業務を変える前に実際のワークフローを把握する

The FDE's first job is to learn how work really flows

Forward-deployed engineering (FDE) here means technical work that goes deeply into a customer's organization and operations. An FDE does not simply hand over a general AI tool from outside. They investigate who employees actually check with, where they make decisions, and whom they hand work to when something goes wrong.

Look at one department first, not the whole enterprise

According to the speaker, an FDE first enters the customer's company and narrows the scope of the investigation to one department. Trying to understand the whole company at once would make the relationships among operations too broad. The FDE interviews process leads in that department: people who know the flow of each operation well.

For example, the talk names the following operations as investigation targets in a finance department.

• AP processing
• AR processing
• Card reconciliation
• Banking operations
• Billing
• FP&A

The FDE does not ask only, "How does this usually proceed?" They also ask, "What do you actually do when the normal procedure fails?" In other words, they investigate both the procedure written in documents and the procedure employees follow in the field.

The normal path alone cannot reveal how a company operates

An operations document may describe an ideal normal path, or golden path. It may contain the steps for a problem-free case and only a few exceptions. In practice, exceptions overlap. A person may change, required information may be missing, or another department's response may be delayed. These small responses are easy to leave out of formal documents.

In the speaker's example, after a normal AP flow fails, the work is passed from Sarah to Chris. Reconciling the purchase order and invoice between them takes four days.

The important point in this example is not only the four days. A process that looks like one operation in a document may not specify who actually takes responsibility for the problem. The handoff from Sarah to Chris is also more than a simple change of person; it shows a dependency that crosses department or role boundaries. Without knowing these hidden owners and handoffs, an AI agent may not fit the real flow of work, no matter where it is placed.

Design automation after discovery

What the FDE discovers is each company's unique way of working. Even when an operation has the same name, companies may differ in who approves it, which department holds the information, and which person receives the work after failure. The FDE organizes the normal flow together with these exceptions and cross-department dependencies so they can later be represented as workflows and graphs.

Therefore, the goal at this stage is not to build a new agent immediately. The FDE first enters the workplace and gathers different perspectives from several process leads. They then understand the normal path, exceptions, decisions, handoffs, and waiting time as one picture of the operation. Only in the next stage can that understanding be used to design an AI-centered workflow.

Note: Scope of this chapter

AP, AR, and FP&A are labels for finance operations used in the talk. This chapter does not guess their formal expansions or company-specific definitions. The Sarah and Chris story is also an example the speaker gave from customer discovery, not a case study independently verified elsewhere. The important pattern is the discovery that real enterprise operations contain handoffs and exceptions between people that are absent from the documents.