Every small business has more automation candidates than it can act on: the inbox that needs sorting, the invoices that need chasing, the weekly report nobody enjoys putting together. This guide gives you a repeatable way to find those candidates, score them the same way every time, and cut the list down to the two or three actually worth testing.

Identifying High-Impact Automation Opportunities

Do not start from a list of AI features. Start from a week of your team's actual work, and write down every task that repeats.

The best candidates tend to have the same shape. Someone receives information in one place (an email, a form, a voicemail), applies a rule they already know, and moves the result somewhere else (a spreadsheet, a CRM record, a reply). The thinking is light; the coordination is what takes the time.

In practice, most of them show up in four places. Intake: inquiries and orders that a person reads and enters into another system. Chasing: payment reminders, missing documents, status checks, anything where the whole job is following up. Reporting: anything assembled from two or more systems on a schedule, like the Friday client update. And record upkeep: typing the same information into the CRM, the job tracker, and the billing tool.

There is real evidence that this unglamorous list is the right place to look. MIT's 2025 State of AI in Business research found that more than half of generative AI budgets go to sales and marketing tools, while the biggest measured returns came from back-office automation. In other words, most businesses point their AI spending at the visible work and miss the paperwork that actually eats the week.

One thing to keep in mind while you build the list: at this stage you are collecting workflows, not solutions. "Automate the proposal follow-ups" belongs on the list. "Buy tool X" does not.

The Use Case Scoring Framework

Once you have a list, score every candidate the same way, 1 to 5 on each of five factors. The scoring doesn't need to be precise. It just needs to be consistent, so you're comparing workflows on evidence instead of on which one annoyed you most recently.

Score risk in reverse: low risk earns the 5. A workflow that is frequent, expensive, predictable, valuable, and safe to get wrong occasionally is a strong candidate. A workflow that fails badly on two of these usually isn't, no matter how tedious it feels.

It also helps to put a dollar figure on the time, so the comparison stays grounded. The method is simple: weekly volume × minutes per instance × the share that's predictable enough to automate. For example: 40 inquiries a week × 6 minutes each × 70% automatable comes to about 2.8 hours a week. At the U.S. private-sector median compensation of $34.78 an hour (BLS, March 2026), that one workflow is worth roughly $97 a week, or about $5,000 a year, and that's before you count what a faster response is worth. Run the same arithmetic on every candidate and the ranking usually sorts itself out.

What Not to Automate

Some workflows score poorly for reasons worth spelling out, because they're often the ones owners want to automate first.

Judgment-heavy, low-volume decisions are a bad fit: pricing an unusual job, responding to an upset long-term client, deciding whether to take on a project. There's no pattern to hand off, and the stakes are high.

Workflows you haven't standardized yet should wait. If three employees do the task three different ways, automating it just locks in one person's version. Fix the process first, then automate it.

Be careful with anything where a mistake is expensive and hard to catch: money movement, tax filings, commitments to customers. Partial automation can still help there, but a person should stay in the loop.

Rare tasks usually aren't worth it either, because the setup and upkeep cost more than the twice-a-year job they'd save. And some work that looks like a task is actually relationship-building. A monthly check-in call is not an inefficiency. Automating the scheduling around it is fine; automating the conversation misses the point.

The common thread: automate the predictable middle of your operation, and keep people on the edges, where volume is low, stakes are high, or the work is the relationship itself.

Narrow Your Opportunities to the Best Two or Three Workflows

Add up the scores, then use three tiebreakers before you commit to a shortlist.

Prefer workflows with a clear start and a clear end. "An email arrives, and it ends with the record updated and a draft ready" is easy to test and easy to judge. "Improve our communication" is neither.

Prefer workflows you can measure today. If you can say "this takes six hours a week and produces two errors a month," you'll be able to prove whether the automation worked. If you can't measure it now, you won't be able to measure the improvement either.

And prefer workflows that live with one person or one small team. Those get adopted. A workflow spread across everybody belongs to nobody.

Then stop at two or three. A shortlist of ten means you haven't really decided, and running five automations at once means nobody is watching any of them properly. The rest of the list isn't wasted; it's the queue for next quarter, once your first workflow has proven itself.

What comes next is a different job from this one. Picking the single pilot from your shortlist, setting its permissions, and rolling it out to the team is covered in the step-by-step implementation plan. Before you build anything, it's also worth checking which level of AI each workflow actually needs, because several of your candidates will turn out to need plain automation, not AI at all.

This scoring exercise is the first thing we do at Unprompted with a new client: we sit down with your list, run the numbers, and tell you which two workflows an agent should own first, and which ones don't need us at all. Send us your candidate list and we'll score it with you.

The Bottom Line

• The best candidates repeat often, follow a pattern, and move information between systems: intake, chasing, reporting, record upkeep.
• Score every candidate on the same five factors, and put dollars on the time so the ranking stays honest.
• Leave judgment calls, unstandardized processes, and relationship work alone.
• Narrow to two or three workflows you can measure and assign to one owner; the rest is next quarter's queue.
• From here: the full picture in AI for small business, the three levels of AI a task can need, and how to roll out your first pilot.