What to automate first: a scoring framework

Most owners stall in the same place.

They know AI can help. They’ve seen what it does for workflow automation. They have a list of ten things that could probably be automated. And then they freeze, because picking wrong wastes two months and makes the next project harder to justify.

The answer isn’t to deliberate longer. It’s to score the candidates against criteria that actually predict success.

Here’s the scoring matrix we run at the start of every engagement. You can use it in forty-five minutes on your own shortlist.

Why most AI project selection fails

The most common mistake isn’t picking a bad project. It’s picking a project for the wrong reason.

We wrote the details in what 95% of AI pilots have in common, but the short version is this: projects get picked for how they’ll look in a demo, not for how much time they’ll return to the team. The workflows that actually pay off are quieter and more measurable than the ones that get boardroom attention.

The scoring matrix is the corrective. It weights every criterion toward durability and measurability, not impressiveness.

The scoring matrix

Score each workflow candidate on five criteria. Use a 1-5 scale. Sum the scores. Rank.

Criterion	1 (weak)	3 (moderate)	5 (strong)
Weekly time cost	Under 30 min/week total	1-3 hours/week	4+ hours/week
Input predictability	Input varies unpredictably	Input varies but has patterns	Input is consistent and structured
Output verifiability	Hard to tell if output is correct	Spot-checking works most of the time	Output is objectively right or wrong
Failure and privacy cost	Failure is public or damages a client relationship	Failure is internal but recovery is slow	Failure is internal, caught fast, cheap to fix
CFO-explainability	ROI is indirect or hard to quantify	ROI requires some explanation	ROI is immediate and easy to count in hours or dollars

Total possible: 25 points.

• 20-25: Strong first-project candidate. Build it.
• 15-19: Viable. May need scoping work before the build starts.
• 10-14: Not ready. Fix the low-scoring criteria first or move it down the list.
• Below 10: Skip for now. The workflow isn’t automation-ready yet.

A worked example

Take a real workflow pattern we see at almost every professional services firm: recurring client status reports.

A team member pulls data from three systems every Monday morning, normalizes it, and drops it into a template. It takes around two and a half hours, every week, without fail.

Criterion	Score	Reasoning
Weekly time cost	4	2.5 hours/week, one named person
Input predictability	5	Same three systems, same fields, every Monday
Output verifiability	5	The report is either accurate or it isn’t
Failure and privacy cost	5	Failure is internal, one person is alerted, recovery takes minutes
CFO-explainability	5	2.5 hours/week times 52 weeks is 130 hours/year. Easy to count.

Total: 24 out of 25. That’s a first project. The build is well-defined, the savings are easy to measure, and the failure surface stays inside the building.

Now compare it to a customer-facing intake chatbot that the same team has been thinking about.

Criterion	Score	Reasoning
Weekly time cost	3	Hard to measure; intake volume varies
Input predictability	2	Customer messages are unpredictable in form and content
Output verifiability	2	Hard to tell if the bot handled an edge case correctly without reading transcripts
Failure and privacy cost	1	Failure is in front of the customer, in writing
CFO-explainability	2	”Faster intake” is hard to count before and after

Total: 10 out of 25. Not a first project. Maybe a fourth. That low score on failure and privacy cost alone should move it down the list.

This is the same logic we walked through in why your first AI project shouldn’t be a chatbot. The matrix just puts numbers to it so you can rank your options instead of debating them.

How to build your candidate list

The matrix is only useful if you’re scoring the right candidates. Here’s how to build the list.

Pull last week’s calendars. Yours and the two or three people whose time matters most to the operation. Highlight every recurring block of work. Status reports, posting routines, intake reviews, data exports, follow-up sequences. Memory is generous about how long these take. Calendar is honest.

Write the input and output for each one. One sentence per. “Input: a new inbound inquiry email. Output: a classified and routed summary with urgency flag.” If you can’t finish both sentences in one sitting, the workflow isn’t ready to score. It’s ready to scope.

Take the top five candidates into the matrix. You don’t need to score everything. The highest-time-cost, most-recurring blocks are your candidates. Score those five. The ranking usually becomes obvious by the third or fourth row.

If you want a structured version of this process, the 45-minute audit walks through it step by step and produces a ranked list with written success criteria for each candidate.

What to do after the matrix

Your top-ranked workflow is your first project. Before you build anything, write down four things:

1. The input, in one sentence.
2. The output, in one sentence.
3. The hours-per-week savings at stake.
4. The name of the person who owns it when it breaks.

If you can write all four, you’re ready to build. If you can’t, the project needs more scoping. Tighten the input and output sentences until they’re unambiguous. That work is not a delay. It’s the build, done early and cheaply.

A team ready to start can use the process mapper to go from candidate to documented workflow before the build begins. That documentation becomes the spec the build runs against.

The first project sets the tone

Owners who pick a 20+ project for their first automation build something that returns time in the first week, holds up unattended, and is easy to explain at the next finance review. That project earns the credibility to take on harder ones.

Owners who pick a low-scoring project first often spend two months building something that nobody can tell is working. That project makes the next one harder to fund.

The matrix doesn’t guarantee a successful build. It does guarantee that you’re starting in the right place.

---

Free 30 minutes. Bring your candidate list and your matrix scores to a blueprint call with us. You walk away with a ranked shortlist, a written success definition for your top project, and a realistic cost and timeline. No deck. No pitch. Book the call.