How do I decide which of my processes to automate first?

List every process you are considering. Score each one from 1 to 5 on three things: frequency (how many times a week it runs), cost of error (what one failure costs in money, customers, or compliance), and measurability (whether you can count the before and after). Multiply the three scores and sort. The top of the list goes first.

Frequency is the multiplier people underrate. A build that saves four minutes on a task that runs forty times a day pays itself back on a schedule you can predict; the same build on a monthly task takes years to matter. Cost of error covers both directions: the cost of doing the work wrong, and the cost of the work not happening at all. A sales call nobody returns is an error, even though no system logs it.

Measurability is a gate, not a tiebreaker. If you cannot count the process today, you cannot prove the automation worked, and you cannot hold any vendor to a payback claim. When a high scorer is unmeasured, instrument it first: even two weeks of tally marks in a spreadsheet turns a guess into a baseline.

Two tiebreakers when scores are close. Prefer the process that has not changed in months over the one still being argued about, and prefer the one with a single clear owner, because every automation needs a named person who answers when it stops.

Owners almost always nominate the process that interrupts them: the report they hate building, the inbox that ruins Monday morning. But the expensive failures happen where nobody is standing, which means they never generate a complaint, which means they never make your shortlist on instinct alone.

Phones are the cleanest example we have. At Hall's Heating & Air, a measured 80% of the calls the AI now handles arrive outside business hours. Before anything was automated, those calls were invisible: nobody heard the phone ring at 9pm, nobody logged the miss, and the cost never reached a report. The visible problem was the busy front desk. The expensive one was the empty desk at night.

So for every candidate on your list, ask one extra question: what fails silently here? Quotes that never went out, leads never called back, invoices sent late. Those belong in the cost-of-error score even though nothing currently records them. The opposite trap exists too: a dramatic once-a-quarter task feels urgent because it is stressful, but it scores low on frequency and usually deserves a checklist, not an automation.

If you have fewer than ten candidate processes and one person who genuinely knows them all, do not pay anyone. The framework above is the whole method, and a spreadsheet will produce the same ranking a consultant would.

An audit earns its fee when the knowledge is spread across departments, when nobody trusts anyone else's scores, or when the numbers you need, volumes and error rates, do not exist and someone has to go count them. That is what Benian's AI Audit is: four weeks, fixed scope, $4,500, built on interviews with the people who do the work, ending in a ranked plan with what each step takes and what it should pay back. The plan is yours whether Benian builds any of it or not, and buying the audit commits you to nothing after it.

Whoever you hire, including us, check three things. Ask them to label every number they show you as measured, client-reported, or a projection; a vendor who blurs those labels is marketing, not diagnosing. Ask who owns the plan if you walk away after the diagnosis. Ask what they would refuse to automate in your business; a vendor with no answer is selling builds, not judgment. And if you handle patient or health data, ask any vendor to put their breach notification window in writing in a BAA and to name where your data lives; a logo on their website is not an answer.

After diagnosis, builds are priced to the work, which is why we publish no build prices. The honest cost drivers: how many systems have to talk to each other, whether a human approval step sits inside the loop, how clean your data is, transaction or call volume, and any compliance review your industry requires. Two businesses automating the same process can land far apart on all five.