A bookkeeping app that doesn't feel like an ERP.

The app helps small business owners (UMKM) in Indonesia track sales, expenses and cash flow. The product worked, but new users dropped off during onboarding and rarely came back.

We ran a heuristic evaluation (Nielsen's 10) and a task-flow review of the live app, using an AI-assisted audit to cover every screen quickly. From there, the team went straight into high-fidelity design, and the public release became the real test.

The “ERP look” was only the surface. Three problems reinforced each other:

• Scattered information architecture — the same action lived in several places under different names, and core reports were buried in Profile.
• Accountant language for shop owners — Balance Sheet, Debit/Credit and VAT greeted small-shop owners with no explanation.
• Passive empty states — new users saw “Rp 0” everywhere, with no hint of what the app would give them.

Our users were owners aged 30 and up who rarely open email and don't want another password. So sign-in became a WhatsApp number and a one-time code, with Google as the alternative, and new users only type their name and business name to get in.

The data backed it up: in the old flow, almost 9 out of 10 people who reached the one-time code step left the app right there.

We ranked 15 recommendations by impact on early retention and by effort. P0 went after the root causes first, because they were cheap to change and touched every screen.

1. P0 · 1–2 sprints →Plain-language vocabulary, one locale, one primary action per screen, empty states that explain.Language, CTA, empty states.
2. P1 · 2–4 sprints →Quick-add transactions, undo, a calmer Home and a guided first session.Quick-add, undo, Home.
3. P2 · Long termReceipt scanning, repeat transactions and a simple vs. accountant mode.OCR, templates, modes.

• Language first — accounting jargon replaced with words a warung owner uses every day, with a “?” hint for unavoidable terms like PPN.
• One focus per screen — Home now leads with a two-step setup mission (complete your business profile, add your first product) and one main action: record a sale.
• A guided first sale — New owners are walked through their first sale step by step, or they can just say it: “Add 2 mineral waters and 1 blender.” AI fills the form, the owner confirms.
• Onboarding you can skip — Sign up with WhatsApp or Google, fill in three fields, or skip straight to Home. Nothing is locked.
• Empty states that invite — every empty screen says what will appear and offers one clear next step.
• Start with a photo — Adding a product starts with a picture, because seeing it helps owners remember the name and price. A barcode scanner removes typos, and AI can fill in the details from the photo.

Three weeks in, the client told us the redesign still looked like a reskin: new fonts and colours on the same technical, B2B flows. They were right.

We went back to first principles: ask owners their business type up front and personalize from there, cut the add-sale, add-product and add-expense flows down to the essentials, let AI do the typing through voice and photo, and give owners a reason to come back.

Every real sale earns XP and keeps a daily streak alive. Tiers from Silver to Diamond unlock tools owners actually want: weekly reports, restock alerts, more staff seats. XP never resets; a missed day only costs the streak, and a streak freeze softens it.

We shipped in two steps. A quick fix first, cutting two welcome screens, then the redesigned onboarding.

Daily active businesses followed: from single digits a day in June to 20+ a day by late July.

The full redesign was then documented as one PRD and handed to the client's engineering team, ahead of the app's launch in new markets.

Interview 5–10 shop owners to validate the new vocabulary, and test a quick-add flow for owners who log 30+ transactions a day.