Insights
Written from delivery rather than from a content calendar. No client is named and no project figure appears here that is not already published elsewhere on this site — these are patterns, not case studies.
Migrations rarely fail on the code. They fail on data nobody profiled, mapping rules nobody signed, tests run on samples and rollbacks that exist only on paper. The ten stages and what each has to prove, followed through a worked CRM example, with ten scenarios where it goes off-plan and five questions to ask of any plan.
Marketing reports rarely fail on the visuals. They fail on definitions: which channel a lead belongs to, which attribution rule applies, and which month the revenue lands in. How I would structure the model so each is answered once.
Where to put the review gate, how to make review cheaper than doing the task by hand, and why an acceptance rate you actually measure is the only honest way to report on an AI workflow.
Standardising a metric across units is a negotiation, not a modelling exercise. How to run the definition workshop, where to allow local variants, and what to put in the KPI dictionary so the agreement survives.
The dashboard is usually not the problem. It gets built against available fields instead of a named decision, ships without retiring the spreadsheet it replaced, and has no owner when a number looks wrong.
Integration projects rarely fail on the connector. They fail on identity, on late-arriving corrections, and on the fact that nobody agreed what a record means. Six failure modes, and the design decision that prevents each.
Dat Tran writes these between engagements. They are practice notes, not a publication schedule — new ones appear when there is something worth writing down.
These notes describe patterns. Applying one to a specific estate is what an engagement is for.