Quick summary · Chapter 17
The Relationship Discipline
First editionThe essential ideas, decisions, and diagrams from this chapter. Follow the section links for the examples, calculations, and fuller argument.
First-edition visual companion · Selected from the current chapter · 8x10-inch PDF
The main idea
Whoever can warn us about a change in demand is probably doing something other than waiting for the capacity team’s questionnaire. They may be trying to close a sale, decide which customers get a feature first, or work out whether a migration is going to finish at all. Your deadline is another deadline in that person’s day, and before answering, they have a perfectly reasonable concern: what are you going to do with the answer?
That concern usually has a history. If what they told us last time was later held against them, or reporting a change earned criticism instead of help, we have taught them that keeping quiet is easier. We can call this a communication problem, but we helped create it.
Most of these people do not report to the capacity team, and nobody can be ordered to mention a cancellation they are still arguing about. What we control is what happens to the information once it reaches us, and that decides whether we hear about the next change in time.
Make early information worth sharing
I would begin by asking what decision the customer is trying to make, then work backward to the information that could improve our answer. An earliest plausible date, the work likely to change, a credible range, and somebody who can update it may be enough to preserve an option. A central template can help us remember these questions; it should not turn every empty field into a reason to send the request back. If I cannot explain how a field affects this decision, I should be careful about making someone else spend an afternoon filling it in.
The return trip matters just as much. Show the customer the supply range or conditional cost estimate their information made possible, including the part we still cannot answer. After the event, come back with what happened and what we learned, even if their adjustment made the forecast worse. Otherwise, from their perspective, information disappears into a process that sends another questionnaire next month.
Work through the specialists
A capacity team needs access to specialists who have lived with their systems, although hiring one person who has suffered through a difficult database does not somehow give us mastery of all databases. The tools in the earlier chapters can make it tempting to imagine a central team measuring everything independently. At enterprise scale, I would build partnerships with the people who understand the streaming platforms, databases, AI infrastructure, and applications, then hire central staff who know enough to ask them worthwhile questions. Those are already demanding jobs.
I want the specialists to disagree with me when my question is incomplete. If I ask how much throughput a database can support, the useful answer may begin with a discussion of maintenance that I had not included. A test with maintenance paused can tell us something about foreground throughput, but if the customer needs both activities to happen, we have another experiment to discuss. Similarly, an AI team may have measured how many requests meet a latency target without yet evaluating answer quality. I would rather learn that distinction from the team than discover, after making a commitment, that I had combined two results that never held at the same time.
When the answer has a sponsor
| Claim | Question that makes it inspectable |
|---|---|
| “Efficiency improved.” | Is the useful outcome unchanged? Were workload mix, latency, quality, durability, maintenance, and recovery conditions comparable? |
| “This is the demand forecast.” | Is it a forecast, sales ambition, funding request, or commitment? Who supplied the assumptions, and are contrary scenarios preserved? |
| “We saved money.” | Was cost removed, deferred, reallocated, or avoided in a hypothetical future? Are migration, parallel running, remaining commitments, and shared costs included? |