How a financial adviser moved her clients from a shared spreadsheet to a WhatsApp budget app.

• Each family got a copy of her spreadsheet template in Google Drive, so she could see what was going on and help.
• She loaded their statements with them by hand. She went through all their bank statements with each family on calls of about two hours, from her own time.
• Families recorded every expense in the spreadsheet, which meant opening the template and finding the right row every time they spent money.
• The adviser checked the sheets herself to see who was over budget and where.
• She did the chasing. Prompts to plan the next week or review the month came from her.

Edo calls the spreadsheet "an overhead and a huge friction" for both sides.

The adviser wanted logging a daily expense to be easy, so that families would become more aware of what they spent. Edo offered to build an app that put WhatsApp first, since that was where her clients already were every day.

• Each family has the app's WhatsApp number alongside a desktop app, which handles the bigger jobs like uploading statements.
• Statements are uploaded in the desktop app, where they are analysed with a mix of manual work and some AI and turned into the family's budget. The planned onboarding agent in workflow 2 will take this over.
• A family logs one or several expenses in one message. A short typed message, like "shopping 500", goes straight in with no AI involved, and anything else goes to an AI model that works out what they meant, so families don't have to learn a format.
• The app replies after every expense with how much is left in that category, or a warning that the category is over.
• Scheduled messages do the chasing. At the end of each week the app asks the family to plan the next one, and it does the same at the end of the month or when a category passes a set amount. One tap opens the plan to edit.

The AI layer came after the first version. With it, families can write freely, record a voice message or send a photo of a receipt, and the AI recognises the shop and the products and records the expense.

The AI also tells an expense from an income or from a question about their finances. It offers a category when it meets a merchant it hasn't seen before.

Edo kept the AI's job as small as possible. "This is financial data, so you don't want AI handling everything," he says, so the code handles every message it can understand and the AI only sees the rest.

Build time. Edo started designing the app in August 2025 and building it in late October, and the first version launched in late January 2026. A product manager and a second developer were on the team for part of that time.

Difficulty, hard. It took an experienced developer three months to get WhatsApp, a database, scheduled jobs and AI working together. A no-code version of the core idea, a WhatsApp message that adds a row to a sheet, is a medium build.

Prerequisites. You need a WhatsApp Business account connected through Meta's Cloud API, a database that can run scheduled jobs, and an AI model for the messages code can't read. You also need to be comfortable directing an AI coding agent and reading what it writes.

• Families log an expense, or several at once, in one WhatsApp message, where they used to open a spreadsheet and find the right row.
• Families are much happier reporting expenses by WhatsApp than in the old spreadsheet.
• The adviser saves about two hours a week on reviewing families' financial data before meetings. This is an estimate, not a tracked figure.
• Advisers see a family's numbers more clearly than in the spreadsheet, and can ask the app questions about them before they talk to the family.
• Families pay to use it. The paid tier has been live since early September 2026.

• A new family's statements needed reading line by line, first by the adviser on her two-hour calls, then in the app as manual work mixed with some AI to read the files.
• Every transaction needed a category, which meant sorting merchants one by one.
• Commitments and assets had to be picked out by hand, including loans, the mortgage and card purchases paid in instalments.
• The budget was built from all of that, so a family waited for the whole job before seeing the full picture.
• Onboarding takes about three hours today, and families can't complete it on their own.

The app offers new families 30 days free, and not every family stays after that. Edo estimates that a regular AI agent reading all of a family's files and chatting with them "can cost several dollars" in AI usage per family, which adds up fast for people who may never pay.

The onboarding agent is planned functionality and isn't live yet. This is how Edo has designed it to work.

Click to expand

Step 1 of 5, commitments. The agent has read a mortgage summary, a loan agreement and a card statement, and proposes each commitment with its source file and a confidence level for the family to approve or edit. A demo screen with a made-up family and made-up figures, supplied by Edo.Close

• The family uploads everything at once in the desktop app, whether that's bank statements, card statements or loan documents.
• AI reads the files and pulls out the transactions.
• Code builds a ledger of every transaction and works through it, sorting commitments, assets, instalments and merchants into categories.
• It works through three months of a family's money. That covers expenses, commitments, assets and income, in about 25 back-end steps of file and data analysis that mix code and AI.
• It looks up each merchant on the bank statements and saves it, so the app recognises it next time.
• Code asks most of the questions. Depending on the family's data, the agent has up to 45 topics to resolve, with up to 60 possible questions, and code rather than AI poses most of them.
• AI steps in only where code can't decide, like a merchant it doesn't recognise or an answer in the family's own words.
• It offers a first budget built from the family's expenses and income and the adviser's own budgeting rules.
• The family approves every item before it goes into the budget, step by step through commitments, assets, expenses, income and the budget itself.

"Whatever I can do in code, I did in code. Whatever is not in code, the AI does," Edo says.

Click to expand

Step 2 of 5, assets. The family gives the car's make, model and mileage in the chat, and the agent suggests a value range and records the figure they agree. A demo screen with a made-up family and made-up figures, supplied by Edo.Close

Click to expand

The annual overview. The family names the bigger yearly costs, like summer camp and a holiday, and the agent spreads them into a monthly amount to set aside. A demo screen with a made-up family and made-up figures, supplied by Edo.Close

The same agent is designed to stay useful after onboarding. At the end of a month a family will be able to upload their statements again, and the agent will reconcile them against what they reported by WhatsApp and show what was missed or has changed.

Build time. The agent has taken two months so far. Edo built and tore it down several times, and improved the architecture with every build.

Difficulty, hard. Edo calls the onboarding agent "a very complex architecture", mostly code with AI for anything ambiguous and for the conversation itself. Deciding which steps code can do and which need AI was, in his words, "a lot of architectural work" that he had to think through rather than leave to the AI.

Prerequisites. You need financial data already structured in a database, a clear set of categories, and a list of the questions an adviser would ask a new client.

• Onboarding should take about 30 minutes once the agent is ready, down from about three hours today.
• Families will be able to do it themselves, which they can't today.
• AI is used only where it is needed, which keeps the cost of each free trial down.
• The code answers immediately, so a family won't wait for an AI model to call its tools before seeing the next question.
• The sums come from code, so the budget adds up every time.
• The standalone version waits on it. The adviser plans to open the app to people without an adviser once the agent is finished.