The Forward Deployed Engineer, Chapter 7: Customer Discovery and the Messy Reality

This is Part 7 of a series walking through my book The Forward Deployed Engineer. In the previous chapter, we closed Part II with the soft stack. This one opens Part III — the deployment playbook — with the most important fourteen days of any engagement: discovery.

The first two weeks of an FDE engagement decide most of the rest of it. The team that uses them well wins the next twelve months. The team that doesn’t spends the rest of the engagement paying down the debt of skipped discovery. I have watched both, and the difference is not how much technical work gets done in the first fortnight — it is whether the team produced the three artifacts that a well-run discovery is supposed to produce. None of them is the deployment itself. All of them are prerequisites to it.

The first artifact is the stakeholder map I described in the previous chapter, now populated with real names from real conversations. The second is the workflow inventory, a structured documentation of the operational reality the deployment is meant to change. The third is the Eval-Customer Split, the deliberate decision about which parts of the workflow are eval surface (the FDE owns the bar) and which are customer surface (the customer owns the bar). Each of these has its own construction discipline, and each has its own characteristic shortcut that quietly fails later in the engagement.

The dominant tool for building all three is the customer interview, and the chapter argues against the default version of it. The conference-room interview, with five operators and an executive sponsor and a sixty-minute slot, produces low-density signal and high social pressure to converge on the executive’s view of how things work. The version that produces denser, more honest information is the async interview — short, written, structured, conducted over the operator’s schedule rather than yours. It feels less efficient. It is more efficient. It also protects the operator’s actual workday from being eaten by the deployment they’re supposed to benefit from, which matters more for trust than most engagements realize.

The Weird Tuesday Problem

What I want to spend a moment on is what I’ve come to call the Weird Tuesday problem. Every workflow has one. The operator handles it in twenty minutes every few weeks, deals with it routinely, and would never bring it up in a structured interview because it’s not what they do “normally.” And yet the Weird Tuesday is precisely what breaks every greenfield deployment design, because the deployment was built around the normal case and the model doesn’t know what to do with the case that wasn’t documented. The technique for surfacing Weird Tuesday is to ask operators not what they do in general but what was on their screen at 3pm last Tuesday. The answers are always more interesting than the abstracted process diagrams the executive sponsor produces, and they are the most useful single output of a well-run discovery.

The workflow inventory captures all of this in a format that doesn’t bury Weird Tuesday in a process diagram. And the Eval-Customer Split — the most strategic of the three outputs — uses the inventory to draw the line between what the FDE team is going to be measured against (the eval surface) and what the customer team is going to own (the customer surface). The split changes how the team builds, how the success metric is written, and where ambiguity gets resolved when the model behaves unexpectedly in week eight. Functions that skip the split end up arguing about ownership for the rest of the engagement.

Common Mistakes, and What Discovery Produces

A few discovery mistakes deserve explicit mention because they appear in almost every failed engagement I’ve reviewed. Allowing the executive sponsor to speak for the operators, when their picture of the work is usually three years out of date. Conducting interviews only with the demo-ready operators — the ones the customer routinely puts in front of vendors — and missing the ones whose work is messier and more representative. And the Friday-afternoon kickoff, in which discovery launches into a weekend and loses momentum before it begins. At the end of two well-run weeks, the FDE team has four artifacts in hand: the stakeholder map, the workflow inventory, the Eval-Customer Split, and a one-page deployment thesis the executive sponsor has signed off on. Tomorrow, the inner loop begins.