only view.” Let stakeholders see how much of their “win” survives the filter.

Big prompt fail → the five-card fix (for AI-assisted analysis)

Prompt fail:

“Summarize why Gen Z bought SKU-14 last week.”

Why it fails:

It assumes stable agency and flattens daypart/state differences. The summary will happily average contradictions.

Five-card fix: (paste these as bullet “cards” before your question)

Context card: “Segment events into Day (6a–10p) vs Twilight (10p–4a), local time.”
Construct card: “Treat Twilight as potential reduced-agency; weight stated rationale from Twilight by 0.5 unless corroborated.”
Confounds card: “Flag sessions >20 min continuous scroll as fatigue risk.”
Join card: “Join conversions to 7-day returns and next-day unlikes; compute Sleepwalk Index.”
Output card: “Report separate drivers for Day vs Twilight; include stickiness deltas and ethical considerations.”

Now ask:

“Given the cards, what are top purchase drivers that persist after 7 days?”

Cross-generational, bi-directional learning (skip the stereotypes)

Gen Z can teach teams in-situ capture and UI dayparting; they know the terrain.
Gen X/Boomers can teach commitment safeguards and longitudinal stickiness.
Blend both: rapid, contextual capture + slower validation.

A practical 90-day plan

Days 0–30 (Instrument & baseline)

Add local-time daypart and alertness micro-item to intercepts/diaries.
Build a Sleepwalk Index for one priority funnel.
Create the Conscious-only view in your dashboard.

Days 31–60 (Method & metrics)

Pilot next-morning recontact on late-night converters (n≈150).
Run a micro-RCT: “confirm tomorrow” vs “buy now” for high-consideration items.
Update reporting to show stickiness deltas by daypart.

Days 61–90 (Scale & ethics)

Roll daypart analysis across major trackers.
Publish an internal guideline: when to adjust weights or exclude twilight noise.
Co-create with the client a Trust-by-Design checklist (no dark patterns at night; reminders for big actions).

Checklists you can paste into your next brief

Instrumentation

Local wall-time stamp
Daypart label incl. Twilight window
Self-rated alertness (1–5)
Session length; first exposure time
7-day return/unlike/usage join

Analysis

Conscious vs Twilight cohort cuts
Cued recall where memory is needed
Sleepwalk Index trended by cohort
Drivers of persistent behavior, not just conversion

Ethics

No exploitative night-window UX
“Confirm in morning” tested for big decisions
Transparent reporting on daypart effects

Bottom line

Digital sleepwalking isn’t a moral failing; it’s biology meeting design. If we keep analyzing twilight actions as if they were fully conscious preferences, we’ll keep mistaking noise for signal—and recommending tactics that won’t stick.

Measure the when and the state, not just the what. Guard agency under fatigue—for your respondents, your clients, and your reputation.

Next in the series: What executives and brand teams should change in product and UX to build trust (and better LTV) in the twilight hours.