← All field notes

Engagement model · Oct 2026 · 9 min read

The First 90 Days

What actually happens before dashboards, AI, and automation become credible.

02

The misconception

Discovery first, build later is too slow. Build first is too blind.

The first ninety days are often described as discovery followed by delivery. That split sounds disciplined, but it can create a false sequence: one team writes documents, another team later discovers production reality, and the project spends month three reconciling the two.

The opposite failure is just as common. A team starts building screens immediately because stakeholders need to see momentum. The interface arrives quickly, but every serious question becomes harder: what does this field mean, which system wins, how are duplicates handled, who owns the exception, what happens when the source changes?

The useful first ninety days combine modelling and delivery. The model is tested by building a narrow working slice. The slice is constrained by the model. Each improves the other.

03

Month one

Make the operation legible.

The first month is about nouns and evidence. We identify the objects the business actually runs on: party, place, asset, order, event, document, measure, rule, user, workflow, exception. Then we inspect where those objects appear, where they disagree, and where people have built local workarounds.

This is not abstract architecture. It means looking at real feeds, spreadsheets, exports, inboxes, photographs, call logs, documents, and legacy tables. It means asking why a field exists, who changes it, whether anyone trusts it, and what happens when it is wrong.

The output is an initial Mantle: a governed model with enough specificity to build against. It should be readable by engineers and business owners. If only one of those groups can understand it, it is not finished.

04

Months two and three

Prove one slice, then widen under pressure.

The second month should produce a working slice that is narrow enough to finish and real enough to hurt. One region, one exception class, one workflow, one document type, one decision loop. The point is not coverage. The point is to expose the assumptions that a presentation would miss.

By the third month, the slice widens. More sources. More users. More operational pressure. This is where brittle assumptions show up: a source that changes shape, an entity match that needs human review, a rule with a hidden exception, a process owner who has a different definition from the dashboard owner.

The best outcome is not a swollen backlog. It is a clearer one. By day ninety, you should know what the core model is, which decisions can be automated, which require review, which data sources are reliable, which are politically sensitive, what the client team will own, and what should deliberately not be built yet.

05

Use this

What you should have by day 90

01

A named model of the operation

The core objects, relationships, rules, and exceptions should be documented clearly enough to build and challenge.

02

A working slice on real data

The system should have survived production-shaped inputs, not just sample data.

03

A visible exception path

Every real operation has ambiguity. The question is whether the system exposes it or hides it.

04

A handover path

Ownership should be designed from the beginning, not negotiated at the end.

05

A smaller roadmap

The first phase should remove bad ideas, defer premature ones, and sharpen the few that matter.

Talk it through

Does your system look like this?

Fifteen minutes with an engineer, not sales. Bring the architecture, leave with a straight read on it.

Book a 15-min architecture call