Automation projects usually go wrong at selection, before anyone writes a line of code. A team picks the task that annoys them most, discovers halfway through that it happens twice a month, and ends up with a tool nobody opens. The build was fine. The choice was wrong.
We start every workflow automation engagement with the same screening pass. It takes an afternoon of counting, and it decides most of the project's value.
Start with the multiplication
A 20-minute task that runs 40 times a week is 13 hours a week. Do that multiplication for every repeated task in your operation and the ranking writes itself.
| Task length | Times per week | Hours per week | Hours per year |
|---|---|---|---|
| 5 minutes | 100 | 8 | 430 |
| 10 minutes | 20 | 3 | 175 |
| 20 minutes | 40 | 13 | 675 |
| 45 minutes | 10 | 7.5 | 390 |
The five-minute row is the one people miss. Nobody complains about a five-minute task; it is over before it registers as work. At a hundred times a week it costs more than two working months a year.
Frequency beats duration almost every time. A single task has to be genuinely large before it competes with a small one that repeats all day. This is also why the task everyone hates is rarely the right first choice; the ones worth money are usually invisible.
Four tests for a first candidate
It is frequent and short. The arithmetic above. Anything under a couple of hours a week is a bad first project, however irritating it is.
It follows rules you can write down. If you can explain the decision to a new hire in one paragraph, software can make it. If the explanation ends with "you'll get a feel for it," that judgment belongs to a person, and the automation should route the case to them.
It moves data between systems. The most reliable automations take something out of one place and put it into another with a transformation in the middle. There is no ambiguity about what correct output looks like, which makes the work testable and the result easy to trust.
A person currently does it by copy-paste. This is the strongest signal on the list. Copy-paste means the data is already structured, the mapping is already known, and a human is serving as the integration layer between two systems. That person can tell you every edge case in ten minutes, because they hit all of them last month.
A task that passes all four is worth scoping. A task that passes three is worth discussing. A task that passes one belongs in a process review before anyone quotes software for it.
What people automate first
The same handful of workflows come up across industries, because they sit at the seams between systems.
Email triage. Document intake. Reporting. Follow-up scheduling. Moving data between a CRM and accounting tools.
Each one is frequent, rule-shaped, cross-system. Reporting is often the easiest sell, since the output is visible to whoever approves budgets. Email triage usually returns the most hours. Document intake tends to have the messiest inputs, which makes it a better second project than a first.
If your list looks nothing like this one, that is fine. The tests matter more than the examples.
Exceptions are the actual work
A demo handles the straightforward case. The build handles everything else, and that is where the cost lives.
A good automation works through the exceptions it can define and escalates the judgment calls it should not make. An invoice with a missing PO number goes to a person with the reason attached. A message the system cannot classify goes into a review queue with the ambiguity flagged. The failure mode you want is a task landing on someone's desk with an explanation; the failure mode you do not want is a confident wrong entry in your accounting system.
Consistency is the other return, and it gets undersold. Software holds the same standard at 3 a.m. as at noon. The quality floor goes up and stays there. Hours saved is the number that gets a project approved; the consistency is what people notice six months later, when the weekly cleanup meeting stops happening.
What it costs and how to start
Our published range for workflow automation is $8,000 to $30,000. Where a project lands depends on how many systems it touches, how clean the inputs are, how many exception paths the process really has.
Set that range against the arithmetic. A workflow costing 13 hours a week is roughly 675 hours a year; at any loaded hourly rate you want to use, the payback window is short and easy to defend to whoever signs. A workflow costing two hours a month does not clear the bar, and we would rather say so than build it.
Then do it in order: one workflow, measured before and after, then expand using the first project's actual numbers instead of a projection. The first build also shows you where your data is inconsistent, which is worth knowing before you automate five things at once. Automation is one part of our AI services, and it is usually the part with the clearest payback, because the before-and-after is countable.
Bring us the task you did three times today. If it clears the arithmetic we will scope it, and if it does not we will tell you that instead. You can start a project whenever you have the numbers.