Talk to My Business: Why the Next Interface for Finance Is a Conversation

Metabase, Tableau, Looker. The dashboard market crossed $15 billion last year. Yet the most common financial question a founder asks on a Friday night — "Can I afford to hire next month?" — can't be answered by any of them.

The data is all there. Revenue trends, burn rate, runway projections, outstanding invoices. Spread across four tabs, three filters, and a spreadsheet exported twenty minutes ago because the dashboard couldn't do the arithmetic.

I built dashboards like this for seven years at Qonto. Hundreds of them. For 200,000+ businesses. Here's what I never admitted out loud: dashboards are where questions go to die.

The Read-Only Trap

Every fintech pitch deck in 2024 promised "visibility." Connect your bank account, see your cash flow, get a beautiful chart. The industry spent a decade building increasingly sophisticated ways to look at your money.

But looking isn't understanding. And understanding isn't acting.

A founder doesn't open their banking app thinking "show me a bar chart of Q3 expenses by category." They think: "Can I make payroll if that invoice doesn't land by Friday?" or "What happens to my runway if I sign this new vendor contract?"

These are conversational questions. They require context, memory, and judgment. No dashboard answers them. We built the entire SMB finance stack as a read-only interface. That was a mistake.

Why Chatbots Aren't the Answer Either

Here's where I'm supposed to tell you that AI fixes everything. Slap an LLM on top of your banking data, let founders "chat with their finances," problem solved.

I wish.

I've tested every AI finance tool that launched in the last eighteen months. Most of them do the same thing: take your question, convert it to a database query, return a number. "What was my revenue last month?" → €47,200. Done.

That's a search engine with a chat skin. It's not intelligence.

The question a founder actually asks is: "Revenue was €47,200 last month — is that good? Why did it drop from €52,000? Is the trend seasonal or did we lose a client? And what does that mean for the hire I'm planning?"

Each follow-up requires something an LLM wrapper doesn't have: context about your specific business. Not generic financial knowledge. Your contracts. Your burn pattern. Your hiring plan. The invoice you've been chasing for three weeks.

Most AI finance products fail here because they treat every conversation as a fresh start. No memory. No accumulation of understanding. Every Monday morning, the AI has amnesia.

The Three Layers Nobody Talks About

After building two core banking integration systems — one that scaled to handle transactions for hundreds of thousands of businesses — I noticed a pattern in what actually works.

Financial intelligence isn't one thing. It's three layers, and almost everyone is building only the first.

Layer 1: Record. Connect accounts, ingest transactions, normalize data. This is plumbing. It's unglamorous, brutally hard to get right, and absolutely necessary. Most fintechs stop here and call it a product.

Layer 2: Intelligence. This is where raw data becomes business understanding. Not "you spent €4,200 on software" but "your software costs increased 34% quarter-over-quarter, driven by scaling your engineering toolchain after your last hire." Intelligence requires a data model rich enough to capture relationships between transactions, vendors, contracts, and business events.

Layer 3: Action. This is the layer that barely exists today. "Your invoice to Client X is 15 days overdue. Based on their payment history, a follow-up email on Tuesdays gets 40% faster responses. Want me to draft one?" That's not a dashboard. That's not a chatbot. That's a financial copilot.

Most products live entirely in Layer 1. A few are attempting Layer 2. Almost nobody has cracked Layer 3, because action requires trust — and trust requires the AI to prove it understands your business deeply before you let it act on your behalf.

The Missing Piece: Persistent Business Memory

Here's the part that took me too long to figure out.

The reason most AI finance tools feel shallow isn't the model. GPT-4, Claude, Gemini — they're all smart enough. The bottleneck is memory architecture.

When a founder tells their AI "we're planning to hire in Q2" — that fact needs to persist. It needs to connect to burn rate projections, runway calculations, and cash flow forecasts. When that same founder asks "can we afford the hire?" three weeks later, the system shouldn't need the context re-explained.

This is what I call the Business Context Graph: a persistent, structured representation of everything the AI knows about a specific business. Not a chat history. Not a vector database of past conversations. A living graph that connects financial data to business decisions, plans, and context.

I'll be honest: we haven't fully solved this. Nobody has. The Business Context Graph today captures roughly 70% of the context needed for reliable financial intelligence. The remaining 30% — implicit knowledge, founder intuition, market context — is genuinely difficult to formalize.

But 70% turns out to be transformative. Because 70% business context plus a strong language model beats 0% business context plus the same model every single time.

What Changes When Finance Becomes Conversational

The shift from "look at your data" to "talk to your business" isn't incremental. It changes the operating model for how small businesses interact with their finances.

Frequency increases. Nobody opens a dashboard daily. But people chat with tools constantly. When the interface is a conversation, founders engage with their finances 5–10x more often. More engagement means fewer surprises.

The skill gap closes. Reading a P&L statement is a learned skill. Asking "am I profitable this month?" is not. Conversational interfaces democratize financial understanding in a way dashboards never could.

Decisions get faster. The gap between question and action collapses. Instead of: notice something → export → analyze → decide → act, it becomes: ask → understand → act. Three steps instead of five. Minutes instead of hours.

The Uncomfortable Truth About Building This

I should be clear about what makes this hard — not from a pitch perspective, but from an engineering one.

Conversational finance requires getting three things right simultaneously: data infrastructure, intelligence modeling, and conversational UX. Most teams are good at one, maybe two. Almost none are good at all three.

The AI community is obsessed with model capabilities. But in finance, the model is maybe 20% of the problem. The other 80% is data quality, entity resolution, enrichment accuracy, and building a context layer that doesn't hallucinate about your bank balance.

When an AI gets a general knowledge question wrong, it's embarrassing. When it gets your cash position wrong, you miss payroll. The error tolerance in financial AI is effectively zero, and that constraint shapes everything about how you build it.

Where This Goes

The future of SMB finance isn't a better dashboard. It isn't a better spreadsheet. It isn't even a better bank account.

It's a conversation — one where the AI knows your business deeply enough to answer the questions that keep you up at night, and eventually, to act on the answers.

We're building this at Well because I spent seven years building the infrastructure layer and watching founders struggle with the interface layer. The data is there. The understanding is not. That's the gap.

The next interface for finance is a conversation. Not because conversations are trendy. Because they're how humans actually think about their businesses.

— Maxime Champoux, co-founder and CEO of Well