← case study
1 / 24

ROVER · PRODUCT DESIGN · 2025–26 · IN FLIGHT

Relationship-Based
Graduated Fees

Aligning sitter fees with how and when Rover delivers value — to curb diversion and grow per-relationship revenue.

Live
Canada · March 2026
~40%
higher take potential
US
targeting June '26
01 / NARRATIVE

Same fee.
Very different
value.

Rover charges a flat 20% (15% in Europe) on every booking — whether it's the first ever with a new client or the 20th with a long-term trusted relationship.

Our internal assessment found Rover delivers ~76% of booking value on a new relationship — but only ~18% on an established one. The fee doesn't reflect this.

BOOKING #1 — NEW CLIENT

Discovery, trust-building, insurance, payment infrastructure, profile visibility. Rover earns its cut.

20% ~76% value delivered
BOOKING #20 — TRUSTED RELATIONSHIP

Owner and sitter know each other well. Rover is mostly the payment rail and safety net. Same cut anyway.

20% ~18% value delivered
Social listening confirms: sitters feel they're getting less value over time while still paying a consistently high fee — especially on repeat relationships.
02 / CONTEXT

The numbers
shaped the test.

Top sitter earnings concentration
73–77%

Top 20% of sitters generate 73–77% of total earnings — consistent across all geos. The top 1% alone account for 15–19%.

North America: earnings from repeat clients
59–63%

Top NA sitters are repeat-driven. In the US: 63%. In Canada: 59%. This is the behaviour we wanted to protect and extend.

Source: Analytics deep dive · Sep 2025 · David Argelich

Repeat vs. acquisition sitter split by geography

Repeat vs. acquisition split — NA vs. Europe

03 / WHAT WE TRIED

The blunt tool
didn't work.

A previous test cut the sitter fee from 15% to 5% for all recurring bookings in the UK and all dog walking in Spain. After 4 months:

OVERALL DIVERSION
No change
STAT. SIG. SIGNAL
DDC only
+16.9% recurring units
4 LEARNINGS THAT SHAPED THE NEXT TEST
Test in North America

Canada is a closer proxy for the US than UK/ES in terms of repeat booking patterns.

Include boarding-type services

Boarding accounts for 53% of top US sitter earnings. High ABV services are where the fee pain is most acute.

Design for net neutrality

A flat fee cut is a pure margin give-up. The next test needs to be financially neutral or better — higher early fee offsets the reduction on repeat.

Target the right sitter segment

Include both new and established sitters. Previous test lacked segmentation to read the signal clearly.

04 / HYPOTHESIS
BASED ON

Sitter feedback and the fact that top providers get more than half their earnings from repeat bookings — while still paying the same fee as on a first booking.

WE BELIEVE THAT

By setting up a fee structure that more closely maps how and when Rover delivers value to sitters

WE WILL

Increase top sitter satisfaction and increase Rover take.

SITTER SUCCESS

Most loyal sitters are better off financially and satisfied with the new structure.

ROVER SUCCESS

New relationship booking volume is maintained — CBR not significantly hurt.

05 / STRATEGY

Three models. One test.

Click to explore →
New vs. Repeat model exploration
Model A New vs. Repeat

Lower fee on all repeat bookings, higher flat on all first bookings with any client.

✓
Relationship-based GBV exploration
Model B · Selected Relationship-Based GBV

Fee drops as cumulative booking value with a specific client hits milestones.

✓
Account-based GBV exploration
Model C Account-Based GBV

Fee drops as the sitter's total annual Rover earnings hit platform-wide milestones.

✓
COMPREHENSION

High mental model match. Sitters track clients, not annual totals. 'This client, this relationship.'

GAMING RISK

Moderate. GBV threshold mitigates booking-split gaming better than count-based thresholds.

VERDICT

Best match to sitter mental models. Testing first in Canada.

06 / EVALUATION

How they scored.

Dimension New vs. Repeat Relationship-based GBV Account-based GBV
Comprehension
Fairness match
Gaming resistance
Diversion reduction
Revenue upside

↑ Relationship-based wins on fairness match — the dimension most directly tied to diversion behaviour

07 / THE INSIGHT
Sitters think about fees per client,
not per year.

Account-based is operationally cleaner — one annual GBV number, simpler engineering. But when we talked to sitters, every grievance was relational: "I've cared for this family for 3 years and I still pay 20%." Relationship-based maps to how unfairness actually feels.

Concept test finding
28

Sitters across 4 segments confirmed: the per-client framing resonated more than an annual platform total.

Account-based risk
High

Tenure-only incentives don't address diversion — sitters could use Rover to acquire clients then move them off-platform as their annual GBV grows.

Account-based future
Layer

Backlogged as a complementary loyalty track once the relationship model is validated. Not abandoned — sequenced.

08 / THE MODEL

The 3-tier
relationship model.

TIER
RELATIONSHIP GBV
SITTER FEE
Tier 1
$0 – $499
30%
Tier 2
$500 – $999
15%
Tier 3
$1,000+
10%
  • Per relationship — each owner–sitter pair tracked independently
  • GBV = service total (sitter rate + fee, excl. owner fee & tips)
  • Permanent — milestone unlocked means lower tier forever
  • Pro-rated fee when a single booking crosses a tier boundary
  • Backfilled at launch — sitters started at their already-earned tier
SIMULATE A RELATIONSHIP — $120/BOOKING
30%
TIER 1
$0 – $499
15%
TIER 2
$500 – $999
10%
TIER 3
$1,000+
$0
RELATIONSHIP GBV
30%
CURRENT FEE
$500
$1,000
Pro-rated fee applied — part of this booking was at 30%, the rest at 15%.
09 / TIMELINE

18 months,
start to live.

From the first analytics deep dive to a live pilot in Canada, then the US next. Click any milestone to expand detail.

Sep 2025
Analytics deep dive published
David Argelich published the sitters' earnings distribution analysis — top 20% generate 73-77% of earnings; NA top sitters are repeat-driven (59-63%). This shaped everything that followed.
Oct 2025
Design exploration + model evaluation
Bernardo published the general design exploration and the alternative monetization evaluation framework. Three models assessed: new vs. repeat, relationship-based GBV, account-based GBV.
Oct–Nov 2025
Concept test — 28 sitters, 4 segments
Lizeth Herrera led qualitative concept testing across new, growing, established-loyal, and established-diverting sitters in Canada. Relationship-based GBV selected as the live test model based on findings.
Nov 2025
Deep-dive pages published
Bernardo published the relationship-based deep-dive (Nov 13) and the account-based deep-dive (Oct 31) — detailed analysis of mechanics, gaming risks, threshold options, and UX implications.
Dec 2025
Fee transparency pre-req + workshop
Fee transparency A/B test designed as a pre-requisite — to isolate the impact of showing sitters their payout before adding the tier structure. Cross-functional monetization workshop held with PM, design, analytics.
Jan 2026
Landing page + comms design
Bernardo shared landing page skeleton and Figma prototype. Brand Studio (Defne Kavas) delivered first round of copy for the landing page and launch email.
Feb 2026
MVP relationship page spec finalised
Bernardo published the relationship page MVP spec (Feb 9) — 3 core questions the page must answer, full MoSCoW component priority. Engineering planning begins.
Mar 4, 2026
🚀 Smoke test live — 4 Canadian cities
Launched in Nanaimo, Regina, Saskatoon, and Québec. Email + push notification sent to treatment sitters. 6+ bugs surfaced in the first week; all fixed within 2 weeks.
Mar 24, 2026
🚀 Full Canada test — Vancouver, Calgary, Ottawa
Scaled to three major Canadian cities. Ongoing monitoring via DataDog and Mode dashboards. Social listening active. Results pending at end of 8–10 week test window.
Apr 2026
Training/Grooming edge case resolved
Average training relationship GBV is ~$500 — trainers would be stuck at 30% permanently. Decision: exclude Training and Grooming from the graduated take. Flat 20% applies; they don't count toward progress.
Jun 16, 2026
US test — Chicago, Dallas, Seattle
Tentative. Conjoint analysis for US tier values pending. Training and Grooming excluded from day one. Plan to update future bookings to lower tier fee immediately (unlike Canada).
Q2 2026
UK MVP
TBD. European market has a different acquisition vs. repeat split — learnings may not transfer directly. Translation and localisation work required before any European expansion.
10 / SCOPE

9 user stories.
One cohesive system.

Every surface a sitter touches was considered — from the first announcement to a booking that crosses a tier mid-stay. Click any story to expand.

US1 · Fee logic
US2 · Contacts page
US3 · Relationship page
US4 · System messages
US5 · Earnings breakdown
US6 · Service settings
US7 · Announcements
US8 · Landing page
US9 · Payment history
US3 · RELATIONSHIP PAGE
The source of truth for per-client GBV progress

5–6 sections: header (Rebook + Profile CTAs), progress (tier visualisation + progress bar + banner), upcoming bookings, completed bookings, archived, and pets. Entry points: Contacts tab, conversation header, conversation menu, and booking earnings breakdown. Dark green = completed GBV; light green = pending GBV.

User story design
11 / KEY ARTIFACT

The Relationship
Page.

Without it, the fee model is invisible. Sitters can't verify their tier, track progress, or understand why a fee changed on a specific booking. It's not a nice-to-have — it's what makes the model real.

3 NON-NEGOTIABLES

Sitters must know how and when they'll gain more value

New and existing sitters must know they're rewarded for keeping bookings on-platform

Everything focuses at the relationship level — not platform-wide

MUST HAVE
  • Client ID + current tier
  • GBV progress bar
  • Transaction history
  • Rebook / Message CTA
SHOULD / WON'T (MVP)
  • Upcoming bookings
  • Refund visibility
  • Conversations
  • Celebratory milestones
DESIGN PRINCIPLE

You don't need to understand the model to use it — you just need to see your progress with this client.

Relationship page MVP design

Relationship page MVP · Figma · Feb 2026

12 / TOUCHPOINTS

Every surface
tells the story.

The fee model only works if sitters encounter it at the right moment — not just once in a landing page. Five distinct surface types, each with a specific job.

Landing page
Carousel + FAQs. Source of truth for the full model. Auth-only. Localised en-ca / fr-ca.
Dashboard announcement
First-login callout on Daily Dashboard and web sitter dashboard. Two variants: with / without existing clients.
Conversation system message
Contextual callout in inbox. Content varies: new relationship, in-progress, tier-unlocking, maxed out.
Booking earnings breakdown
Bottom sheet in booking details showing fee breakdown by tier. Links directly to relationship page.
Relationship page
The hub. Per-client GBV progress, transaction history, tier status, and rebook CTA. The model's source of truth.
Touchpoints system overview

Touchpoints overview · Figma · deep-dive doc

13 / DESIGN CHALLENGES

Four hard problems.

Click any card to see what we considered and what we decided.

01 Model comprehension

Cumulative GBV, milestones, permanent reductions — that's four mental models at once.

Options: (A) Show all the maths up front. (B) Progressive disclosure — anchor on 'this relationship with [client name]', reveal mechanics only when needed. (C) Full onboarding flow before first booking.

→ Progressive disclosure. The relationship page is the source of truth; sitters don't need to understand the model to use it — they just need to see their progress. Avoided overwhelming with rules at announcement.

02 Perceived fairness across services

GBV thresholds favour boarding ($300/stay) over walking ($20/walk). Is that fair?

Options: (A) Accept asymmetry in v1, anchor messaging on relationship not maths. (B) Service-specific thresholds. (C) Points-based system where different services earn at different rates. (D) Count-based milestones (5th booking) — equal across services but gameable.

→ Accepted asymmetry in v1. Boarding drives 53% of top sitter earnings; it's the right place to test. Messaging focused on the relationship benefit, not the threshold value. Points system backlogged as a future iteration.

03 Sense of agency

Progress toward a milestone depends partly on whether the owner chooses to rebook — not just sitter quality.

Options: (A) Surface progress as purely dependent on the relationship (honest but demotivating). (B) Focus UI exclusively on actions the sitter controls — booking on-platform, sending post-stay messages, rebooking nudge. (C) Explicitly tell sitters that owner repeat rate is a factor.

→ Focus on sitter-controlled actions. Never show progress as dependent on owner behaviour. UI surfaces: "Book on Rover", "Remind client to rebook", "Send post-stay message" — avoid showing "waiting for client" as a progress state.

04 Threshold-crossing bookings

A single booking crosses a tier boundary mid-transaction. How is the fee applied?

Options: (A) Split fee — first portion charged at old rate, remainder at new rate. Accurate and honest, but creates a blended fee on one ledger line. (B) Cashback — full booking at the old rate; credit issued for the next booking. Simpler ledger display but delayed reward, requires a credit wallet.

→ Split fee. Most transparent — sitters get the benefit the instant they earn it. The ledger complexity is manageable with a clear breakdown modal (US5). Cashback backlogged; at relationship level threshold crossings are recurring, making it a meaningful future incentive loop.

14 / EDGE CASE

Training &
Grooming.

Average training relationship GBV is ~$500 — meaning virtually all trainers stay at 30% forever. Off-platform, trainers earn $300–350/client vs. $100–130 on Rover. A 30% Tier 1 makes that gap worse.

Five paths were explored. Click each to read the trade-off.

CHOSEN PATH

Training and Grooming excluded from the graduated take. Flat 20% applies; they don't count toward relationship GBV progress. No fee callouts in Training conversations. Mixed relationships (core + training) get two separate sections.

Option A ✓
Exclude Training & Grooming
Complexity: Low
Flat 20% for Training and Grooming; these services don't count toward GBV progress. Already live in Canada. Clean, avoids permanent Tier 1 problem.
Option B
Include as-is
Complexity: Very low
Same tiers, no exceptions. Trainers permanently stuck at 30%. Creates resentment; makes fee gap vs. off-platform worse.
Option C
Service-specific thresholds
Complexity: Medium
Different GBV thresholds per service type — lower milestones for Training. Fair, but adds significant config complexity and comms burden.
Option D
Service-specific fee %
Complexity: Medium
Same thresholds but lower Tier 1 fee for Training. Reduces the gap but doesn't solve the stuck-in-Tier-1 problem long-term.
Option E
Points-based system
Complexity: Very high
GBV replaced by points accruing at different rates by service type. Most equitable across services — but a completely different model. Future iteration territory.
Mixed relationship (core + training) design exploration

Mixed relationship (core + training) · design exploration · Apr 2026

15 / RESEARCH

Concept test.
4 segments.
28 sitters.

Qualitative 1:1 interviews, 60 min each, remote via Google Meet. Canada market. Oct–Nov 2025 — before a single line of code was written.

5 LEARNING DIMENSIONS
Cognitive load  ·  Fairness  ·  Alignment with Rover's decreasing value  ·  Perceived benefit  ·  Agency & control

Researcher: Lizeth Herrera
Stakeholders: Mégane Martinez, Bernardo Prudêncio, Guillem Pons (CPO)

New sitters
< 1 year tenure · trial & exploration phase
MotivationPractical & flexible — supplemental income, lifestyle fit
React to 47% start feeFinancial survival risk. Feels unsafe to start a business at such high initial cost. Many would not join at all.
Fairness perceptionStructurally unfair — lack leverage, pricing power, or volume to reach milestones quickly
Verdict for live testMotivation collapses. 47% start acts as a deterrent to taking new clients.
Growing sitters
~1 year tenure · stabilising activity
MotivationOptimising for life — a job that fits their life, not business growth
React to 47% start feePenalty perception. Enthusiasm drops; the reward feels too distant. 50/50 split feels "too much".
Fairness perceptionService-dependent — fair for boarders, a grind for walkers who need many bookings to hit $500
Verdict for live testMotivation decreases. Shifts focus to keeping existing clients rather than acquiring new ones.
Established — loyal
Long tenure · high repeat usage · risk-averse
MotivationLove & routine — genuine affection for animals; frustrated by market saturation
React to 47% start feePunitive. Mechanics clear, but emotional reaction strongly adverse. Concerns about one-off clients who'll never return.
Fairness perceptionConditional fairness — fair if you break through the milestone; unfair if progress depends on owner repeat behaviour
Verdict for live testIncreases motivation post-milestone. Highly motivated once fee drops; sees 10% as a genuine partnership.
Established — diverters
Long tenure · suspected off-platform bookings
MotivationStrategic boundaries — strict selectivity, boundary-setting, value-based pricing
React to 47% start feeConfirmation of bias. Confirms suspicion the model is designed to protect Rover's margins, not partner with sitters.
Fairness perceptionFavours "power users" — high-volume sitters. Seasonal or selective sitters feel excluded.
Verdict for live testMotivation decreases or stagnates. Pre-milestone phase is seen as a "bad deal". Stays only to avoid abandoning existing clients.
16 / FINDINGS

What we heard.
What we changed.

RESONATED

The principle of "higher fees at the start, lower as the relationship deepens" was understood and felt fair across all segments in theory. Sitters intuitively grasp that Rover provides most value upfront.

DIDN'T LAND

The mechanics — cumulative GBV, milestone triggers, permanence, scope per client — created significant friction independent of the starting fee. When 47% was shown, fairness perceptions collapsed.

WHAT IT CHANGED

Starting fee tuned to 30% (not 47%). Comms strategy shifted to lead with the "why" before the maths. Backfilling historical GBV at launch was a direct response — sitters enter already winning.

"I've cared for this family for 3 years. I know their dog, I know their home, I know their schedule. And I still pay the same 20% as on day one."

Established sitter · concept test · Canada · Oct 2025

Customer feedback and social listening signals

Social listening & customer feedback · design exploration · Oct 2025

17 / SMOKE TEST RESEARCH

100 conversations.
Zero mentions
of fees.

After the smoke test launched, I manually reviewed 100 real sitter–owner conversations from the treatment geo. Dia reviewed each thread against key themes including any organic mentions of tiered incentives or rewards.

KEY SIGNAL

No explicit mentions of tiered or graduated fees, rewards programs, or loyalty schemes in any conversation. All pricing treated as a flat per-service outcome of the standard Rover flow. This confirmed the model needs to be actively surfaced — sitters won't discover it organically.

3 PATTERNS THAT MATTERED
Relationship CRM behaviour

Same sitter and owner, lots of repeat coordination, high trust. "So happy you're available again", "the dogs will be very happy to see you". These are the relationships the fee model is designed to protect — but sitters never reference a program.

Off-platform gravity — subtle, not explicit

Nobody wrote "let's go off Rover." But payment method drift (e-transfer mentions), adjustments handled via SMS, and granular changes made outside of bookings all signal that Rover's perceived marginal value drops as relationships mature.

Modification opacity creates friction

"I was charged for 2 visits but I counted 3" — no visible before/after diff. Sitters explain platform rules from memory. This reinforced the decision to show clear tier-crossing breakdowns in booking details (US5).

Author: Bernardo Prudêncio · Mar 12, 2026 · 100 conversations · smoke test geo

19 / 21 · Measurement

Measuring Impact

Primary metric
Per-relationship GBV growth
GBV earned per active sitter–owner pair in the 90-day window post-launch, treatment vs. holdout control. Clean signal of whether lower Tier 2/3 fees are actually pulling repeat bookings back on-platform.
Break-even logic
≤7%
Diversion can increase by up to ~7% on new relationships before the take-rate gain from repeat relationships is neutralised. Our hypothesis: loyalty effect more than compensates for the higher starting fee.
Secondary signals
SignalTargetWhy it matters
Sitter comprehension≥70% correct on key mechanicsModel only works if sitters understand the incentive
Perceived fairness (top sitters)Net positive sentimentCore retention risk if fairness scores collapse
Diversion — new relationshipsNo material increaseValidates that higher Tier 1 fee isn't killing acquisition
Repeat booking rate (90-day)+3–5pp vs controlBehavioural proof that Tier 2/3 is changing sitter actions
Real-time · DataDog
Fee calculation errors, p95 API latency, client-side crashes. First-week triage was run against DataDog dashboards before any analytics pipeline was ready.
Weekly cohorts · Mode
Cohort-level GBV analysis segmented by sitter tier, service type, and relationship age. Analytics built before Canada launch; first read at 4 weeks.
20 / 21 · Reflection

What I'd Do Differently

Four specific things — not regrets, but what I'd change if I ran this again.

01 · Test the fee numbers earlier

47% of concept test participants showed motivation collapse at a 47% starting fee. We knew this going into the Canada launch but treated it as a business constraint, not a design problem. I'd push for a joint design–finance session to stress-test rate options against comprehension data before they become locked inputs.

02 · Answer "do past earnings count?" in the product

The single biggest confusion at launch. We knew it was a question — it was in our FAQ backlog. I'd build a "you're starting at Tier X because of your history" state into the Relationship page from day one, not retroactively via a comms FAQ on April 21.

03 · Scope the training/grooming edge case sooner

The mixed-service relationship (e.g. boarding + training with different sitter licences) surfaced late in the project and needed five option cards and a difficult trade-off conversation. I'd pull edge-case discovery into the first sprint, not the last one before handoff.

04 · Bring content design in at discovery

System messages, announcement copy, landing page, and FAQ all needed more iteration than I'd scoped. Content was brought in at execution. For a model this complex — where comprehension is a primary success metric — content strategy needed to be a first-class input from the first prototype.

21 / 21

Bernardo Prudêncio

Senior Product Designer · Monetisation & Provider Experience

LinkedIn bernardo.prd@gmail.com View case study →

Relationship-Based Graduated Fees · Rover · Canada live · March 2026