Trading Margin for Moat: Why the Forward Deployed Engineer Is the Hottest Job in Startups

It turns out dethroning Salesforce isn’t as simple as spinning up an OpenAI-enabled voice agent. Companies attempting to replace core workflows owned by legacy systems of record with lightweight, wrapper-like integrations too often see their agents break or fail outright.

Paradoxically, we’re witnessing explosive growth for complicated AI products in sales, support, and legal. Enterprise software companies that tackle complex workflows are regularly growing from $0 to $5 million, $10 million, or beyond $20 million in ARR in their first two years. So how are new AI startups solving for these intricate use cases?

Enterprises buying AI are like your grandma getting an iPhone: they want to use it, but they need you to set it up. 

The PLG obsession vs. historical reality

For the better part of the last decade, it’s been broadly assumed that product-led growth (PLG) is superior to implementation-heavy enterprise software. The allure is obvious: PLG promises greater scalability and higher margins. This obsession has been driven by success stories like Atlassian, Slack, Figma, Notion, and Dropbox (and more recently, ChatGPT and Cursor). All of these products offer simple, single-player modes, are easy to adopt without needing a sales call, and can be purchased directly with a credit card — no lengthy scoping or enterprise contracts required.

During platform shifts, however, companies have room to experiment and build more intricate products that don’t follow this standardized formula. Salesforce, ServiceNow, and Workday did this during the transition from on-premise to cloud platforms. Each of these companies sells an enterprise platform requiring significant implementation, services, and support, which is the antithesis of bottoms-up PLG. However, in nailing complex implementations, these companies achieved dominance with impressive market capitalizations: $254 billion for Salesforce, $194 billion for ServiceNow, and $63 billion for Workday. Their combined value dwarfs that of the top PLG companies, and it’s not even close.

The implementation necessity

Salesforce, ServiceNow, and Workday were able to achieve success because they did a significant amount of implementation work. Enterprise AI products have an even more pronounced implementation requirement, as they require deep integrations and context. To navigate this complexity, services organizations handle the heavy lifting of securely connecting the AI application to internal databases, APIs, and workflows, ensuring models have the context they need — historical records, business logic, and more — to deliver value. 

Category-defining companies like Salesforce and ServiceNow became indispensable largely because of their ability to integrate with a company’s internal systems and context:

• Salesforce is only valuable when configured with the right fields for tracking sales data
• ServiceNow only works when ITSM workflows are properly integrated and orchestrated
• Workday requires extensive HR data migration and configuration

This customization effort initially results in lower gross margins and higher burn rates. At IPO, ServiceNow’s gross margin was 63.2%, and Workday’s was just 54.1% — far below the ideal ~80% for software. Even Salesforce, generally considered the gold standard, reportedly burned over $52 million to generate $22 million in revenue before developing its partner ecosystem.

These complex businesses are easiest to build early in a platform shift, when workflows are still taking shape and the payoff for replacing an entire system of record is highest. No company would replace a core system like a CRM without a massive estimated ROI. This was the wave Salesforce rode to disrupt Siebel amidst the transition from on-premise to cloud. Implementing Salesforce required a much lower IT investment and no on-prem installation, enabling a lower cost and better experience for every user.

The AI platform shift is different from — and in some ways more exciting than — the previous transitions to cloud or mobile because the implementation work required to make agentic experiences can itself be streamlined and automated by AI. Historical integration work might require outreach and collaboration with partners, mapping data fields, navigating data transfer between different coding languages, and understanding various internal guidelines. This is the kind of work that can now be done much more efficiently (and in some cases, entirely!) with AI. Imagine an agentic browser instance being spun up to retrieve information from systems lacking APIs, or a conversational agent streamlining workflows across internal and external communications. 

Once those workflows and behaviors are established, these companies possess “moats” that allow them to increase prices and build implementation ecosystems. Such pricing power is reflected in increased gross margin percentages over time; in 2024, ServiceNow had 79% gross margins; Workday was at 75%. 

Agents become active coworkers

Today, even benchmark systems of record like Salesforce, ServiceNow, and Workday are vulnerable, as the workflows and human behaviors that once drove lock-in can now increasingly be handled by AI agents. This shift changes everything about how to most effectively own the data ingestion point. 

Software is no longer aiding the worker — software is the worker. Software doesn’t need the same graphical user interface on top of a database to operate; it can autonomously complete tasks end-to-end. But as those tasks get more complex, fulfilling them becomes more challenging. For AI agents to truly be on par with human workers, companies will need expert services to redesign job functions and processes around this new approach. Without hands-on implementation support, AI risks falling short of the standards set by a dedicated employee. With the right support, however, agents can take on a more holistic role, unlocking far greater business value than basic task automation ever could.

We’re seeing early signs of companies adopting implementation service models — similar to those used by Salesforce, ServiceNow, and Workday — to address this problem. Just examine the products and technical docs put out by the leading AI application companies. They have more in common than their landing pages indicate. Much of the product differentiation comes from how the same underlying technology is implemented and applied differently across customer sets. 

Agent onboarding can learn from human onboarding

Onboarding human employees is hard. “Onboarding” AI agents that solve for use cases in areas like sales, support, or legal, end-to-end, is no different. To be effective, advanced agents require active management, guided learning, and rich context that comes from read/write access to internal systems. For humans, this role is typically played by a manager. For the best complex AI application companies, it’s often filled by a professional services employee, sometimes rebranded as a forward deployed engineer or an implementation/solutions specialist. 

Take a company like Decagon, whose agents automate customer support. The company has a sizable team of human “Agent Product Managers” who work closely with customers to stand up AI support agents. These PMs are relatively technical and help implement, integrate, and problem-solve so Decagon’s agents can resolve a wide range of customer support requests autonomously. 

Professional services teams can help customize AI applications to a company’s specific needs and ensure they’re successfully implemented. AI can be magical, but only when it’s properly leveraged. That’s where forward deployed teams come in: to operationalize the model into a real-world solution.

Startups should aim to become the system of work

Incumbent systems of record are valuable because they own workflows that capture valuable data, which helps them become their company’s source of truth. If emerging AI applications really want to succeed, they should similarly position themselves to become the system(s) of work that generate, capture, and store valuable company data. As they race to own the data ingestion layer, it’s shortsighted to be optimizing for 80% gross margin. The only thing businesses should be optimizing for today is growing total gross profit as fast as possible. They should aim to build a durable moat by controlling where and how the data enters the system.

It’s not just emerging application companies that are seeing the opportunity here, either. As the foundational models have become more interchangeable and owning the application layer has grown in importance, even model providers like OpenAI and Anthropic have been hiring forward deployed engineers and solutions engineers to win at the enterprise. As of this writing, 22 of the 311 open roles on OpenAI’s career page fall into these categories. 

Best practices for teams building their first services team

In talking with leaders of companies that have built elite services organizations, several key themes emerged: 

1. 1. Sell smart. Pick the right customers and start small in scope. A young company and product can’t be everything to everyone. Identify an ICP with at least some commonalities across systems and use cases to maximize learning.
   2. Optimize incentive structures: Forward deployed teams need to be closely aligned with account executives. Pricing and compensation should reflect that. It’s rare to find implementation leads that want to carry a quota, and that can also lead to counterproductive behavior. Instead, design incentive structures that allow services to be sold at cost. As ACVs grow, all that matters is success (and retention).
   3. Design systems thoughtfully: Build common software libraries and design APIs in a way that is easy to integrate with and understand so that these integrations are easier. Consider what public APIs and surfaces are available to customers. While being thoughtful about this up front may slow down development in the early days, it lays the foundation for rapid growth in the future.
   4. Leave a trail. Outsourcing to ecosystem providers requires clear and consistent documentation. Design with that future in mind and document everything.
   5. Build/buy tools to automate: Optimize for maximum speed and scalability in your services motion. This is the major unlock that will enable complex AI application companies to scale faster — and potentially with a lower minimum ACV — than the previous generation. Best-in-class companies are automating as much of the process as possible, including tasks like process mining, data pipeline automation, system integrations, and combing through API documentation.  This speed compounds and will create a competitive advantage over time. 
   6. Hire curious hustlers: This kind of implementation work doesn’t require a PhD, but it does require hustle. The best implementation leads are high agency and insatiably curious. In some instances it can be helpful to also pair these qualities with some exposure to the industry being served — such as legal or automotive — but it’s critical to hire people with a healthy disregard for the status quo.
   7. Create the right feedback loops: Forward deployed teams — technical or not — are closest to the customer. It’s critical to have the right feedback loops and clean information sharing between the front lines and product. Cut out games of telephone to accelerate development.
   8. Be there in person: It’s cliche for a reason. Obviously, going to the customer can make the sales process easier, but it also improves your chances of success with professional services. Being physically present can help as you sort through internal power dynamics and drive adoption of new tools. 

Addressing the counterarguments

Critics of this approach might argue that relying on a professional services motion limits scalability, that lower gross margins reflect a commodity product, and that professional services should be done by ecosystem partners, rather than the company itself. That’s because every company, in a perfect world, would like to be a beautiful, bottoms-up PLG winner. The reality, however, is that in the vast majority of instances this approach falls short. The more durable and integrated models consistently have some form of hands-on support, such as a forward deployed or solutions engineer. 

Hands-on support can also enable young companies to accept larger contracts, potentially powering faster and more durable topline growth than lighter-touch approaches. These early services contracts can also validate customers’ willingness to pay and hint at what future contract values might look like. Over time, ecosystems can be powerful tools for scalability and margin protection. However, much like the transition from founder-led sales to a dedicated sales team, young companies need to learn how to do it themselves before they can teach others to do the same. Salesforce, ServiceNow, and Workday have all demonstrated that high gross margin businesses await the sticky system of record. The same high gross margin outcomes are in reach for these new systems of work, and it’s likely even closer than historical precedents, given how much more scalable these services operations are today. 

A critical element of any elite agentic application today: services

The rise of AI has created a once-in-a-generation opportunity to build new systems of work, but successful implementation requires embracing a truth that Silicon Valley has long rejected: sometimes human-intensive services are necessary to create transformative software. Just as Salesforce, ServiceNow and Workday all had to invest heavily in implementation, support, and integration services before becoming the massive software businesses they are today, the most promising AI applications should follow the same playbook. 

Founders who obsess over gross margin and scalability, while neglecting the hard problem-solving required to own the system of work, risk missing the forest for the trees. We believe the companies that will dominate this platform shift won’t be the ones with the prettiest interface or the cleanest P&L in year one — they’ll be the ones that effectively replace human workflows with AI, capture critical company data, and solve genuine business problems, even if that means digging in with implementation work along the way. By embracing the professional services motion today, AI startups are positioning themselves to become the systems of record tomorrow, with the market capitalization and high margins that eventually follow. 

A big thanks to Joe Morrissey, Santiago Rodriguez, Justin Kahl, Peter Doyle, John Solitario, Ben Scharfstein, Charlie Beall, Yousef Hindy, Robby Allen, Perry Ha, Adam Erickson, and others who contributed.