Palantir And Forward Deployed Engineering: What Should We Believe?

1 hour ago 2

WAIC World Artificial Intelligent Coference

According to one former Palantir executive I talked to, in 2022, Palantir only had three nongovernmental/nonmilitary customers – what they call commercial customers. Today, 46% of Palantir’s business, roughly $2.1 billion, comes from commercial customers like Wendy’s and General Mills. Broadly speaking, many of these commercial deployments are focused on supply chain management.

Palantir’s success came from its ability to deploy AI, which, in turn, is driven by both its technology stack and its deployment model. This deployment model is called “forward-deployed engineering.”

Drawing inspiration from military terminology (forward-deployed forces), a forward-deployed engineer is a software engineer who works on-site to solve complex, real-world problems using Palantir’s platforms, such as Foundry and Gotham. Unlike traditional software engineers, these engineers focus on tailoring solutions for a specific customer, integrating data, configuring workflows, and ensuring the software works effectively in the customer’s operational environment.

Because of Palantir’s success, other enterprise and supply chain software firms now also offer forward-deployed engineering services. Kinaxis, a provider of supply chain software solutions, is one example. But, they argue, they do it better because they have domain expertise.

Other enterprise vendors, such as Anaplan, think the whole concept of forward-deployed engineering is oversold and underdelivers.

Back to Custom Solutions?

Decades ago, companies developed their own software. But this was risky; the failure rate was high. And the total cost of ownership was high. Hence, enterprise software development companies arose. These companies provide standard solutions that they sell to a wide range of customers.

“Palantir has a very different approach,” Danny Lutkus, a deployment strategist at Palantir Technologies, explained. Palantir is 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.”

Manik Sharma, the chief of agentic solutions at Kinaxis, says that while these AI solutions are customer-specific, they should not be considered “custom” solutions. Custom solutions were created when consultants used software development tools to create code. Now, AI tools are being used to stitch a solution together. In effect, a company like Palantir is a combination of an AI tool company and a system integrator that co-builds a customer-specific solution with its customers.

However, like the custom solutions of yore, these forward-deployed AI solutions have high failure rates. A 2025 report from MIT's Media Lab revealed that 95% of generative AI pilot projects fail to deliver business value. This study analyzed over 300 AI deployments and found that only 5% resulted in clear revenue growth or operational improvement.

However, when these unique AI solutions work, they can provide robust differentiation. CH Robinson, a leading 3PL, brags to Wall Street about its improved margins stemming from its “lean AI” journey. For example, they created automated quoting capabilities. The automation improved the win rate and margins associated with quotes.

Anaplan Does Not Believe Forward-Deployed Engineering Provides Lasting Value

I interviewed Charlie Gottdiener, the chief executive officer of Anaplan, a few weeks ago. Anaplan is a large player in the enterprise market, one of the relatively few enterprise software companies that generate over a billion dollars annually. Anaplan offers a cross-functional scenario planning and analysis platform whose applications span finance, supply chain, human resources, and sales and marketing.

One of the things we talked about was forward-deployed engineering. “So, here's why I think (forward-deployed engineering) is attractive to customers; they do a very good job of showing up and doing a POC (proof of concept).” They do this rapidly using a customer's data, and “they create a nice-looking dashboard. They use that to good effect; there's value in that in the selling process.”

In terms of the buying process, power has shifted from business leaders, such as a Chief Supply Chain Officer, to the CIO. Business leaders want results. But CIOs are being tasked with using AI because CEO’s believe AI could be transformative. AI budgets are controlled by the CIO. Quick POCs and pretty dashboards then get business leaders to acquiesce to an AI project. Also attractive is that these POCs are often sold under a fixed-fee arrangement.

Companies, Gottdiener went on to say, “like this until the deployment is done. Now they've got forward-deployed engineers that they have to pay for and rely on in perpetuity. If I want to make a change, I've got to pay them for it. I've got to pay them to keep it running. It's very hard to get off that platform. And they're running all around my organization trying to sell the next thing, so they've embedded themselves.”

Further, the capabilities are often limited. It is often “much more of a visualization and BI capabilities, as opposed to a new capability.” Despite the limitations, the CIO, who has championed the project, says ‘Give it time.’ “We don't, we don't want to trap people with our technology, we want them to use it because it's creating value for them.”

“It's a good selling model,” Anaplan’s CEO went on to say. “It's a good initial deployment model. It's not a great model for running the software. Not at all.” In time, as the word gets out, this methodology “will run its course.”

Kinaxis Believes FDE Is a Good Model That Has Been Poorly Executed

Before Manik Sharma was the chief of agentic solutions at Kinaxis, he was the head of supply chain digital solutions at Palantir. He understands forward-deployed engineering, its heritage, its shortcomings, and its potential.

Kinaxis is a leading provider of supply chain solutions. They are the fourth largest supplier in the supply chain planning market. They now refer to themselves as a leading provider of supply chain orchestration.

“Orchestration” is a good term for describing solutions built using an agentic AI framework. The company has increased investments in its Maestro platform to make it more agentic. The goal is to use Maestro to drive horizontal orchestration and continuous planning across not just SCP – which they provide – but also procurement, logistics, and finance – applications they don’t own. For better demand planning, marketing and customer relationship management components may need to be part of the end-to-end operational orchestration.

The agentic framework will allow them to use microservices to connect to other vendors’ applications and components. Generative AI will be used to create new applets and workflows, as well as to drive a much more robust user experience. These “outcome flows” will continuously sense, analyze, respond, learn, and then again sense. This they call a “beat.” To do this requires both a technology shift and their version of forward-deployed engineering.

The technological shift will require a knowledge graph to drive semantic intelligence grounded in a unified ontology.

Palantir introduced the notion that AI deployment, maintenance, and robust outcomes are owned by a highly technical team. But, Sharma said, “putting 25-year-old engineers with the customer, there is a problem, and that was the big problem I had in my previous company.”

Kinaxis wants to be responsible for the outcomes and maintenance of these composable solutions by deploying teams that provide customers with domain, industry, and technical expertise, depending on the stage of the journey.

A new supply chain solution should not be written by “a 25-year-old engineer writing Python code.” They don’t understand basic supply chain concepts like constraints or the bullwhip effect. They can’t match models from Kinaxis and combine them with other supply chain software companies' applications or components to drive robust execution.

Further, the use of AI tools is not solely the responsibility of the technical team members. Traditionally, integrating applications was about linking to the right rows and columns in a database. But semantic intelligence is about objects, their relationships, and their attributes. Domain experts can help technical experts better understand what these relationships need to be. A supply chain knowledge graph needs to be built with the right constraints, connections, and interrelationships.

“Steve, we are well into the journey, but it is a heavy lift. We have really increased the R&D budget for this.”

Read Entire Article