UX Case Study · Self-initiated · 2025

StaySense

A memory layer for hotel booking. Capturing intent before booking and learning after the stay.

Domain: Indian OTA · Hotels
Context: MakeMyTrip, self-initiated
Role: Research · Strategy · UX
Designer: Joseph Anthraper

StaySense: capturing intent before booking and learning after stay

The problem

OTAs remember what you booked. None remember why.

The idea

Let users save the reason, then check it against the stay.

The bet

Validated reasons become data for travellers and for hotels.

01 · The problem

Travellers rebuild their reasoning every trip.

Traveller frustrated while comparing hotels

“During trip planning I had 11 browser tabs, a Notes entry that read ‘the one near the metro — quiet street, café on corner,’ and no memory of why I rejected the other seven. On the next trip I did it again. From scratch.”

The observation behind StaySense

80%

of travellers visit an OTA before booking, even if they book elsewhere.

Expedia Group, 2023

81%

OTA cart abandonment. Users research but don’t commit.

Condor Ferries, 2024

4 wks

Average trip research time. Platforms capture none of it.

TripAdvisor, 2023

55–60%India OTA share (MMT)

$9.8BGross bookings / yr (FY25)

36.4MMonthly active visitors

100M+Combined app downloads

02 · Insight

People already do this work by hand.

Before formal testing, I checked the instinct against 8 informal conversations with frequent travellers.

Pattern 01

They keep manual notes on why.

Notes app entries, screenshots, WhatsApp-to-self. Homemade memory systems to hold the reason a place made the shortlist.

Pattern 02

Or they re-research from scratch.

Without notes, they re-open listings they’d already evaluated and redo the same comparison to remember why they cared.

StaySense isn’t teaching a new habit. It’s absorbing one that already exists.

The behaviour is there. No platform has given it a home.

Scope note: n=8, convenience sample from my network. Directional signal to justify a prototype, not a substitute for the structured study defined under Metrics.

Platform	Saves listings	Captures why	Post-stay validation	Personalisation loop
MakeMyTrip	✓	✗	✗ shallow	✗
Booking.com	✓	✗	~ open review	✗
Airbnb	✓	✗	✓ structured	~ partial
Google Hotels	✓	✗	✗	✗
StaySense	✓	✓ Tagged note	✓ Binary prompt	✓ Full loop

The gap isn’t storage.
The reason never becomes data.

03 · The solution

A 4-part system that closes the loop.

Embedded in the existing booking flow. Scroll through the four steps; the phone follows.

Research Notes

Capture

While browsing a listing, users tap “Save Research Note” to open a bottom sheet. Select predefined tags like quiet area, walkable cafés, scenic surroundings, and add an optional note. Zero friction, inside the existing flow.

What caught your interest?

Tag + context

Tags let users structure their reasoning without effort. The “Add Street View” button saves neighbourhood context directly into the note.

Save Research Note

Confirm + save

Once tags are selected, Save activates. One tap saves the structured note (tags, free text and Street View context) to the listing. The reasoning is now data the platform can learn from.

Street View snapshot

Visual context

One tap saves the view by coordinates, not cached imagery, keeping it compliant with Google Maps Platform ToS. The snapshot becomes visual evidence of the user’s intent, tied to the note.

After the stay

Quick check-in

Binary yes/no prompts, not open reviews. Questions come straight from what the user noted before booking. Under 15 seconds. The return is personal: your answer improves your next recommendation.

The loop

Recommended for you

With enough validated signal, StaySense surfaces hotels matching the user’s criteria, with a transparent reason. Trust through explanation, not black-box ranking.

04 · Key decisions

Three calls that shaped the outcome.

Binary prompts over open reviews

Considered: Open-text review (existing model)
Chose: Yes/No prompts tied to the user’s own notes

Trade-offLower richness, but a target of 12–18% completion vs ~5%. Structured signal beats sparse narrative.

Tags over free-text notes

Considered: Free-text note field only
Chose: Predefined tags + optional note

Trade-offTags constrain expression but enable aggregation. Free text needs NLP; tags work on day one.

Street View via coordinates

Considered: User uploads own photos
Chose: Re-fetch by coordinates, never cache imagery

Trade-offMaps ToS compliant. Legally scalable, lower storage cost, no licensing risk.

The hard question

I diagnosed a motivation problem and initially prescribed an effort fix.

The contradiction

My problem statement says post-stay reviews barely happen. My solution depends on post-stay validation.

Reducing effort only works if effort was the constraint. For open reviews it’s partly motivation: writing a public evaluation for strangers returns nothing to the reviewer.

Binary prompts tied to the user’s own notes are different. The question references something they said, and the answer improves their next recommendation.

How I’d test it

Hypothesis: Binary prompts tied to the user’s notes achieve 3× completion vs a generic review (baseline 5–8%).
Method: A/B: generic 3-question review vs note-derived binary prompt. n ≥ 500 per arm, 30-day window.
Success: ≥ 18% completion on the binary arm (SMS survey benchmark 12–18%, FeedbackRobot 2024).
Guardrail: No increase in time-to-book for note-capture users.

05 · Success metrics

What we’d measure, and what we’d protect.

Targets are proposed, not measured results.

Metric	Type	Baseline	Target	Source / rationale
Note capture rate (5+ listing sessions)	Adoption	0% (new feature)	≥ 22%	Comparable to Airbnb Wishlist save rate
Post-stay validation completion	Adoption	5–8% open review	≥ 18%	FeedbackRobot 2024; binary format targets upper end
Repeat-session listing overlap	Efficiency	Not tracked today	↓ 25%	Direct re-research reduction measure
Recommendation CTR (returning users)	Business	AI personalisation in dev	↑ 15%	Personalisation CTR lifts in adjacent verticals
Shortlist → book conversion	Guardrail	Current MMT baseline	No decline	Must not add funnel friction
Time-to-book (note users vs control)	Guardrail	Current session length	No increase	Must not extend the booking journey

Sources: FeedbackRobot 2024 · MMT Q4 FY24 Earnings · Revinate Reputation Report 2024

06 · Business case

Not a feature. A data asset that changes both sides.

Consumer side

Users find what they want, and stop forgetting why.

Reduces cognitive fatigueUsers stop rebuilding criteria every trip. Research time drops on repeat sessions.
Increases booking confidence68% of customers pay more when they feel certain about their choice (Virdee 2024).
Drives repeat useThe platform becomes the memory layer. Switching cost rises without lock-in tactics.

Supply side: hotels pay

Validated expectation data is a product for hotels.

“73% of guests who shortlisted you for Quiet Area reported it wasn’t met.”

Illustrative example of the insight hotels would receive.

Operational insightNo star rating produces this.
Fix by expectation gapHotels prioritise by gap, not gut feel.
New revenueA B2B analytics dashboard beyond commission.

Personalisation loop

“Recommended for You”

Once the validation loop has enough signal, recommendations reflect criteria the user has confirmed.

↑15%

Rec. CTR target

68%

Pay more when certain

4 trips

To full personalisation

07 · Next steps

Three things before shipping. One thing I’d reconsider.

Validate the motivation gap

A/B test: note-derived binary prompt vs generic review. n ≥ 500, 30-day window. If completion stays below 15%, the feature needs a stronger value hook.

Interview 8–10 high-research bookers

Users who view 5+ listings in one session. Confirm re-research is universal or segment-specific.

Maps licensing check

Google Maps Platform restricts caching Street View imagery. The coordinate re-fetch approach needs formal legal clearance.

What I’d reconsider

Whether tags are the right primitive, or whether inferring intent from browsing behaviour (time on listing, scroll depth, return visits) beats asking. Passive capture removes the feature’s own friction entirely. It needs a separate prototype cycle to evaluate.

The next competitive layer in OTA is memory, not inventory.

The problem is real

Decision fatigue and re-research are documented behaviours in academic literature and OTA data.

The design is testable

Every assumption has a metric, a test and a success threshold.

The business case holds

Consumer personalisation plus B2B hotel analytics: two revenue angles on the same data.