How to Choose an AI Workflow Worth Implementing

Once you have a list of things AI could help with, the hard part starts. Nearly every business we speak to has more candidate ideas than capacity, and the ideas that sound most impressive are frequently the worst places to start.

Here is the scoring approach we use, and you can apply it with a spreadsheet and an honest hour.

Score four things, not one

For each candidate workflow, score four dimensions from 1 to 5.

Value

How much does solving this actually help? Use whichever unit you can defend:

• Hours per week currently spent
• Delay it removes from a client-facing process
• Revenue at risk when it goes wrong, such as unanswered enquiries

Be careful with "it would be nice". Nice is not a unit.

Feasibility

Can this realistically be built with the information and tools you already have? Feasibility drops fast when:

• The inputs are inconsistent or handwritten
• The information lives only in someone's head
• A key system has no way to connect to anything else
• The rules have exceptions nobody can state clearly

Risk

What is the consequence of a wrong output, and how quickly would you notice? A wrong internal summary is a nuisance. A wrong figure sent to a client is a problem. A wrong submission to a regulator or a bank is something else entirely.

Score risk so that 5 means low risk, which keeps the arithmetic in the next step simple.

Effort

How much work is it to build, test and document? Consider your own effort too: interviews, examples, decisions, and the time your team spends learning the new way.

Score effort so that 5 means low effort.

Add them up, then read the shape

Add the four scores for a total out of 20. But do not just take the highest number. Look at the shape.

• High value, high feasibility, low risk, low effort: start here. This is the first project.
• High value, low feasibility: usually a data or process problem wearing an AI costume. Fix the underlying issue first.
• High value, high risk: worth doing, later, with a firm human approval step and a proper test plan.
• Low value, low effort: tempting because it is easy. Skip it, unless it is genuinely annoying enough to affect morale.
• Low value, high effort: write it down and let it go.

Apply two disqualifiers

Regardless of score, two things should stop a candidate from being first.

No owner. If nobody in the business will own it after handover, it will drift and quietly stop being used. Find an owner or choose something else.

No agreed process. If two people describe the current process differently, you are not ready to automate it. Document the process first. That work is not wasted: it is most of the value, and it makes the eventual build far cheaper.

Define success before you build

Write one sentence you will be able to test in six weeks. For example: "Quotes are drafted within one working day of the enquiry, without me writing them from scratch."

That sentence does two jobs. It tells whoever builds the workflow what "done" means, and it stops the project from being judged on how clever it feels.

Choose the smallest useful version

Almost every workflow has a smaller version that delivers most of the benefit. Instead of a system that handles all enquiry types, build one that handles the two most common. Instead of a report covering everything, build the section people actually read.

The smaller version reaches real use sooner, and real use tells you things no planning session will.

Then re-score

After the first project has run for a few weeks, score the remaining candidates again. Your feasibility estimates will be better, your team's appetite will be clearer, and some items will have moved up or off the list entirely. Prioritisation is not a one-off exercise.

If you would rather not do this alone

This is precisely what an AI Opportunity Audit produces: a scored, prioritised list with a recommended first project, the tools involved, and the risk considerations written down. You can also start with the three-minute AI Opportunity Score for an initial indication, or read 7 repetitive tasks small businesses can automate for candidate ideas to score.