Enterprise AI implementation faces three major challenges, and excellent Forward Deployed Engineers need three core capabilities

This article comes from the WeChat official account "ToB老人家", author: Wang Daiming

Recently, I have been deeply involved in an enterprise AI project, and I have personally experienced just how difficult it is to become an excellent FDE (Forward Deployed Engineer).

At the current stage, there are three difficulties in enterprise AI implementation:

1. Unclear requirements

Customers often have high expectations for AI, but what kind of AI product do they actually need? What is the business logic behind this product? In fact, customers are not clear about this themselves.

For example, a customer hopes to use AI to improve lead conversion rates, but how exactly? What is the underlying business logic? Customers often haven't thought this through.

2. Unclear solutions

Even if the business logic is figured out, how to implement it with AI? Nobody knows.

There is a key point here: enterprises have very high requirements for accuracy. If hallucinations cannot be eliminated through engineering approaches, AI will be very difficult to implement.

For example, different industries and different enterprises all have specialized terminology and rules, and enterprises also have a large number of traditional software systems. How can AI accurately understand these terms and rules? And how can AI correctly invoke the corresponding software functions?

These are all entirely new territories.

3. Unclear path to standardization

Even if the AI solution is successfully implemented, AI projects still face the biggest hurdle: product standardization.

In fact, even abroad, purely custom AI projects easily lose money, and the key to turning losses into profits is product standardization.

However, many enterprise AI projects are exploring uncharted territory. Can they be standardized? How to standardize? These are all major challenges.

Because of the above three difficulties, an excellent FDE must possess three capabilities:

1. Business expert

Since customers themselves are unclear about their requirements, someone must sort out the business logic based on the customer's expectations.

For example:

Which is the key node in improving lead conversion rates?

For salespeople to do a good job at this node, what kind of suggestions does AI need to provide?

To ensure the quality of these suggestions, what data needs to be provided to AI? What industry know-how needs to be communicated to AI?

In the era of traditional software, although implementation consultants would also sort out business processes, they often stayed at the "process" level and did not truly go deep into the "strategy" and "SOP" levels.

However, for enterprise AI to deliver results, the business logic must be sorted out in much greater depth, which requires the FDE to have the knowledge and capabilities of a business expert. Otherwise, it can easily lead to extended research cycles, or even cause the AI system to become a "white elephant" that users simply abandon.

2. AI expert

For enterprise AI projects to be implemented, on the one hand, the FDE needs AI technical capabilities, such as prompts, RAG, and Vibe Coding.

On the other hand, the FDE also needs engineering capabilities. For example, how to build the semantic layer? How to represent business rules? How to trigger actions?

Because AI is developing so rapidly, talent with these capabilities is actually very scarce.

3. Product manager

The key to success or failure of the FDE model lies in whether the project can be standardized.

This requires the FDE to, like a product manager, examine from the very beginning whether the project has the potential for standardization, to invest cautiously in projects that cannot be standardized, and for projects that can be standardized, to think about the path to standardization from the very start.

As you can see, any one of these three capabilities is very rare on its own, yet an FDE must possess all of them simultaneously. This shows just how difficult it is to be an excellent FDE—no wonder some people say that FDEs are the CEOs of the future.

So how can qualified FDEs be cultivated? I think there are two ways.

The first way is to select and cultivate them from product managers/project managers.

For example, currently the most common source of FDEs is product managers, because relatively speaking, product managers are the role closest to FDEs.

Product Planet has invited many heads of enterprise AI products to share their experiences in the community. In my view, they are all actually performing the functions of an FDE: charging to the front lines and co-creating brand-new AI products with customers.

Product Planet currently has 20+ enterprise AI courses, all practical sharing from frontline AI practitioners. Welcome to click "Read Original" to view the full course schedule.

Another way is to form a joint team.

For example, a delivery-focused FDE may lack the ability to standardize products, so a senior product manager can be specifically assigned to oversee the project, thereby increasing the success rate of product standardization.

The AI project I have been involved with is a joint team, which includes a high-level salesperson from a major international company who plays the role of business expert.

Everyone leverages their respective strengths. As long as the first project is successfully implemented and a standardized product architecture is established, the difficulty of implementing subsequent projects will be greatly reduced.