Before It Had a Name

It's 2006. A first-year analyst at a top-tier London consulting firm, doing work nobody had quite named yet. Twenty years later, that work has a name — and AI just changed everything about who can do it.

It's 2006. I land my first job at a top-tier consulting firm in London.

The work is painful. It is grand. It is powerful. Some of the most meaningful work of my career. Little did I know it would take twenty years to get a name.

Fast forward to today. Forward-deployed engineering. FDE. The most in-demand profile in tech right now.

This is that story. And where it goes next.

The Work Nobody Had Named Yet

Some allocations took days to run. This was 2006, my second project as a first-year analyst, fresh out of school, working out of the London office on a team that was building something nobody had quite named yet.

I had studied business, not engineering. The manager who led the team, Francisco Barea, is still a mentor today. He was the brain behind the methodology. The kind of professional who could hold three industries in his head at once and still answer your question without losing his place. He was my Shyam Sankar before I knew who Shyam Sankar was. Palantir hires kids straight out of high school as apprentices. This was mine.

The brief was simple to describe and brutal to deliver. Take any process industry with thousands of SKUs and customers scattered across geographies. Map every cost the company incurred. More than mapping. Understand what was actually driving each cost. Down to the invoice line. Customer by customer, product by product, shipment by shipment. Then aggregate back up and show what was actually profitable.

In plain terms: reconstructing the company's P&L from the invoice line up.

We called this Cost to Serve. The label undersells it. We anchored ourselves inside a company for eight to twelve weeks at a stretch, sitting with every head of department, often their extended teams. How long does this product sit in the warehouse. What does it cost to switch the line from one color to another. How many invoices do you cut for this customer in a month.

Those engagements were intensely painful. Consultants were selected on two things: the technical chops to model a multi-thousand-SKU business down to the invoice line, and a tolerance for pain. You can teach the first. You can't teach the second.

Memory

I remember the 4am calls with Jonathan trying to debug a multi-thousand-line allocation that wouldn't converge. The pain was vivid. The work was also addictive. Once you have built that granular mapping, you have established Memory. Memory is not data. Memory is data solved, as my friend and NEXT CEO Moodi drove home to me on a call this week. That Memory becomes the company's source of truth. It links operations to financial outcomes at a granularity nobody inside the company ever had. Executive teams stop guessing. They start making surgical decisions.

The shape this revealed was almost always the same.

Top customers drove most of the profit. The long tail dragged it back down. The biggest customer on the books was often the most value-destroying one. We would walk into the CFO's office with a model that contradicted twenty years of intuition, and the room would go quiet.

The output was a model, the same across clients. The input was hundreds of company-specific rules, never the same twice. The model was general. The rules were the company.

The Gravel Road

Twenty years later, I was listening to a podcast about Palantir's forward-deployed engineer model. Engineers sit inside client operations for weeks, find the specific gravel that defines that client's problem, then feed it back to a platform that absorbs it. They call it main road versus gravel road. The main road is what every company shares. The gravel road is what makes each company itself.

I almost missed my stop.

We were doing this in 2006. Smaller scope. Different tools. What Palantir's Ontology does now at industrial scale, we were doing inside one company at a time. The essence was the same.

For thirty years, software couldn't go this deep. Building into one company's gravel was prohibitively expensive, so everyone settled for the same compromise: ship a generic tool, let each company twist itself into the tool's shape, charge a subscription, repeat. The scheduling tool treats scheduling. The CRM treats the contact list. The question of why scheduling broke at this specific company stayed outside the product. SaaS treats the symptom. The economics never allowed it to treat the root cause.

What AI Changed

AI changed the math. Building the engine used to take armies of engineers and budgets to match. Today a small team builds it at a fraction of the cost. Knowing what belongs in the engine and what belongs to each client remains the scarce part. That knowledge comes from the gravel. The gravel stays painful. The gravel stays where the value gets made. This is what I meant when I wrote that SaaS is mid-platform pivot.

At Exponential Partners we have worked closely with companies that, on the face of it, share nothing in common. A tennis academy. A personal security and surveillance company. A fleet of rental cars.

They all have the same nightmare.

Scheduling.

And underneath the scheduling, invoicing.

The engine we are building handles the common denominator. We call it Factory. The rules we collect on-site, customer by customer, head-of-department by head-of-department, are the gravel. Same playbook as 2006. Different stack.

I was in a meeting this week with a CFO who wanted to figure out his cost of acquiring a customer. One number. He couldn't get to it. The data was there. The barrier was the long string of follow-up questions nobody in his company had the energy to answer.

That's the SMB reality. The questions that unlock the most value are the ones the company is least equipped to ask itself. Enterprise can grind through the gravel with internal teams. SMBs need someone to come in with the pain tolerance the work requires, do it with them, and leave behind an engine that runs on rules nobody else could have written.

OpenAI made this thesis explicit in December. An equity stake in Thrive Holdings, paid in engineers embedded inside portfolio companies rather than cash. A few months later, a joint venture with TPG, Bain Capital, Brookfield and Advent International, backed by over four billion in PE equity. Anthropic is doing the same with its own group of financial institutions.

PE is where ownership and management are aligned around making this work. The PE-backed engagements were a bliss, because everyone in the room wanted the answer. PE is the cleanest go-to-market for the FDE model in the mid-market. The largest AI labs in the world are telling you so by signing the checks.

The Engine and the Road

The way we enter is small. We find an acute symptom, treat it, prove the fix. We earn our way toward the harder problems our clients need solved. Forward-deployed engineers sell the first yard. If the work is good, the next eight come.

This is muscle. A product manager, no matter how brilliant, has been trained to ship custom features for specific customers. The common-denominator question gets learned somewhere painful. Once you have it, you keep it.

What we're building, what Palantir built before us, is a meta product. General at its core. Hyper-specialized at its edge. It sits on top of everything else the company already runs. It is the layer SaaS was never able to build, because SaaS was paid to treat the symptom and stay there.

The anti-SaaS.

Too bad for the legacy mammoths who built their margins on selling the main road and pretending the gravel didn't exist.

The work doesn't scale. Sitting inside real companies, learning their gravel, mapping their rules. None of that scales.

The engine does.

Paul Graham wrote that the path to building something that scales is to do things that don't scale. We're using that advice to build the layer that gets bigger every time we sit with one more company and absorb one more set of rules. That same logic makes one-of-one at scale finally possible.

The work is gravel.

The leverage is the road.

The road is the product.

We're Exponential Partners. We sit inside real companies, learn their gravel, and leave behind an engine that runs on rules nobody else could have written. If you're trying to figure out where the trapped value lives in your business, reach out.

Originally published on Sacha's Substack. Republished with permission.