AI Workflow Readiness Scorecard

Dimension 1. Workflow suitability

Whether this particular piece of work is a sound candidate at all. This dimension fails more pilots than the technology does, and it fails them before anyone writes code.

1. The workflow has a defined trigger and a defined end state.

   Score 0

   The workflow is described as an area of activity rather than a process. Nobody can say what starts it.

   Score 1

   The trigger is clear but the end state is contested, or the process branches into several endings that have not been enumerated.

   Score 2

   A single sentence names what starts the work and what condition means it is finished, and the room agrees with it.

2. Volume and frequency justify the build.

   Score 0

   The workflow runs occasionally or unpredictably, and no one has counted it.

   Score 1

   Volume is known and moderate. The case rests on quality or turnaround rather than on quantity, and that case has not been written down.

   Score 2

   Volume and frequency are measured, and the measurement is written down. Whether the case rests on volume, on turnaround, or on consistency, it is stated in numbers somebody stands behind.

3. The work is language, document or data shaped rather than dependent on actions the solution cannot take.

   Score 0

   Completing the work requires physical action, or changes in systems that offer no interface, or judgment that depends on information nobody has recorded anywhere.

   Score 1

   The core of the work is language or document based, but one or more steps depend on a system with no usable interface, and no plan exists for those steps.

   Score 2

   Every step either operates on text, documents or structured data, or has a documented interface. Steps that fall outside are identified and deliberately left with a person.

4. The current process is documented well enough that a new joiner could follow it.

   Score 0

   The process lives in the heads of the people who do it. Two practitioners would describe it differently.

   Score 1

   Documentation exists but is out of date, partial, or describes the intended process rather than the one actually performed.

   Score 2

   A current written description exists, and someone who does the work has confirmed it matches what they actually do, including the exceptions.

5. A good output can be described in advance and recognized when produced.

   Score 0

   Quality is a matter of taste, or is judged only by whether a downstream person complains.

   Score 1

   The people who do the work can recognize a good output reliably, but the criteria have not been written down or agreed between them.

   Score 2

   Written criteria exist, and a set of past examples has been marked good or not good against them, so the criteria have been tested on real cases.

If this dimension scores low

• If W1 or W3 scored 0, stop here. The workflow as described is not the unit of work to automate. Split it until each piece has one trigger, one ending and a usable interface, then score the pieces.
• If W4 scored 0 or 1, run a process walkthrough with two practitioners before anything else. It is the cheapest work in this entire list and it changes the design of everything that follows.
• If W5 scored 0, build a set of twenty past cases labeled good or not good. Without it there is no way to tell whether the built solution works, and no way to tell when it stops working.