What Should You Automate First? The Impact-Ordered Answer

TL;DR: Impact-Ordered, Not Sequence-Ordered

The question I get on nearly every strategy call, phrased almost identically every time: "Where does the time actually go in my business, and which workflow do we automate first?"

The second half of that question is the one people get wrong, and they get it wrong in a predictable direction. They pick the workflow that's most interesting, or the one that's first on the list because someone happened to write it down first, or the one that annoys the owner personally. All three are sequence-ordered. None of them are impact-ordered.

Impact-ordered means you pick by a number: hours per week × repetition × low judgment required. The workflow that scores highest goes first, even when it's boring. Especially when it's boring. And a line from that same call has stuck with me as the whole philosophy in seven words: AI does the dishes so the team can do the art.

First, Find Out Where the Time Actually Goes (30 Minutes)

You can't score what you can't see, and most owners genuinely cannot see their own week. Not because they're careless — because the recurring stuff is invisible by design. You stopped noticing the forty minutes of inquiry replies years ago.

You don't need a time-tracking rollout to fix that. Here's the 30-minute version:

• Open the last two weeks of your calendar and your sent folder. List every task that shows up more than twice. That list is roughly 80% of your recurring load, and it's more honest than any survey.
• Ask each person who touches those tasks one question: what do you do every week that you'd hand to a competent assistant tomorrow? People answer that one accurately, because they've already fantasized about it.
• Write the overlap down. Anything appearing in both the calendar sweep and the answers is your shortlist. Ten to fifteen items is normal.

You're not after minute-level precision. You're after the shape of the week — which chores are structural and which just felt loud this month.

The Score: Hours × Repetition × Low Judgment

Now score every item on the shortlist across three 1–5 scales and multiply them for a number out of 125.

• Volume (1–5). Hours per week the task eats across everyone who touches it. Under 2 hrs = 1 · 2–5 hrs = 2 · 5–10 hrs = 3 · 10–20 hrs = 4 · 20+ hrs = 5. Count the whole team, not just you.
• Repetition (1–5). How identical is each instance? 5 means same shape every single time. 1 means every one is bespoke and you're improvising from scratch.
• Low judgment required (1–5). How teachable are the rules? 5 means a new hire could follow them after a ten-minute explanation. 1 means the rules live entirely in one person's head and they'd struggle to write them down.

Read the result like this: above 60 is a first candidate. 30–60 is a second wave, after something boring is already working. Under 20 you should never automate — the install cost exceeds the return no matter how much you personally dislike the task.

The multiplication matters more than the individual scores, because it's brutally unforgiving. A task that eats 25 hours a week but requires real judgment on every instance scores 5 × 2 × 1 = 10 and is disqualified. That's correct. Volume alone never justifies automation.

Worked Examples: What Wins, What Always Loses

Run a few real ones so the scale calibrates.

• Inbound inquiry triage and first response. Say 6 hrs/wk (3), same shape every time (5), rules teachable in ten minutes (4) = 60. This or something adjacent wins the sort in most owner-led businesses I look at.
• Meeting notes into the CRM with a next step. 4 hrs/wk (2) × identical structure (5) × almost no judgment (5) = 50. Unglamorous, and it disappears an entire category of "I forgot to log that."
• Lead and customer follow-up sequences. 5 hrs/wk (3) × (5) × (4) = 60. Often the highest-revenue automation in the building, because the follow-ups that don't happen are invisible losses nobody grieves.
• First-draft quotes and proposals from a template. 5 hrs/wk (3) × (4) × (3) = 36. Second wave — worth doing, not worth doing first, because the judgment score drags it.
• Pricing decisions. 1 hr/wk (1) × every case is different (2) × pure judgment (1) = 2. This is the art. Never automate it.
• Hiring decisions. 2 hrs/wk (2) × (2) × (1) = 4. Same verdict, plus real legal exposure if you get creative with screening.
• Monthly financial review. 3 hrs/wk (2) × (4) × (2) = 16. Under the floor. You can get AI help preparing the inputs; the review itself stays yours.

Notice the pattern. Intake and follow-up win over and over. Pricing and hiring lose every time, in every business, no exceptions. If your scoring produces a different answer, check whether you scored judgment honestly or optimistically.

The First Automation Candidate, by Business Type

Different industries, same math. Here's where the score usually lands:

• Professional services (agencies, consultancies, accounting, legal, marketing): inbound inquiry triage plus first-response drafting. Second: meeting notes to CRM with the next step attached.
• Field and trades businesses (HVAC, electrical, contracting, property management): after-hours and missed-call response. Missed calls are the leak everyone knows about and nobody has staffed for; a same-minute reply with a booking link beats a callback tomorrow.
• Practices (dental, medical, veterinary, therapy): recall and unscheduled-treatment follow-up. High volume, scripted, and it maps directly to production that already exists in your own records.
• Brokerages and agencies (insurance, real estate, mortgage): lead follow-up cadence, plus document intake summarized into a client record. Keep anything touching eligibility, pricing, or screening on the human side of the line.
• Ecommerce and retail: routine customer-service replies — order status, returns, the same eight questions in the same eight shapes.
• Software and product companies: support-ticket first drafts and release notes, which are both template-shaped work that engineers resent doing.

If your business isn't on this list, the sort still works. Ask which chore your team would describe as "the thing we do constantly that a smart temp could learn by Thursday," and score it.

What Not to Automate First

Five categories, and they fail for different reasons — worth knowing which one you're looking at.

1. Judgment-dense, low-volume decisions. Pricing, hiring, firing, partnerships, whether to fire a client. Low volume means little payback; high stakes means every mistake is expensive. This is the art. Protect it.
2. Anything whose rules live only in one person's head. If the expert can't write the rules down in an afternoon, you'd be automating a guess and calling it a system.
3. Anything with regulatory exposure on the first pass. Tenant screening, lending decisions, clinical advice, legal filings, anything with privilege or fair-housing implications. Not never — just not while you're still learning what the system gets wrong.
4. Broken processes. Automating a broken process gets you broken output faster and at higher volume. Fix the process manually first; if you can't describe the good version, there's nothing to install.
5. Whatever you personally hate most. This is the one that stings. Irritation is not volume. The chore you resent is usually rare — that's part of why it stays irritating — and it will score in the teens. Score it anyway, then put it in the second wave where it belongs.

AI Does the Dishes So the Team Can Do the Art

That line came out of a strategy call last spring and I've used it in every conversation since, because it settles the two arguments that stall these projects.

The first argument is prioritization. Dishes are the recurring, low-judgment, high-volume work: the replies, the follow-ups, the notes, the status updates. Art is the judgment work: the pricing call, the hard conversation, the design decision, the read on whether this client is a fit. You automate the dishes. The whole scoring framework above is just a way of finding which dishes are stacked highest.

The second argument is the one your team is actually having in the hallway — is this how we get replaced? The honest answer is the same sentence, read the other direction. Nobody's identity is built on typing the same reply for the ninth time this week. When the dishes leave, what's left is the part of the job people wanted when they took it. That framing does more for adoption than any rollout memo, because it's true and they can check it against their own Tuesday within two weeks.

You Picked One. Now What?

Selection is this post's job; installation is a different discipline, and it has a specific shape: AI drafts, one named human approves, the whole thing lives inside the tool your team already has open, and you measure drafts approved per week for 30 days. The full conversion sequence is written up here — run it on exactly one workflow, and don't start number two until number one is boring.

Two things worth knowing before you start. The tooling floor is lower than people expect: a Claude Team plan runs about $25 per seat with a five-seat minimum, so roughly $125/month, usually less than the half-used subscriptions it replaces. And the ways this goes sideways are well documented — the seven failure modes are almost entirely sequencing and ownership problems, not technology problems, which is why picking correctly matters this much. If you want the quarter-long view of how first workflow becomes fifth, the 90-day integration playbook covers it.

Do this today: open your sent folder, list every task that appears more than twice in the last two weeks, and score the top five on hours × repetition × judgment. Twenty minutes, and you'll have an answer you can defend — instead of a list you'll still be discussing in October.

Getting the Ordering Done For You

If the bottleneck is method, everything above is the method. Run it yourself; it's an afternoon and a spreadsheet.

If the bottleneck is capacity or confidence — you have a dozen candidates, the scores are close, and you want the ordering in writing with the tradeoffs named — that's precisely what the AI Readiness Assessment is: $2,500 flat, 2 weeks, a 15–25 page written roadmap that names which workflow goes first, who owns it, and what it should measure. It is this prioritization, done for you, against your actual week. If we build together within 90 days, the fee becomes your deposit. Capacity is 2–3 a month, so lead time is real. Here's exactly what's inside one, including when you don't need it.

Already know what goes first and just want it running? The done-for-you install starts at $4,500, takes about two weeks and roughly three hours of your time. Or start free: the 30-minute call is a fit gut-check, no pitch, and the AI Readiness Checklist takes five minutes.

Related guides: turning AI ideas into systems your team uses, how to integrate AI into your business, what an AI audit actually includes, why AI implementations fail.