Week One Decides Everything

The phrase “90-day plan” suggests 90 days of forward momentum — building, deploying, iterating. In practice, the first week of a well-run AI engagement looks nothing like building. It looks like a calendar full of one-hour conversations with people at different levels of the organization, most of whom have never spoken to an outside consultant before, and some of whom have strong opinions about what’s broken and why.

That week is not a preamble. It’s the work. A Nebraska distribution company with 180 employees and three operational divisions has problems that no outsider can identify from a demo call and a signed contract. The people who know where the freight reconciliation breaks, where the dispatch process produces the most callbacks, where data entry is slowest and most error-prone — those people are on the floor, not in the kickoff meeting. Week one is about reaching them before the project direction hardens around someone’s best guess.

Week-one principles (talk to the people who do the work first)

The governing principle is simple: talk to the operators before the managers, and the managers before the executives. Not because managers and executives have less useful information, but because the sequence matters for what you hear.

An executive describes the business in terms of strategy and outcomes. That’s useful context, not a source of operational detail. A manager describes the business in terms of problems she’s trying to solve and people she’s trying to manage. That’s closer. The operator — the dispatcher, the credit analyst, the receiving coordinator — describes the business in terms of what she does between 8 a.m. and 5 p.m. and what makes that harder than it should be. That’s where the initiative list comes from.

Going in the other direction — executive first, then manager, then operator — shapes every subsequent conversation. If the CEO has told you the dispatch process is the priority, the dispatcher is going to feel like she’s being evaluated against a predetermined conclusion rather than heard. The conversations close down instead of opening up.

A second principle: don’t bring a solution. Consultants who arrive in week one already knowing what they’re going to build have stopped listening before they started. The discovery process is not a box to check — it’s the primary means of learning whether the engagement is aimed at the right problem.

The interview structure

Each conversation runs 45 to 60 minutes. The questions stay in one lane: describe your actual workday, not your ideal one. Where does the work back up? What takes longer than it should, and why? Where do you spend time on something that feels like it shouldn’t require a person?

The last question — where does it feel like it shouldn’t require a person — is the most productive one in this context. It surfaces the workflows that are high-volume, repetitive, and rule-based: the manual entry that bridges two systems that don’t talk, the report that someone assembles by pulling numbers from three different exports, the approval queue that routes through email because no one has built a better mechanism.

Note what the interviewee does and doesn’t mention. The absence of a complaint about a particular process is information. If no one on the billing team mentions the reconciliation process, either it works fine or it’s so broken they’ve stopped expecting it to be fixed. Both interpretations matter.

After each session, write a 200-word summary: what this person does, what the three most concrete friction points were, and what the one thing they mentioned that surprised you. The surprise note is the most valuable part of the summary. Surprises are where the assumptions being tested turned out to be wrong.

The deliverable at the end of week two

By the end of the second week, the output is a named-initiative list. Not a roadmap, not a strategy deck, not a set of recommendations. A list of four to eight specific things the company could build, each with: a name, a one-sentence description, an estimate of who is affected and how often, and a difficulty rating (straightforward, moderate, complex).

The naming matters more than it sounds. An initiative with a name — “freight reconciliation draft,” “dispatch callback reduction,” “credit limit alert” — becomes a thing people can reference in conversation and argue about specifically. An initiative described in vague terms stays vague and gets argued about vaguely.

The list goes back to the operators who informed it. They should be able to recognize their own problems in it. If they can’t, either the interviews missed something or the synthesis introduced distortion. Both problems are fixable in week two; neither is fixable in week eight.

What we ignore in week one (vendor demos, tool selection, architecture)

Vendor demos, tool evaluations, platform comparisons, architecture decisions — all of this waits until the named-initiative list exists and has been validated. The sequence is not negotiable.

The reason is basic: tool selection before problem identification produces a solution looking for a problem. If you arrive in week one with a preferred AI platform or a model architecture you want to try, you will find ways to apply it regardless of fit. The people on the floor will sense that their input is being translated into a predetermined direction. The trust that week-one interviews depend on evaporates quickly once operators feel like props in a vendor selection process.

The right time to talk about platforms and tools is after the initiative list is agreed upon. At that point, the conversation is grounded: what does this specific workflow require, what integration points exist, what does the company already have in its technology stack that should be used rather than replaced? Those questions have concrete answers once you know what you’re building.

What it looks like on day one for a 200-person firm

Day one at a 200-person logistics firm in central Nebraska. The engagement starts at 8 a.m. with a 30-minute kickoff with the CEO and the VP of operations — not a long meeting, just enough to confirm the firm’s understanding of what week one will produce and who the team will be talking to.

The calendar for weeks one and two has 12 conversations scheduled: two dispatchers, two drivers, the freight billing coordinator, the two credit analysts, the receiving manager, the VP of operations (again, now for depth rather than kickoff context), the IT manager, the HR director, and one executive check-in at the end of week two.

By end of day one, the first three conversations are done. The summary notes are written. One surprise has already surfaced: the freight billing coordinator has been maintaining a personal spreadsheet for 18 months that bridges a gap between the TMS and the accounting system. Nobody in management knows the spreadsheet exists. That spreadsheet is probably the most valuable artifact in the building for the purposes of this engagement.

That’s day one. Week one looks like 12 more conversations. Week two looks like a named-initiative list. Week three is when the building starts.

See how Blue Sage structures the full engagement in how we work.