The workflows worth automating first are frequent, rule-governed, measured, owned by a named person, and cheap to get wrong. Score each candidate from 1 to 5 on those five criteria.
A total of 20 or more out of 25 marks a first-project candidate. A score of 14 to 19 means fix the process first, and 13 or below stays manual.
This guide is for operations leaders at B2B firms of roughly 50 to 500 people: COOs, heads of operations, RevOps, and support leads with more automation ideas than capacity. It gives you the rubric, the thresholds, a worked example, and the baseline to record before anyone builds.
Key Takeaways
- Selection, not supply, is the constraint: most teams have plenty of repetitive work, and the risk is picking the wrong first candidate.
- Score each candidate 1 to 5 on frequency, rule clarity, baseline and ownership, data readiness, and risk.
- A total of 20 to 25 means automate now, 14 to 19 means fix the process first, and 13 or below means leave it manual.
- No named owner or no measured baseline means not ready, whatever the total.
- Measure the baseline before you build, or you won't be able to show what the automation changed.
Why the First Workflow You Pick Decides the Project
The first workflow you automate sets the budget, the internal credibility, and the template for everything after it. Pick badly and the whole program looks like a failed experiment.
Gartner predicts that more than 40% of agentic AI projects will be canceled by the end of 2027, due to escalating costs, unclear business value, or inadequate risk controls (Source: Gartner, reported by MarTech).
Gartner describes the root cause as early-stage experiments or proofs of concept that are driven by hype and often misapplied (Source: Gartner, reported by MarTech). That's a selection problem. A project aimed at the wrong workflow can't show value, however good the build.
The supply of candidates isn't the constraint. Asana's Anatomy of Work Index, a survey of more than 10,000 knowledge workers, found that 60% of a person's time at work goes to "work about work" (Source: Asana, Anatomy of Work Index).
The same research counts 209 hours a year lost to duplicative work, 352 hours a year spent talking about work, and 103 hours a year in unnecessary meetings (Source: Asana, Anatomy of Work Index).
And 88% of those knowledge workers agree that time-sensitive projects and large initiatives have fallen behind or through the cracks (Source: Asana, Anatomy of Work Index).
There's plenty of repetitive work to automate. The hard part is choosing the one workflow that will prove the approach, and the rest of this guide is about that choice.
Where Mid-Sized Firms Stand on Automation
The US Census Bureau's Business Trends and Outlook Survey shows AI adoption rising with firm size. In data collected from December 14, 2025 to May 3, 2026, 37% of firms with 250 or more employees used AI in business operations, against 32% of firms with 100 to 249 employees (Source: US Census Bureau, Business Trends and Outlook Survey).
Among firms with fewer than 20 employees, use stayed under 20%, and national use hovered between 17% and 20% (Source: US Census Bureau, Business Trends and Outlook Survey). Between 20% and 23% of businesses expected to use AI within the next six months (Source: US Census Bureau, Business Trends and Outlook Survey).
The trend matters as much as the level. Use rose among firms with 20 or more employees and did not change significantly among firms with fewer than 20 (Source: US Census Bureau, Business Trends and Outlook Survey).
A firm of 50 to 500 people sits in the bands where adoption is rising, so the pressure to do something is real. In our experience, the projects that stick start with a contained, measurable workflow rather than an ambitious one.
Step 1: Inventory the Work Nobody Chose to Do
Good candidates hide in the work nobody chose to do. It piled up one request at a time, and now someone spends part of every week on it.
Look in five places:
- Recurring reports. Weekly status decks, monthly client reports, and pipeline summaries assembled by hand.
- Copy-paste handoffs between tools. Data retyped from a form into the CRM, from the CRM into a spreadsheet, and from the spreadsheet into an invoice.
- Requests routed by hand. Support tickets, media inquiries, and inbound sales leads that someone reads and forwards.
- Document production from templates. Proposals, customer stories, and digests that follow the same structure every time.
- Checking and monitoring tasks. Someone opens a dashboard every morning to confirm nothing broke.
Then run a two-week log. Ask each team to add one line every time a recurring task happens: the task name, the owner, how often it occurs, minutes per instance, and the tools it touches.
Two weeks is short enough that people keep logging and long enough to catch weekly work. Monthly work needs a note from its owner. The finished log is the list you score in Step 2.
Step 2: Score Each Candidate on Five Criteria
Score every inventory line from 1 to 5 on each criterion; use 2 and 4 between anchors.
Frequency and volume
Score 1 for a few times a year, 3 for weekly, and 5 for daily or many times a day.
Rule clarity
Score 1 if every case is a judgment call, and 3 for mostly rules with known exceptions. Score 5 for a written procedure a new hire could follow, with one agreed definition of "correct".
Baseline and ownership
Score 1 if there's no owner and nobody measures it, and 3 if an owner exists and time is estimated. Score 5 for a named owner and measured minutes per instance.
Data readiness
Score 1 if inputs live on paper, in heads, or in chat, and 3 if they're digital but scattered. Score 5 if they're structured and reachable by export or API.
Risk and reversibility
Score 1 if an error reaches a customer, regulator, or ledger before anyone sees it, and 3 if a review step catches errors. Score 5 if errors are cheap and reversible.
For an AI decision, run the candidate through the NIST AI Risk Management Framework functions, Govern, Map, Measure, Manage, before building (Source: NIST, AI Risk Management Framework 1.0).
| Criterion | Score 1 | Score 3 | Score 5 |
|---|---|---|---|
| Frequency and volume | A few times a year | Weekly | Daily or more |
| Rule clarity | All judgment calls | Rules, known exceptions | Written procedure, agreed "correct" |
| Baseline and ownership | No owner, no measurement | Owner, estimated time | Named owner, measured minutes |
| Data readiness | Paper, heads, or chat | Digital but scattered | Structured, export or API |
| Risk and reversibility | Error reaches customer, regulator, or ledger | Review catches errors | Cheap and reversible |
These anchors are our own working definitions; copy the table into a spreadsheet and score every line from your inventory.
Step 3: Read the Score
Add the five scores for a total out of 25, then read it against three bands:
- 20 to 25: automate now. This is a first-project candidate.
- 14 to 19: fix the process first. Write the procedure, name an owner, measure it, and rescore.
- 13 or below: leave it manual for now. Revisit it when the process or the data changes.
Two overrides apply regardless of the total. No named owner means not ready. No measured baseline means not ready. A high score can't make up for either, because without them nobody runs the workflow or proves it worked.
One more filter: a one-trigger, one-action job is a native integration, not a project. If the tools you already use can connect directly, switch that on and move to the next candidate.
Step 4: Measure the Baseline Before You Build
The baseline is how you'll prove the automation worked. Record it for the workflow you picked, before anyone builds:
- Instances per month. How often the work actually happens.
- Minutes per instance. Use the median, so one unusual case doesn't distort the number.
- Elapsed cycle time. Calendar time from request to done, including waiting.
- Rework or error rate. How often the output comes back for correction.
- Who touches it. Every person and handoff between request and delivery.
Here's what that looked like for us. Before automating customer stories, our manual baseline was 10 to 14 hours of work spread over 2 to 4 days per story.
The automated portion now takes about 30 minutes, measured across our first 20 production case studies. Interviews and final approval stayed manual.
The baseline made that split easy to defend, because it showed where the hours went: gathering sources, reconstructing the narrative, and drafting. The full build is in our n8n production guide, and the stage-by-stage detail is in our customer-story pipeline breakdown.
Worked Example: Scoring Three Common Workflows
Here's how three workflows common in professional-services firms score against the rubric.
| Workflow | Frequency | Rules | Baseline | Data | Risk | Total | Verdict |
|---|---|---|---|---|---|---|---|
| Customer-story production from interview notes | 3 | 5 | 5 | 3 | 5 | 21 | Automate |
| New-client onboarding setup (folders, accounts, kickoff docs) | 3 | 5 | 1 | 3 | 5 | 17 | Fix the process first |
| Bespoke pitch or proposal writing | 3 | 1 | 1 | 3 | 3 | 11 | Leave manual |
Illustrative scores from our own experience; your numbers will differ.
Customer-story production scores well because we had a named owner, measured time, and an editor sign-off that keeps errors cheap to reverse. It's the workflow behind our Customer Story Generator.
New-client onboarding has clear rules and low risk, but nobody measured it, so it scores 1 on baseline. Measure a handful of instances and it clears the automate-now band.
Bespoke proposals depend on judgment in every case, and a weak proposal costs a deal. Automating research inputs might help later; automating the writing now would not.
What Not to Automate Yet
Some candidates score well on paper and still make poor first projects. Hold these back:
- The workflow that changes every month. You'll spend more time rebuilding the automation than you save running it.
- The one where two teams disagree on what "correct" means. The automation will encode one team's answer and start an argument.
- The one with no owner. When it breaks, nobody fixes it, and often nobody notices.
- The one whose only value is a demo. If nobody would miss it after the demo, it isn't a first project.
Once a candidate passes the rubric and clears these filters, the next decision is how to deliver it. Our guide on whether to build, buy, or assemble your AI stack covers that choice.
FAQ
How many workflows should we score?
Score every recurring line from your two-week inventory. Scoring is fast once the anchors are agreed, and a longer list makes the top candidates easier to trust. If the list is very long, score the most frequent items first, since frequency drives most of the value.
Should the first workflow use AI at all?
Only if the task needs it. Many strong first candidates are plain routing, copying, or formatting, and deterministic steps are easier to validate. If an AI step is involved, run it through the NIST AI RMF functions, Govern, Map, Measure, Manage, and keep a person reviewing output until you know its error rate.
What if two candidates tie?
Pick the one with the higher baseline and ownership score, because that's the one you can prove. If they still tie, choose the one whose errors are cheaper to reverse. A first project's job is to show a measured result, not to solve the biggest problem.
How long should the baseline measurement run?
Long enough to see the work's normal rhythm. For daily or weekly work, the same two weeks as your inventory usually works. For monthly work, cover at least one full cycle and ask the owner whether it was typical. Record the median time per instance, not the average.
Score a Workflow With Us
If you'd like a second pair of eyes on your inventory, see how we design and run automation for clients.
Bring one line from your inventory and we will walk through the five criteria with you.
