// OPERATIONAL INFRASTRUCTURE FOR BUSINESSES DOING $1M+/YEARBOOK A DISCOVERY CALL →

BLOG

How to Measure Automation ROI Before You Hire an Agency

Nov 2, 2026

Measure the process before you automate it. Record volume, minutes per item, error rate and the cost of each error. Then add the full yearly cost of the build, the running costs and the human review that remains. Net benefit divided by total cost is your ROI. Ask any agency to agree that baseline in writing before it builds.

What should you measure before you hire an automation agency?

Measure five numbers for one process, over a normal few weeks (not the week before a holiday):

  1. Volume: how many times the process runs each month.
  2. Touch time: minutes of human work per run, including re-typing, chasing and checking.
  3. Error rate: the share of runs that need fixing, and what each fix costs in time, credits or lost goodwill.
  4. Delay: how long a run waits between handoffs, which matters when a customer is waiting.
  5. Exception rate: how often a person has to step in because something is unusual.

Time a sample instead of guessing. Ten to twenty real runs, timed with a stopwatch or read from the timestamps in your tools, beat any estimate made in a meeting. Ask the person who does the work to confirm the numbers, because they know which cases are quick and which are not.

This is the same job as the analyse phase on our services page: every process, role, tool and data flow mapped, with the bottlenecks and handoffs that cost the most written down, and a precise build scope as the output. Good business process automation consulting starts here, not with a tool demo.

Are hours saved a real return?

Hours saved count as cash only when a decision turns them into money. That means a temp contract not renewed, overtime removed, a planned hire not made, or a backlog cleared that was holding up invoices. Otherwise the hours are capacity. Capacity is valuable, but it belongs on its own line.

A small saving spread across ten people rarely becomes cash. A large saving concentrated on one role often can. Before you sign anything, decide which of the two you expect, and who will make the decision that moves the money.

What does a worked ROI example look like?

Take a 25-person services firm that hands every sold job from quote to project board to invoice by hand. The numbers below are illustrative. We chose them for the example. They are not a client result and not a price.

LineBeforeAfter
Jobs per month300300
Human time200 hours (40 minutes a job)23 hours (60 exceptions at 15 minutes, plus a 2-minute approval check on the other 240)
Errors5% of jobs (15)2% of jobs (6)
Cost of one error$80$80
Running cost$0$350 a month (platform, monitoring, upkeep)

Now the monthly arithmetic, with a loaded labour cost of $40 an hour:

LinePer month
Hours freed (200 minus 23)177 hours
Value of freed hours at $40$7,080
Error cost avoided (15 × $80 minus 6 × $80)$720
Running costminus $350
Net benefit if every freed hour counts$7,450
Net benefit if only half the freed hours turn into cash$3,910

Over twelve months the cautious figure is $46,920. That is the most you can spend on build, set-up and training in year one and still break even on the conservative view. The generous figure is $89,400. The gap between the two is the reason to agree up front what counts as money.

Notice what the example does not claim. It does not assume the automation never fails, and it keeps a person in the loop: every automated job gets a short approval check, and every exception goes to a human.

Which costs do most ROI calculations leave out?

Most calculations include the build fee and forget everything that happens after launch. Use this list when you compare quotes:

CostWhat to ask
Build and set-upWhat is in scope, and what is not?
Your team's timeHow many hours of working sessions, testing and training do you owe?
Running costsWhat does one run cost at your real volume?
Human reviewWho checks the output, how long does it take, and what does that time cost?
MaintenanceWho fixes it when a tool or an API changes?
Remaining errorsWhat do the mistakes that still get through cost you?

Running costs are getting harder to guess, and a recent change shows why. On 6 October 2026 Zapier announced that its standalone Agents product is now an AI step inside the Zap editor. Zapier says AI steps bill as tasks, and that a step can cost 1, 3 or 5 tasks depending on the model tier. By default a single run that passes 75 tasks pauses for a person, and Zapier's help guide (updated on 1 October 2026) notes that tests of a full Zap currently use tasks the same way live runs do. The same workflow can cost very different amounts to run depending on how it is built.

So ask for expected run volume and a cost per run, not a flat guess. Keep predictable steps as plain rules, and use an AI step only where the input varies, such as free-form email.

Gartner said on 25 June 2025 that over 40% of agentic AI projects will be cancelled by the end of 2027, citing escalating costs, unclear business value or inadequate risk controls. It is a prediction, not a measurement. But its three causes are exactly what a baseline exposes early: a cost nobody tracked, a value nobody defined and a risk nobody owned. Gartner also recommends pursuing agentic AI only where it delivers clear value or ROI.

If you cannot fill in the first table above for one of your own processes, that is the first job. On a free 45-minute discovery call, Yonas looks at how your business runs today, where the ceiling is coming from and what one system would look like for you. No pitch. Book a free discovery call and bring the one process you suspect costs the most.

What should an agency agree with you before it builds?

Get these seven things in writing before work starts:

  1. The baseline: the five numbers above, signed off by the person who does the work.
  2. One owner on your side, someone who feels the queue when it backs up.
  3. A definition of "working", such as an exception rate below an agreed level, measured the same way before and after.
  4. A human approval step wherever the work sends, pays or deletes something.
  5. A run-cost forecast with expected volume, a cost per run and a cap.
  6. A review date, with early indicators first (exception rate, review time, whether your team actually uses it) and the money judged later.
  7. A stop rule: what you will do if the numbers do not move.

This is how we work. On our about page the principle is stated plainly: structure first, automation second, because automating an unstructured process just makes the mess faster. The analyse phase is the product, and every process gets mapped before anything is built. After go-live we stay on as a partner and keep the system current as the business grows.

What do results look like when they are counted properly?

Real results are stated in units a baseline can check. Our results page lists R3Tract at 2,000+ hours saved per year, from proposals that used to be assembled by hand from CRM data and now generate themselves. It lists Red Laser Scanning at 150+ documents automated per month, and MuchGroup at 75+ workflow steps automated per product launch.

Those are results for those clients. Yours depends on your starting point: volume, touch time and the cost of an error. That is why the baseline comes before the quote, and why we would rather count with you than promise a percentage.

How long before you can judge the return?

Read the early indicators in the first weeks and judge the money after a full, normal cycle of your business. Early indicators are the exception rate, the review time and whether people use the new flow or work around it. Compare like with like. December against January tells you about the calendar, not the automation.

Write down anything else that changed in the same period, such as a price rise or a new hire, so you do not credit the automation for something it did not do.

Frequently asked questions

What ROI should I expect from automation consulting?

Nobody can honestly promise a percentage before the baseline exists. The return is set by volume, touch time and error cost in your own process. A good consultant gives you a range with the assumptions written down, and shows you how to check it after launch.

What if I have no data on how long things take?

Sample it. Time ten to twenty real runs of the process, note every handoff and count how many needed fixing. Two weeks of rough, honest numbers are a workable starting point, and the analyse phase fills in the rest.

Should I count extra revenue in the ROI?

Only where you can trace it, for example a faster reply that is followed by a booked job. Keep it on a separate line and do not build the case on it. The cost side of the ledger is easier to prove and easier to defend.

Does AI change the maths?

It adds a running cost per item and a review step, and it makes both harder to predict. Fixed rules are cheaper to run and easier to explain, so use AI where the input is messy and keep a person approving anything with consequences.

Can you tell me the price on a call?

Cost depends on scope, and a scoped quote comes from the discovery call, after we have looked at how your business runs. We do not publish package prices, so any figure in the example above is an illustration of the method, not an offer.

Next step

If you want this table filled in for your own business, book a free 45-minute discovery call. Bring one process and rough volumes. You leave with a clearer picture of where the cost sits and what one connected system would change.

// NEXT STEP

Build the system
your business is missing.

YOUR OPERATIONSAUTOMAGENCYONE SYSTEM

A 45-minute discovery call. We map where your operational ceiling is coming from and what a functioning system looks like for your business. No pitch.