Salem J. AbabnehHow I workMethod 02

Deciding where AI actually belongs

Most AI strategies are a list of things that sound possible. The useful question is narrower: given this organisation's data, regulation and appetite for change, what should it build first, and what should it not build at all.

Use cases scored on impact against feasibility Every candidate use case is placed by how much value it creates and how ready the organisation is to build it. The shaded band in the upper right is what gets built first. Build first Feasibility → (data, tech, regulation, change effort) Impact → Service triageDocument processingCase summarisationPredictive demandAutonomous agentsKnowledge search

The method

  1. Start from the value chain, not from the technology

    Walk the actual chain of work end to end and ask where judgement, waiting or volume sits. Those are the only places AI has anything to offer. A use-case list assembled from vendor decks describes the vendor, not the organisation.

  2. Score impact and feasibility separately, and never blend them

    Impact is what the organisation gains. Feasibility folds together four things that each kill a use case on their own: whether the data exists and may lawfully be used, whether the technique is proven for this task, what the regulatory exposure is, and how much change the operation can absorb. Collapsing them into a single score hides the reason a use case is hard, which is the only part that tells you what to do about it.

  3. Be honest about data readiness

    This is where most AI programmes actually die. A high-impact use case sitting on data that does not exist, or cannot be used lawfully, is not a use case yet.

  4. Separate agents from prediction from automation

    They have different risk profiles, different governance, and different failure modes. Putting them in one bucket called AI guarantees the governance fits none of them.

  5. Build the first one properly, then use it as the argument

    One working system in production changes an organisation's willingness to fund the next ten more than any amount of prioritisation analysis.

Where this comes from

An AI use-case map across a full government value chain, a first-of-its-kind exercise that shaped the AI agenda across 20 entities, and appointment to two nationwide ministerial AI committees. On the delivery side: a retrieval-grounded LLM support assistant over a bank's own operational data, wired into core-banking APIs.

The other methods

Changing how a large organization works? Let's talk.

[email protected] LinkedIn Download CV