The Palantir Approach

“What Palantir does is a bit different,” Mr. Lukus said. He described how decades ago, companies developed their own software. But this was risky; there was a high failure rate. And the total cost of ownership was high.

Hence, enterprise software and, eventually, supply chain software development companies arose. These would include companies like SAP and Oracle on the ERP and supply chain side, as well as best-of-breed supply chain software companies like Blue Yonder and Manhattan Associates.

First, these companies sold in a software license model, then they moved to software-as-a-service. These companies provide standard solutions that they sell to a wide range of customers.

“Palantir has a very different approach,” Mr. Lukus explained. First, they are skeptical that the standard solution actually aligns with how companies want to run their businesses. Most companies end up using Excel and other offline workflows, which is an implicit admission that the standard solution does not work for many. The ERP and SCM solutions are just too “rigid.” “We have humans doing manual data integration and creating their own logic. There are a lot of bad things that happen as a result of that.”

The second thing Palantir argues is that if you use a standard template to run your business, your capabilities are not differentiated from your competitors'. “You have ceded” your strategic differentiation. And every time you ask your software partner to develop functionality that will help you run your business better, you are giving those same capabilities to your competitors.

Palantir uses “forward-deployed engineers” to provide solutions that are specifically tailored to their customers. These engineers determine the capabilities and processes their customers want to develop to fill the gaps left by their standard software. Essentially, the Palantir solution is built on top of solutions like SAP and Oracle.

The Palantir platform is used to build these new company-specific capabilities. This toolkit approach builds custom software. Initially, Palantir engineers do this, but their customers can also use the toolkit for this purpose. These capabilities can be developed without upgrading existing software systems or replacing legacy software, a process that could take years.

Palantir has “an abstraction layer,” Mr. Lukus continued. “It's an operating system that allows us to essentially sit on top of whatever the underlying infrastructure is, and the architecture is, and define data integration, both structured and unstructured, define the logic integration, both deterministic and non-deterministic.” The platform also defines which decisions can be made by which users and how the data they generate is written back into the underlying systems.

This abstraction layer is based on Palantir's Ontology. “The Palantir Ontology, according to Palantir’s website, is an operational layer for the organization.” The Ontology not only integrates with a wide range of applications, datasets, virtual tables, and models, but also with the real-world objects and logic needed to run a large enterprise. This includes physical assets - like plants, equipment, and products - to concepts like customer orders or financial transactions. “In many settings, the Ontology serves as a digital twin of the organization, containing both the semantic elements (objects, properties, links) and kinetic elements (actions, functions, dynamic security) needed to enable use cases of all types.” Their ontology layer may be the most differentiating feature of this solution.

Palantir and AI

In the “age of AI”, Mr. Lukus continued, it “is even easier for me to go and build customer-specific S&OP (sales and operations planning) models or to build scheduling and logistics models that read and write with the underlying systems.” Anthropic’s Claude or OpenAI’s Codex can be used for code generation.

In supply chain management, we often refer to AI in terms of machine learning or optimization. Machine learning is a technique that has been used for over 20 years to improve demand forecasting and inventory planning. Optimization has also been used for decades. Optimization models account for supply chain constraints, such as factory line throughput, and seek to maximize throughput based on their understanding of these constraints and company goals.

AI for Software Coding

But Mr. Lukus was using the term "AI" differently. AI is used as a software development tool that can build code much more quickly than was historically possible. The historical way that software was built was very expensive. Because it was so expensive, consultants and system integrators conducted months-long studies to understand all the user and business requirements. Now, forward-deployed engineers can go in, quickly build models and workflows, get feedback, and iterate until they get a solution that fits.

“The speed of being able to build that type of software is dramatically lower and dramatically cheaper,” Mr. Lukus asserted. That is why “SaaS is dead.”

But what about the counterarguments? Once the software is configured, the company is far from finished. The software needs to demonstrate that it can be scaled. It also must be extensively tested. Mr. Lukus said generative AI can do this as well.

Furthermore, companies may not have the internal processes to keep up with changes in laws, such as tariff changes. A software company will have domain experts specifically focused on keeping its software up to date with legislative changes. Software agents, he said, could be directed to the relevant websites, and when legislative changes occur, the right managers in a company could be notified.

Mr. Lukus described a complex supply chain workflow they built, which was then utilized by a few workers within two months. By the end of the quarter, it was fully deployed to a large team.

AI Agents

Secondly, AI is used in the sense of agents. Agents mimic the actions of a human with a particular role.

So, I asked, “It's not that the agent is necessarily incredibly intelligent, like some advanced form of optimization. It's that it's part of this end-to-end process, and it can collaborate with other agents. “Exactly right,” Mr. Lukus said. For example, “I can have an agent say, 'Hey, is this a new purchase order or not? And if it's a new PO, I should parse it, match it to the master data, and determine, 'Hey, it's this customer, and they want this product.'”

Other agents could be used to determine if the products are in stock or need to be manufactured. And if the product needs to be manufactured, how long would it take? In this case, a scheduling agent using linear programming provides the answer.

I expressed some skepticism about building optimization or forecasting agents with code-generation tools. But Mr. Lukus asserted that Claude and Codex can do it quickly. I am still skeptical about this. Quickly does not mean well. Some planning software companies have spent decades developing cutting-edge mathematical algorithms to solve various optimization problems. Mr. Lukus admitted that most manufacturing clients initially used the solution for Sales & Operations Execution. S&OE is based on sensing exceptions and creating automatic responses. In short, it is more about creating a new workflow than advanced math.

Final Word

With most solution companies, I want to verify the claims the supplier makes by speaking with one of their customers before writing about them. But Palantir is a large company generating a lot of buzz, so I proceeded without garnering proof. While I can’t verify these claims, their website does offer quotations from Wendy’s, Tyson's, General Mills, and other companies touting its effectiveness.

I also know that Aera Technology offers a similar solution to solving the S&OE problem. I did hear an Aera customer who said the agentic approach to solving these types of problems is effective.