Choose a workflow before choosing an AI tool

An organization can usually name several places where people search for information, prepare drafts, sort material, or repeat an administrative step. Turning that list into an AI project requires a choice. The most visible task is not automatically the one with the clearest purpose or the most useful outcome.

Describe each candidate as a workflow rather than a tool category. State what starts it, who participates, which information is used, and what result must be passed onward. That description creates a common basis for comparison. It also helps distinguish a recurring problem from a single frustrating incident that may have a simpler explanation.

For each candidate, identify the information required and where it comes from. Explain who maintains it, how complete it is, and whether the people using the proposed result are allowed to access it. A task can sound straightforward while depending on information that is scattered or not consistently recorded.

Include the knowledge people contribute outside the written process. They may interpret a customer’s request, recognize an exception, or know which record is current. A proposed workflow should account for that contribution. Otherwise the comparison assumes that the available information contains an answer that actually resides in someone’s judgment or experience.

Consider what a person might do because of the result. An internal draft that receives careful review has a different role from an answer that directly changes a business record. The significance of an error should affect both the proposed system role and the evaluation effort.

Describe the intended boundary in ordinary language. The system might help locate a source, prepare a draft, or organize a queue while a person retains a decision. That boundary makes the comparison more useful than asking whether a whole job can be automated. It shows what assistance would be valuable and what responsibility would remain.

A tool that demonstrates well may still need information preparation, connections to existing systems, permissions, training, and continuing review. List those dependencies for each candidate. The purpose is to understand what would need to happen around the technical capability for it to become useful.

Review effort belongs in the same account. If an output takes as long to check as the original task took to complete, the proposed improvement needs closer examination. Consider how the result reaches the next person and whether that person can act on it without repeating the same investigation.

Use a small comparison table with the same questions for every candidate: intended outcome, information readiness, review needs, operational dependencies, and responsible owner. Avoid combining these into an apparently precise score unless the team understands what the score means and which tradeoffs it hides.

A useful next step may be focused research, a clearer process description, or the design of a system. The comparison should make that choice easier. It should also make it possible to conclude that a particular task is not currently a useful place to apply AI, without treating that conclusion as a failure of the wider initiative.

Consider two candidate tasks: locating an approved document and deciding whether an exception should be granted. Both involve information, but the second also involves authority. Comparing them exposes a useful boundary: AI may assist with preparation while the organization retains the decision. The distinction should be visible before a tool is selected.

Once a workflow is chosen, preserve the reasoning behind it. Write down the present process, desired improvement, representative examples, and questions still unresolved. Identify who can judge a useful result and who would own the system if it becomes part of everyday operations.

Halsted, Haines & Co. brings together consulting, advisory, research, and AI development. A clear workflow choice connects those disciplines: it gives research a question, strategy a decision, and development a purpose. That continuity is more useful than beginning with a product choice and then searching for a business need to justify it.