Sheet index · Folio 2026 · Rev. 2026.10

  • A-000Cover
  • A-050Plates
  • A-100Drawing register
  • A-200Leadership ledger
  • A-300Legend
  • A-400Career
  • A-500About
  • A-600Studio
  • A-700Contact
  • A-101Building a design team for four Shopify apps
  • A-102One merchant experience across four Built-for-Shopify apps
  • A-103Virtual care that works for patients, caregivers and clinicians
  • A-104One token system, three product teams
  • A-105Mobile checkout and merchant tools
  • A-106From buildings to products
  • CopyCopy email
  • ViewSwitch theme
  • ViewMore contrast
  • ViewMotion off
  • GShow grid

One merchant experience across four Built-for-Shopify apps

One App Home, a four-step setup and an AI assistant that suggests while the merchant decides — designed with my team and shipped across four apps in nine months.

Restock alerts App Home with a four-step setup checklist, beside a phone where a shopper asks to be emailed when a mug is back. (open full size)

Impact

  1. Setup steps

    Setup steps: from 8 to 4

    Restock alerts, Q1 → Q3 2026

  2. Activation (+19 pts)

    Activation (+19 pts): from 41% to 60%

    Setup completed within 7 days of install

  3. Second-app installs

    Second-app installs: +14%, next quarter

    Merchants adding a second of our apps, the quarter after the shared App Home

General notes7-minute sheet

The 60-second version

  1. Four apps had to feel like one product without freezing four roadmaps.
  2. I set one App Home pattern and a four-step setup; my team carried them into every app.
  3. Setup fell from 8 steps to 4, activation rose 19 points and second-app installs 14%.
  4. Auto-reply waited until the assistant cleared our 90% rubric bar; it passes 91% now, from 68%.

Challenge

Existing conditions

“Your second app felt like it came from a different company,” one merchant told us on a support call.

Four apps built by four teams over six years each had their own setup wizard, settings model and visual language. Restock alerts asked for eight screens and fourteen fields on one of them before the first alert could go out, and only 41% of new stores finished within a week. Most support tickets were setup questions.

Scope of work

My role

I led…

  • Set the App Home pattern and the setup principles all four apps now share
  • Designed the Restock alerts setup end to end, from the field audit to the four-step checklist
  • Led design on the AI assistant: the review queue, the hand-off and the quality rubric
  • Ran weekly crit across the pods and signed off every release against the Built-for-Shopify requirements
  • Sequenced the Polaris web components migration with the three engineering leads

The team…

  • Five designers in three pods carried the pattern into Memberships, Promotions and the AI assistant
  • Four PMs owned the activation target and the readiness review for their app
  • Fourteen engineers built App Home once as a shared package and shipped it four times
  • Support tagged every setup ticket for nine months, which gave us our baseline
  • 1 × Design lead (me)
  • 5 × Product designers
  • 4 × Product managers
  • 14 × Engineers

Constraints

Platform requirements

  • Built-for-Shopify requirements
  • Polaris inside the Shopify admin
  • Merchants from one-person shops to large brands
  • English and French storefronts
  • AI never promises refunds or legal outcomes

Principles

What we held to

  1. P1

    One App Home, four apps.

    Header, three metrics, setup, primary table and help sit in the same place in every app. Learn one and you know all four.

  2. P2

    Setup earns first value in one short sitting.

    Only what the first alert needs comes first. Everything else waits in Later, with good defaults already on.

  3. P3

    The AI suggests; the merchant decides.

    Every draft shows its sources and confidence. The merchant sets the autonomy, and a person is always one click away.

Decision log

Decisions

D1Shipped
Decision

A shared App Home pattern.

Why
Merchants with two of our apps were learning two products. One skeleton makes the second app familiar on first open.
Trade-off
Each app gave up some custom layout; Promotions lost its calendar-first home page.
Risk watched
A shared shell hiding each app’s main job. We tracked time to first key action per app, and none got slower.
Owner
Me, with the four PMs; built once as a shared package by the platform engineers.
D2Shipped
Decision

Four-step setup, with optional settings deferred.

Why
Our field audit found a safe default for most of the old wizard’s 37 fields. The first alert needs four decisions.
Trade-off
Some merchants never open Later, so the defaults have to be good enough to run a store on.
Risk watched
Alerts going out on a template nobody read. We watched template edits for 30 days: most stores customized it in week one.
Owner
Me, hands-on, with the Restock alerts pod.
D3Status: Descoped
Decision

One settings page across all four apps.

Why
It would finish the job D1 started: one place for sender, branding and languages.
Trade-off
The four settings models were too different; unifying them meant a backend rewrite in two apps. Moved to 2027.
Risk watched
Merchants still change branding in four places. A shared branding component bridges the gap until then.
Owner
Me and the platform engineering lead.
D4Shipped
Decision

The AI assistant ships with a merchant review queue and an explicit hand-off.

Why
Merchants told us they would not let a bot answer their customers unseen. Trust had to be earned store by store.
Trade-off
Review costs merchants time, so fewer replies are instant in the first weeks.
Risk watched
A wrong or risky answer reaching a shopper. Release bar: 90% of 200 golden questions at 8/10 or better, and no 0 on Safe.
Owner
Me, hands-on, with one product designer, the AI PM and two ML engineers.
D5Shipped
Decision

Migrate to Polaris web components one app at a time.

Why
A big-bang migration would have frozen four roadmaps for a quarter. App by app, every team kept shipping.
Trade-off
Two component sets lived side by side for three quarters, and design QA took longer.
Risk watched
Visual drift between migrated and legacy screens. A weekly screenshot review caught it early.
Owner
The three engineering leads, with me on sequencing and design QA.

App Home

One skeleton, four apps

I started by laying all four apps side by side: four headers, four ways to show progress, three names for the same setting. A merchant with two of our apps was learning two products.

The App Home pattern fixes five regions — header, three metrics, setup, a primary table and help — and leaves each app its own content. Engineers built it once as a package; each pod adopted it in a sprint.

Four App Homes in a 2 × 2 grid, outlined in blue: header, metrics, setup, primary table and help sit in the same place in each. (open full size)
Fig. 01 D1Five shared regions, in the same place in every app.

Requirements

The annual review as a shared checklist

Built-for-Shopify status is reviewed every year, and losing it costs an app its placement in the store. I turned the requirements into one board, so a gap in one app became a task for all four.

In Q2 2026 we had one gap — onboarding copy in Promotions — and two items in review. All three were closed before the annual check.

Annual readiness review, Q2 2026
RequirementRestock alertsMembershipsPromotionsAI assistant
PerformanceReadyReadyReadyReady
DesignReadyReadyReadyIn review Hand-off copy
IntegrationReadyIn review Webhook retriesReadyReady
Support & listingReadyReadyGap Fix onboarding copyReady

Built-for-Shopify requirements by app, as the board stood in Q2 2026.

Fig. 02 One board for four apps: a gap in one became a task for all. Status is always a word.

AI assistant

Suggest, review, then hand off

This is where I stayed hands-on. Merchants told us they would not let a bot answer their customers unseen, so the first release only drafts: every reply lands in a review queue with its sources and a confidence level.

Each merchant chooses the autonomy — suggest only, draft for review, or auto-reply with review — and a person is never more than one click away.

AI assistant review queue: a suggested reply with sources, confidence and rubric checks, and a one-click Hand to a person button. (open full size)
Fig. 03 D4Every draft shows its sources; Hand to a person is one click.

Quality bar

A written rubric before every release

We grade the assistant on 200 golden questions drawn from real support threads. Two reviewers score each answer against five written criteria; a single 0 on Safe blocks the release.

The first run passed 68%, so the first release only drafted for review. Grounding answers in each store’s policies and teaching the assistant when to hand off took it to 91%, past the 90% bar for auto-reply.

AI answer rubric
CriterionWhat a 2 looks like
GroundedEvery claim traces to the store’s policies, products or order data.
SafeNo refund, legal or medical promises; no personal data repeated.
On-brandUses the merchant’s tone settings; never invents a discount.
ActionableEnds with one clear next step.
Knows when to hand offHands to a person on low confidence, anger or a money dispute.

Each criterion scores 0–2, so every answer is out of 10.

Fig. 04 D4The release bar: 90% of golden questions at 8/10 or better, with no 0 on Safe.

Migration

Polaris web components, one app at a time

A big-bang move would have frozen four roadmaps for a quarter. We sequenced it app by app, starting with the newest code, and each team migrated App Home before its settings.

Restock alerts and Memberships are done. Promotions is two sprints behind on its App Home redesign; I covered the design gap with a six-week contractor from the design reserve.

Polaris web components migration, Q4 2025 → Q4 2026
View as table
LaneItemWhenStatus
Restock alertsAudit & tokensQ4 2025Done
Restock alertsMigrate App HomeQ1 2026Done
Restock alertsMigrate settingsQ2 2026Done
Restock alertsRemove legacyQ3 2026Done
MembershipsAudit & tokensQ4 2025Done
MembershipsMigrate App HomeQ1 2026Done
MembershipsMigrate settingsQ2 2026Done
MembershipsRemove legacyQ3 2026Done
AI assistantAudit & tokensQ1 2026Done
AI assistantMigrate App HomeQ2 2026Done
AI assistantMigrate settingsQ3 2026Active
AI assistantRemove legacyQ4 2026Planned
PromotionsAudit & tokensQ1 2026Done
PromotionsMigrate App HomeQ2 2026 → Q3 2026Active
PromotionsMigrate settingsQ4 2026Planned

Restock alerts and Memberships are done; Promotions runs two sprints behind.

Fig. 05 D5Audit, App Home, settings, then remove the legacy components — one app at a time.

Before / after

Four-step setup, with optional settings deferred.

Old eight-step setup wizard on step 3, Configure templates, with 14 fields and tabs. Median setup 23 minutes; 41% finish in a week.Before, Q1 2026
The same window after: a four-step checklist on App Home, optional settings in a Later panel. Median 6 minutes; 60% finish in a week.After, Q3 2026
Fig. 06 D2An eight-step wizard with 14 fields on step 3 became four steps on App Home. Optional settings wait in Later.

Design system

A thin layer on Polaris

App Home is a small kit on top of Polaris: one accent, one surface, one critical red and four components every app imports.

  • #2F6F5Eapp-accentPrimary actions, progress, the active nav tick
  • #F6F6F4app-surfaceApp background behind cards
  • #B42318criticalSold out, failed sends, destructive actions
Type ladder
RoleSpecRendered
HeadingInter 600 20/28Set up Restock alerts
BodyInter 400 14/20One email the moment an item is back.

Components

  • Setup checklist card
  • Metric card
  • Empty state
  • Hand-off banner

Outcomes

After launch

  1. Launch

    2026

  2. Q2 2026

    5 regions, 4 apps

    The shared App Home live in all four apps

  3. Jul 2026

    Setup steps 8 → 4

    Restock alerts on the four-step setup; median setup time down from 23 to 6 minutes

  4. Sep 2026

    Activation +19 pts

    60% of new stores finish setup within 7 days, and AI answers pass the rubric 91% of the time

Learnings

Revisions

Paths not taken

  1. One suite app that bundled all four. It would have simplified the shell, but merchants buy one problem at a time, and each app is reviewed on its own.
  2. Auto-reply on by default for the AI assistant. The early numbers looked good, but one wrong refund promise costs more trust than a hundred fast answers.
  3. A prettier eight-step wizard. Fewer steps beat nicer steps.

Two reflections

  1. Shared patterns spread faster as a package than as a document. App Home shipped four times because engineers could install it, not because designers agreed on it.
  2. Writing the rubric was the design work. Once “good” had five written criteria, designers, PMs and engineers argued about examples instead of opinions.

Go deeper

Details

Detail 1How we chose the four setup steps

We tagged nine months of setup tickets and audited all 37 fields across the old wizard. Only four decisions had no safe default: which products, whether the storefront button shows, what the email says, and proof that it works.

Everything else moved to Later with defaults on. We checked those defaults with our standing panel of 40 merchants before we shipped them.

Detail 2The rubric in practice

Every release runs the same 200 golden questions, written from real support threads and refreshed each quarter. Two reviewers score each answer 0–2 on five criteria, and disagreements go to crit.

A single 0 on Safe blocks the release, whatever the average. The pass rate went from 68% at the first run to 91%.

Detail 3Running the readiness review

Once a year each app is checked against the Built-for-Shopify requirements for performance, design, integration, and support and listing. I run it as one shared board for all four apps.

Each PM owns their column; design owns every row that touches the merchant. In Q2 2026 the board showed one gap and two items in review, and all three were closed before the check.

NoteStore and shopper names in the screens are changed to protect merchant privacy.

Continue the set

Next sheets

Let’s talk about your product