Integration & data readiness
The unglamorous majority of every AI initiative: systems integration, data quality, and identity — and why the data-readiness line decides your roadmap's actual speed.
Here is the sentence every experienced AI program leader eventually says to a board: 'The AI was the easy part.' The model classified beautifully in week one; the next five months went to getting inputs to it and outputs from it reliably, and to discovering what your data actually looks like. Budgeting Module 2's integration multiplier is how you priced this; this lesson is how you manage it.
The three unglamorous fronts
- Systems integration — the AI must sit inside the workflow, not beside it: reading from the TMS, writing to the ticketing system, appearing in the tool the dispatcher already has open. Every 'beside' deployment (a separate tab, a copy-paste step) taxes adoption daily and quietly dies. Count the touched systems per initiative — each is an interface to build, a team to coordinate, a failure mode to design for. This number, more than model choice, sets the timeline.
- Data quality — the inventory's green/yellow/red was an estimate; integration is where you find out. Yellow data (exists, messy) means duplicate customers, three date formats, free-text fields where enums should be, and a 'reason code' column that's 40% other. The initiative inherits whatever the data carries — AI on unresolved duplicates confidently serves the wrong customer's history. Fixes range from a cleanup sprint to a real data-foundations effort (dedicated course, dedicated tooling — entity resolution and pipeline quality gates are product categories for a reason).
- Identity & access — who is the AI acting as? What can it read, and on whose authority? The service account with read-everything permissions is the convenient choice and the one your security lead (and Module 4's charter) will rightly veto. Least-privilege takes a week now or an incident later; it also produces the audit trail governance requires. Decide identity per initiative, in writing, before integration starts.
Data readiness as roadmap strategy
Zoom out and the data-readiness column of your inventory is a shadow roadmap: every red flag is a future initiative blocked, every yellow a tax on timelines. The strategic response isn't per-initiative heroics — it's a deliberate data-foundations investment (warehouse hygiene, resolved entities, quality monitoring, governed metric definitions) charged partly to the portfolio and defended as what it is: the substrate that makes each subsequent initiative cheaper and faster. Alder's version: the TMS appointment-field fix (option #7) plus a customer-entity cleanup, together unblocking three red-flagged candidates for next year. Boards fund substrate when it's framed as pipeline; they starve it when it's framed as IT hygiene. Frame it as pipeline, because it is.
Before funding any initiative, make the owner answer in writing: 'Name every system this touches, who owns each, and what happens when each is down.' Initiatives that can't answer aren't ready to fund — they're ready to scope. Asking this in the portfolio review, cheerfully and every time, will do more for your delivery dates than any vendor selection.