Alexander Neumann All articles
Technology Strategy

Diagnose Before You Prescribe: A 48-Hour Method for Locating the Actual Problem

Alexander Neumann
Diagnose Before You Prescribe: A 48-Hour Method for Locating the Actual Problem

Photo: PEO ACWA, CC BY 2.0, via Wikimedia Commons

A client arrives with a diagnosis already formed. The platform is slow, the integration is failing, the operating model is misaligned. Occasionally the client is right. More often the stated problem is the most visible consequence of a cause located elsewhere in the system, and an engagement that accepts the brief at face value will produce an elegant solution to a question nobody needed answered.

The diagnostic phase is therefore not a courtesy before the real work. It is the point at which the value of the engagement is largely determined. The framework below is designed to be completed in roughly forty-eight working hours, and to produce a problem statement defensible enough to build a proposal on.

Read the System Before You Enter the Room

The first eight hours are spent with artifacts rather than people. Request everything that has already been written: prior strategy decks, architecture diagrams, incident logs, support tickets, the last two audit reports, the vendor contracts that shaped the current stack. Read them in sequence and note contradictions rather than conclusions.

Three scanning passes are usually enough. The first pass establishes the official story, which is what leadership believes. The second pass looks for the operational record, which is what the systems and tickets say. The third pass compares the two. The gap between the official story and the operational record is the most reliable signal available in the first day, and it is nearly free to obtain.

Where the technology estate is instrumented, pull raw telemetry rather than dashboards. A dashboard encodes someone's earlier judgment about what matters. Event logs, queue depths, error rates by endpoint, and cycle times from workflow systems tend to be more candid. A frequent finding at this stage is that the reported problem occurs in a narrow, identifiable slice of transactions, which changes the shape of the eventual solution entirely.

Interviews That Test Rather Than Confirm

Hours eight through twenty-four belong to structured conversations. Six to eight interviews, thirty minutes each, conducted with people at different altitudes: an executive sponsor, two line managers, two practitioners closest to the work, and one person known to be skeptical. The skeptical interview is not optional. It is the cheapest available check against collective self-deception.

Ask for sequences and stories rather than opinions. A question such as "walk me through the last time this happened, from the moment you noticed to the moment it was resolved" yields far more than "what do you think the problem is." The latter invites theory; the former produces evidence.

Three prompts do most of the work. Where does the work wait? Where does it get re-entered by hand? Who has to approve, and what do they actually check? Answers to these tend to cluster, and clustering across independent interviews is the closest thing to confirmation available at this speed.

Hypothesis Testing With the Cheapest Evidence

By the second day, the team should hold three competing explanations, written as falsifiable statements. For each, name in advance what evidence would disprove it, then go collect that evidence first. This inversion is deliberate: confirmation is cheap and misleading, while disconfirmation is decisive.

Suppose the client believes a reporting platform is underperforming. Hypothesis one is capacity. Hypothesis two is data quality upstream. Hypothesis three is that the reports are accurate and the business process they describe is the actual constraint. The disconfirming test for hypothesis one is a load profile at peak, which takes minutes if monitoring exists. If capacity is adequate, an entire workstream has been eliminated before lunch.

Where data is contested, prefer a small direct sample over a large indirect one. Twenty manually traced transactions will usually settle a question that a month of dashboard debate cannot.

The One-Page Problem Statement

Forty-eight hours should produce a single page with five components: the observed condition stated without interpretation; the mechanism that most plausibly produces it; the consequence in financial or operational terms; the strength of the evidence behind each claim; and the specific evidence that would change the conclusion. The last component matters as much as the others, because it gives the client a way to disagree productively.

This page becomes the foundation of the proposal. Scope is then limited to what the diagnosis supports, which typically produces a smaller and more credible engagement than the original brief imagined.

Where the Method Breaks Down

The framework fails in three recognizable conditions. It fails when source data is unreliable, in which case the diagnostic must include an instrumented measurement phase rather than pretending evidence exists. It fails when the sponsor is a cause of the problem, since interviews conducted under that sponsor's auspices will be shaped accordingly. And it fails in politically charged settings where the diagnosis is contested before it is formed, which is a negotiation problem rather than an analytical one.

Recognizing these limits is part of the discipline. The purpose of a two-day diagnostic is not omniscience. It is to ensure that the first prescription written is aimed at a condition that actually exists, supported by evidence the client can inspect, and narrow enough to be governed.

All Articles

Related Articles

Demos Are Not Decisions: Auditing Your AI Portfolio Before the Next Pilot Wave

Demos Are Not Decisions: Auditing Your AI Portfolio Before the Next Pilot Wave

Fee at Risk: Structuring Outcome-Based Compensation Both Sides Can Defend

Fee at Risk: Structuring Outcome-Based Compensation Both Sides Can Defend

Saying No on Purpose: A Fit Test for Prospective Consulting Clients

Saying No on Purpose: A Fit Test for Prospective Consulting Clients