Where Should AI Fit in Your Business Strategy?
A practical guide to choosing AI opportunities from business outcomes, evidence and constraints.
A customer expects a delivery on Thursday. The stock exists, but it has already been allocated to another order. The warehouse cannot dispatch until Friday, and the customer finds out too late to change their plans.
This fictional distributor could ask for an AI delay predictor. First, its leaders need to ask why the Thursday promise was made and when anyone could have spotted the conflict.
The practical rule: Start with the promise you need to keep. Trace where it breaks. Then compare AI with simpler fixes.
Compare the options against the same order cases. A catalogue of ideas is not an approved roadmap.
Choose the outcome and the area to investigate
The distributor could also pursue faster quotes or a broader product range. Reliable promises may deserve attention first if missed dates are harming customers, creating rework or weakening trust. That is a hypothesis to check, not a conclusion the example proves.
Follow one customer journey: the sales commitment, stock allocation, dispatch and the message to the customer. The business needs to manage stock availability, make dependable promises and resolve exceptions. These are capabilities — things it must be able to do across teams, regardless of which software it uses.
Start with evidence: How often do promised dates change? When is the allocation known? Who notices an exception, and when does the customer hear? The current baseline is unknown. Investigating those questions is more useful than asking each department for an unrelated AI wish list.
Compare changes against the same cases
Each response in the diagram solves a different part of the problem:
- Process and ownership: Require an availability check before confirming a date and name who handles exceptions. Check whether this prevents the failure without new technology.
- Rule-based alert: Flag a conflict when the allocation is already known. This needs timely, reliable data and a person who acts on the alert.
- Prediction: Test whether historical patterns reveal risks before a rule can. This needs enough relevant cases and proof that an earlier warning changes a decision.
- Generative assistance: Draft a customer update from verified order information for a person to review. This may improve communication, but it cannot make the original promise accurate.
Set a few principles before choosing: do not call an uncertain date confirmed; tell customers promptly when a commitment changes; give exceptions an owner; use information approved for the task. These guide a checklist as much as an AI system. The team may revise them when real orders expose unclear responsibilities.
Record promising options in an opportunity catalogue with the outcome, evidence, dependencies and open questions for each. Inclusion means worth examining, not approved to build. A polished demo should not outrank a straightforward fix to the real constraint.
Turn the investigation into a decision
An operations lead could review a small sample of changed orders with sales and fulfilment. For each one: when was stock allocated, when did each team know, who could act, and when was the customer told? The same cases can test whether a process check, a rule or an AI-assisted option would have helped.
The result should be a short recommendation: the preferred change, supporting evidence, prerequisites, an accountable owner and the remaining test before wider commitment. If evidence shows that faster quoting or another outcome matters more, leaders can change the focus. The point is to make a better choice, not to defend the first idea.
Your one-page AI focus brief
Use these prompts for one bounded area of your business:
- Outcome: What result matters? What is the baseline, or what remains unknown?
- Focus: Which customer journey or business area has the strongest bearing on it?
- Capabilities: What must the business do reliably across teams?
- Principles: Which commitments, decision rights and information boundaries apply?
- Options: Which process, rule-based, existing-tool and AI-assisted changes merit comparison?
- Dependencies: What data, integration, ownership and adoption work would each need?
- Next decision: Who will investigate, what evidence will they return, and what choice will it inform?
The distributor is fictional. This brief is my synthesis of the sources below, rather than a prescribed framework.
Sources and inspiration
- Richard Rumelt’s strategy kernel, discussed in The Perils of Bad Strategy.
- Eric Lamarre, Kate Smaje and Rodney Zemmel on choosing a transformation bite size.
- Business Architecture Guild, Business Architecture Metamodel, for the distinction between capabilities, value delivery and initiatives.
- Fortuna One, applied-AI session: ideas about recurring decisions, bottlenecks and business context, drawn from the session transcript.