Finance Automation Is Having Its Sourdough Starter Moment

The fundamental lesson isn’t about desire or ingredients, it’s about the gap between growing a starter and maintaining one. These are completely different challenges. The first requires enthusiasm; the second requires changing how your kitchen works.

I’ve spent 2025 watching finance teams get stuck in exactly the same place.

The Kitchen That Can’t Bake

The conversations I have with CFOs follow a pattern. They’re excited about AI, they’ve read the articles, attended conferences, and maybe run a pilot. But then they show me their architecture: legacy ERPs that can’t expose data properly, point solutions that don’t talk to each other, manual reconciliation processes that shouldn’t exist in 2025. They’ve got the starter culture bubbling away – the ambition, the budget approval, the stakeholder buy-in – but the kitchen itself can’t support what they’re trying to bake.

Think about what this parallel actually means in practice. Both involve living systems that resist rigid rules. Sourdough starters are colonies of wild yeast and bacteria – they respond to temperature, feeding schedules, and flour quality in ways that can’t be fully programmed. For twenty years, finance automation assumed the opposite: that every transaction would follow a predefined path, that exceptions were rare bugs rather than daily reality.

The 30% Brick Wall

That’s why rule-based systems automated 30% of work and then hit a wall. They were designed for a world where payments arrive with perfect reference numbers and customers never consolidate invoices. It’s like writing a recipe that only works if your kitchen is exactly 21°C and your flour is exactly 12% protein. Real kitchens – and real finance departments – are messier than that.

What makes AI different is that it’s built for messy, unstructured environments. It learns patterns rather than following rules. When a payment arrives with a typo in the reference field, it doesn’t freeze and wait for human intervention – it recognises intent from context. This is precisely how an experienced baker can look at their starter and know it’s ready, even though it doesn’t match the textbook description of “doubled in size with large bubbles throughout.”

You Can’t Bake Sourdough in a Microwave

But here’s where the sourdough parallel gets uncomfortable: AI adoption requires the same thing successful bread-baking does. You can’t just add a starter to your existing routine. You have to rebuild the routine around the starter’s needs.

Most finance teams I work with are trying to drop AI into their current architecture the way someone might try to bake sourdough in a microwave because that’s the heating appliance they already own and use. The technology is irrelevant if the surrounding infrastructure can’t support it. This is the architecture problem, and it’s why so many pilots produce impressive demos but never scale.

We were at Google’s Gemini Founders Forum when they named it: AI theatre, the gap between what looks transformative on stage and what actually survives in production.

Prove Something and Then Scale

The contained‑value approach we advocate – focused use cases, defined datasets, and measurable outcomes – is the same advice every sourdough guide gives. Start with one simple loaf. Master that. Then experiment. Don’t attempt twelve kinds of bread at once with a starter you fed for the first time yesterday. Yet that’s precisely what happens in corporate finance when companies launch sweeping “AI transformation initiatives” before they’ve automated a single reconciliation workflow successfully.

I tell treasury teams: prove you can automate one thing completely before you write a roadmap for twenty things. Get cash visibility working in real-time for one entity before you roll it out globally. The discipline is in resisting the urge to scale before you’ve proven you can execute at a small scale.

When the Dutch Oven Finally Arrives

The interesting thing about the regulatory wave hitting finance is that it doesn’t actually change the fundamentals. PSD3 and ISO 20022 will help – they’re the equivalent of finally getting a proper Dutch oven after months of failed attempts with a sheet pan. Standardised connectivity will remove fragile data flows, which is one of the biggest technical barriers we see.

But here’s the uncomfortable truth – a tool only matters if you’ve learned the fundamentals. Teams that have been practicing contained-value AI deployments will scale quickly when those regulations hit. Teams that have been waiting for perfect conditions will still be reading articles about transformation.

Nobody Wants to Learn Using a Microwave Kitchen

What all this means for talent is straightforward but nobody wants to say it out loud. The best finance graduates don’t want to work at companies still reconciling CSVs across fifteen spreadsheets for the same reason nobody wants to learn to bake at a restaurant still using a microwave. It signals that the organisation doesn’t understand its own craft.

And here’s where the gap becomes a problem. When I talk to treasurers who can’t provide real-time liquidity visibility, I think about what message that sends to incoming talent. The narrative is clear: if your systems look and behave like a microwave kitchen, top talent will go elsewhere.

Where Reality Bites

The fundamental lesson for 2026 isn’t about technological maturity, it’s about exposure. The technology is already adequate. What changes is that the space between companies that executed and companies that deliberated becomes impossible to hide.

In sourdough terms, you can claim you’re “really into artisan bread” for only so long before someone asks to see your loaf.

And that’s where reality bites. Finance leaders sitting in 2026 with transformation roadmaps instead of working pilots will face a simple reality: their competitors have been baking bread while they’ve been perfecting their starter. And in the treasury department, nobody cares how good your intentions are without proof.