The Palantir Model

Where Strategy Goes Next

I’ve been chewing on something for a few weeks now and I can’t quite spit it out, so I’m going to do what I always do when that happens: write my way through it and see what comes out the other side.

It started with a piece I read by a former Palantir employee. And before we go any further, let me be clear: I am not a fan of Palantir. Not of their moral flexibility, their cosiness with authoritarian power, or the quiet devastation they enable across immigration enforcement, surveillance states, and whatever fresh hell they’re building for the highest bidder this week. But minus the war crimes, sans the ethical void, hold the dystopian overreach - and underneath all of it is a model I can’t stop thinking about, because it’s formidable.

I’ve been practicing some version of it for almost seven years now - largely by instinct, without a name for what I was doing or why it worked when everything else didn’t. Palantir has given me the frame. And now I’m certain: this is where strategic work has to go. The rest of the industry just hasn’t caught up yet.

---

The thing most people get wrong about Palantir is they think it’s a software company. It’s not. Or rather, the software isn’t the product. The product is embedded cognitive capacity. Which sounds like consultant-speak, I know, but stay with me.

Most enterprise software works the way you’d expect. It gets built centrally, generalised for the market, pitched by sales teams in bad suits, lightly customised (”we can change the logo and the colour scheme”), and shipped as-is. The client gets a demo, signs a contract, receives a login, and is left to figure out why none of it maps to how their organisation actually functions. Sound familiar? It should. It’s how agencies work too.

Palantir does something completely different. They take their people - and I’ll grudgingly admit their people are brilliant - and they embed them inside the client organisation. Not for a few weeks of discovery. Not for a “sprint.” For months. Sometimes years. And not as consultants doing quarterly business reviews over sad sandwiches in windowless rooms. As cohabitants. They move in.

And those teams don’t trust briefs. They don’t trust stated requirements. They don’t trust the client’s own articulation of the problem. They assume the organisation is wrong about what’s actually broken.

And it turns out, they’re usually right.

---

I keep thinking about why that is. And I think it’s this: most organisations don’t actually have a clear understanding of their own problems. They don’t have the internal alignment to define them. They don’t have the operating structure to fix them. And they definitely don’t have the culture to adopt the solution, even if someone handed it to them gift-wrapped.

What they have is seventeen competing priorities, three legacy systems held together with prayer and duct tape, a CMO who’s been there eleven months and will be gone in thirteen, and a vague sense that things aren’t working, but no shared language as to why.

Palantir bypasses all of it. They collapse the whole cycle - discovery, diagnosis, design, build, embed, iterate - into one long residency. They sit inside the mess until they understand it better than the people who created it. And then they build something. Not a product. A solution. Bespoke. Specific. Weird if it needs to be weird.

And then - this is the part that turns it into a flywheel instead of just very expensive consulting - anything that works gets productised. Every bespoke solution becomes a module. Every module becomes a building block for the next client. Every engagement makes the whole that much smarter.

This is how they ended up with fingers in every pie. Government, defence, finance, logistics, healthcare. Anywhere complexity is maximal, stakes are existential, and off-the-shelf has already failed. They’ve built what one former employee calls “a private pattern library of the world.” They’ve seen how procurement works inside the DoD. How supply chains break at Airbus. How NHS trusts actually make decisions. That accumulated context is the moat. Not the software. The knowing.

---

Now. Look at advertising.

We’ve sold essentially the same thing for decades. Off-the-shelf solutions from a centralised sales team called the agency. Here is our process. Here is our proprietary framework (which is just someone else’s framework with different words and our branded colour palette). Here is the conveyor belt. Please place your brief onto it and we’ll see you in eight-to-twelve weeks.

This model is breaking. I’ve said this publicly, repeatedly, to the point where I’m boring myself. But it’s true. Because the problems brands face now are so specific and layered and multifaceted that you can’t just buy a campaign anymore.

The challenges are different now. Fragmented audiences. Collapsing media ecosystems. Fandom-driven cultural dynamics that nobody truly understands. Organisational dysfunction so deep it’s basically architectural. Unpredictable algorithms. Legacy tech stacks. Hybrid in-house/outsourced models that create more friction than actual work. Too many data sources and too few people who can synthesise any of it into a decision.

Contrary to what we’re selling, brands don’t need more outputs. What they need is internal cognitive augmentation. They need people who can navigate complexity. People who can find the real problem - not the one in the brief, the actual problem, the one nobody in the room wants to name. People who can design systems, not campaigns. Fix workflows, not decks. Understand culture and operations and technology simultaneously, which is a Venn diagram with approximately eleven people standing in the middle.

The agency model was never designed for this. It was built for problems with clear edges - a brief, a channel, a campaign, a measurement framework. Neat boxes. But the problems now don’t have edges. They’re tangled. Political. They bleed across departments, sit in the gaps between job descriptions, hide inside legacy systems and unspoken tensions. No brief captures them. No scope of work contains them. And no conveyor belt can solve them.

---

But someone has to.

Embedded creative strategists. Fractional CMOs. Fractional CSOs. In-house “innovation residents.” Indie consultants doing six-to-twelve month immersions. Sprint-based problem-finding teams that come in, figure out what’s actually wrong, and leave before anyone can make them sit through a tissue session.

They’re not selling campaigns. They’re selling time inside the mess. Diagnosis as the product. Residency as the model. They’ve figured out - mostly by instinct - that the value isn’t in the output. It’s in the understanding that makes the right output possible.

The industry doesn’t have language for this yet. Doesn’t have structure. Doesn’t have economics. Just a bunch of talented people who’ve stepped off the conveyor belt and are building something new, without a shared name for what it is or any infrastructure to connect them.

---

Now I know what you’re thinking. I can hear it. What about in-house agencies? What about Oliver? What about all those brands who brought it inside?

Right, but that was still the same model. It was still an ‘agency’. Still following agency processes. Still a conveyor belt. Just in a different postcode. Just with a slightly sadder coffee machine and a lot of people who used to work at the agency now wearing lanyards.

But the Palantir model is different because the embedding isn’t about relocating existing structures. It’s not “what if the agency sat in your building.” It’s “what if we threw out the entire concept of the agency and started from what you actually need.”

The job is to get into the weeds. Into the nuance. Into the politics that nobody will admit exist but everyone navigates every single day. Figure out what strings to pull. Figure out what’s connected to what. Figure out why marketing and sales haven’t spoken civilly in eight months and how that’s showing up in every brief. And then build something bespoke. From the inside.

Tyler Cowen has this line I keep coming back to: “context is that which is scarce.” Going onsite, actually sitting in the planning meetings, lurking in the Slack channels, understanding why the CMO and the CEO have completely different definitions of what “brand” means and then capturing that tacit knowledge which no official document will ever contain - no amount of “immersion sessions” or “stakeholder interviews” gets you there. You have to live in it.

The difference between “tell me your problem” and “let me watch you work for six months” is the difference between stated and revealed preferences. And those, as any behavioural economist will tell you, are very different things.

This is the work agencies can’t do. Not won’t, can’t. The model doesn’t allow for it. You can’t bill for six months of watching. You can’t scope ‘figure out what’s actually broken.’ You can’t put ‘earn enough trust that someone finally tells you the real problem’ on a timing plan. So agencies sell what they can sell: head hours, deliverables, things that fit in a SOW. And the actual work - the diagnosis, the understanding, the context - gets skipped. Or faked. Or compressed into a two-day ‘immersion’ that captures the sum total of fuck all.

---

There’s also a structural element to the Palantir model that I think matters here, for those of you who like structure.

They split their engineers into two types. Forward Deployed Engineers go into client organisations and build fast, hacky, whatever-works solutions. Get it done. Solve the problem they’ve identified. Don’t worry about elegance. Then there’s a second group - Product Development engineers - whose job is to take those bespoke solutions and generalise them. Find the patterns. Build reusable infrastructure. Turn the one-off into the scalable.

This is the encoding mechanism. This is how the knowledge doesn’t just stay trapped in the heads of the people who did the work. It gets systematised. It becomes substrate.

And this is where I start to see the problem for the indie practitioners doing this kind of embedded work right now. Because they’re doing the FDE job. They’re in the weeds. They’re diagnosing. They’re building bespoke solutions. But they don’t have the second layer. There’s no one extracting the patterns. The knowledge compounds in their heads, maybe in personal frameworks they scribble in notebooks, but it doesn’t get productised. It doesn’t create leverage.

Every engagement is linear, not exponential. They get paid for their time, not for access to what they’ve learned. The pattern recognition gets sharper with each client, but there’s no flywheel. No “private pattern library of the world.” Just a very experienced person who can only be in one room at a time.

So I keep asking myself: what’s the substrate for the new model? What’s the encoding mechanism that lets accumulated insight become leverage?

Is it AI? Tools that help embedded strategists systematise their pattern recognition and offload the extraction work? Some kind of shared infrastructure across indie practitioners - a network, a commons, a protocol? Or is it the brands themselves, building internal capability that embedded talent augments rather than replaces?

Because without the substrate, you get brilliant bespoke work that doesn’t scale. You get artisanal consulting. Which is insanely valuable, but it’s not a model that transforms the industry. It’s a nicer, far more exciting corner of the market to occupy while everyone else keeps doing what they’ve always done.

And maybe that’s fine. Maybe transformation isn’t the goal. Maybe a thousand indie strategists doing embedded work, each building their own pattern library in their own head, is enough. A cottage industry of diagnostic brilliance.

But I don’t think that’s the case. Because the problems are too big. The complexity is accelerating. And these solo practitioners, however sharp, can’t cover the territory. Someone has to build the connective tissue, the infrastructure that lets this work compound.

---

One more thing, because it keeps nagging at me.

Palantir sent every new hire a copy of Keith Johnstone’s Impro. It’s a book about theatrical improvisation. Which seems odd until you understand what the FDE job actually requires. You’re not visiting a client. You’re moving in. Living inside someone else’s organisation. You have to become wallpaper - present enough to observe, invisible enough not to trigger the immune system. You have to read status dynamics, navigate politics you don’t fully understand, build trust with people who didn’t ask for you to be there. Know when to push and when to disappear. Play political games while looking like you’re not playing political games. It’s not consulting. It’s infiltration, but for their benefit, not yours.

And these skills aren’t optional extras. They’re the whole job. Because what Palantir actually does at its core - the unsexy engine underneath all the hype - is data integration. Getting access to enterprise data. Cleaning it. Putting it somewhere useful. And that work is fundamentally political. A team controls a data source. Their entire existence depends on being the gatekeepers. Their power comes from being the people you have to go through. Getting that data means negotiating with people whose relevance depends on you not having it.

This maps directly to brand organisations. The people who control the research. The brand tracker. The customer data. The “proprietary insights” that justify someone’s headcount. They’re not going to hand that over to some embedded strategist without a fight. The technical work of finding the real problem is often easier than the political work of being allowed to look.

Nobody trains strategists for this. We teach them frameworks and research methodologies. We don’t teach them how to read a room, build trust over months, or make someone feel safe enough to show them where the bodies are buried.

The thinking is broken and so is the access. You can’t fix one without the other.

---

There’s an uncomfortable part of the Palantir analogy that I’ve been circling around, and I should just say it.

They create lock-in. They become so embedded they’re essentially load-bearing. You can’t remove them without the building falling down. That’s not a bug, it’s the business model. Dependency by design.

Does it have to be that way? I don’t think so. But it’s worth naming: the deeper the embedding, the harder it is to leave. That’s a feature for Palantir. It might need to be a constraint for everyone else - capability transfer as an explicit goal, not just a nice-to-have.

And then there’s the client question. This model requires a certain type of organisation. One desperate enough to let strangers see how the sausage gets made. Palantir works in sectors where failure is existential and off-the-shelf has already failed. The client has to be in enough pain to tolerate the intrusion. To let someone see the mess.

Most brand organisations aren’t there. They’re still in ‘hire an agency and blame them when it doesn’t work’ mode. The dysfunction isn’t painful enough to warrant the vulnerability. In-house agencies were the band-aid - same model, different postcode.

But in that scenario, the pain hasn’t peaked yet. So for now, this model is for the outliers. The clients who’ve already tried everything else. The ones ready to be honest about the reality. That’s fine. That’s where it starts. The rest will come when they’re ready, or when they have no other choice.

---

I’m not going to pretend I’ve made peace with Palantir. I haven’t. The model is extraordinary. The company is rotten. But here’s the thing about architecture: you can admire the blueprint while despising the builder.

And what that blueprint is showing us is this: strategy has to move. Out of the agency, into the organisation. Not as a visitor with a deck, but as embedded intelligence. The slow, patient, political work of understanding a place well enough to actually help it. This isn’t strategy as we’ve known it. It’s something closer to organisational anthropology - with the mandate to change things.

That requires skills nobody’s teaching. There’s a reason Palantir hands every FDE a copy of Impro. Reading rooms. Building trust over months. Making yourself useful enough to be tolerated, invisible enough to observe, trusted enough to be told the truth.

Understanding becomes the product. Not the deck that comes after. The understanding itself - the diagnosis, the pattern recognition, the ability to see what nobody inside can see.

And encoding becomes the opportunity. The thing Palantir figured out that the indies haven’t yet. How to turn what you learn in one room into something useful in the next. How to build infrastructure, not just reputation. Right now, there’s no PD to their FDE. That’s the gap.

How do we build it? How do we encode understanding without flattening it into templates? I don’t know yet. These are real questions without easy answers. But the shape of something is emerging. And that’s enough to work with.