Follow the waiting, not the working.
To find the bottlenecks in your business, stop watching how hard people work and start tracking where work sits still. Every bottleneck announces itself as a queue: the approval nobody has opened, the stack of orders waiting to be typed into a second system, the numbers that wait until month-end for someone to assemble them. Walk your main workflow once, write down every place a job waits, put an annual dollar figure on each wait, and rank the list. That list is your bottleneck map — and the top of it is rarely where the overtime is.
The reason this works: effort hides bottlenecks, and queues expose them. People flex — they batch, they stay late, they work around the problem, so a walk through a busy office tells you almost nothing. Queues don't flex. They just grow. The drag you can feel but can't name — late quotes, slow cash, overtime that never quite ends — traces back to a handful of places where work stands in line. This article is about finding those places and pricing them.
A twenty-minute quote that takes four days.
Take any job that felt slow and put two clocks on it. Touch time: the minutes a human actually spent working on it. Elapsed time: the hours on the calendar between arrival and done. A quote might carry twenty-five minutes of touch time inside four days of elapsed time — which means the quote wasn't slow to build. It was slow to start. The other ninety-five hours were queue, and no amount of hustle at the desk touches them.
You can't see queue time by watching people. You can see it on a walk, because waiting leaves evidence — physical piles, unread counts, aging reports, phone calls. Six kinds of evidence worth hunting for:
| What you'll see on the walk | What's actually standing in line |
|---|---|
| Quotes leave days after the request lands | An estimator's backlog. The quote takes minutes to build; the request stands in line behind everything else on one desk. |
| A stack of printed orders beside a keyboard | Re-entry. Work that already exists in one system, waiting for a human to type it into the next one. |
| Approvals that sit unread overnight or longer | A decision. Sign-offs idling in somebody's inbox while the job stands still. |
| Month-end close stretches into week one | Assembly. The numbers existed all month; someone still has to pull them out of every system by hand after the fact. |
| Overtime in one seat, slack in the seats around it | A mid-flow constraint. Upstream keeps producing; one station can't drain fast enough to keep up. |
| Customers calling to ask where things stand | An invisible queue. Every status call is a job that has waited long enough for someone outside the building to notice. |
The month-end row is the easiest one to picture. In the multi-brand reporting automation case study, a monthly finance report that three analysts stitched together by hand became one that assembles itself, and about 18 hours a week of manual reporting came off their desks.
One afternoon. Four moves.
- 01
Trace the last job you finished. Pull one completed order and follow it from first contact to cash in the bank. For every step, note when the work arrived and when someone actually touched it.
- 02
Split the two clocks at every step. Touch time on one side, elapsed time on the other. The gap between them is queue time — the thing this whole exercise is hunting.
- 03
Ask everyone one question: “what are you waiting on right now?” The answers converge fast — the same name, the same inbox, the same system keeps coming up. Where the answers point, a constraint lives.
- 04
Run the vacation test. Whose week off would stall quotes, dispatch, or billing? Every answer that is one specific name marks a place where work queues behind a person instead of a process.
Write each find down the same way: what waits, what it waits on, and roughly how many payroll hours a week it burns — in re-typing, chasing, assembling, and working around. You'll need that last column for the next step.
Turn every wait into dollars a year.
A bottleneck without a dollar figure is a complaint. With one, it's a decision. For each wait on your list, price a year of it: the weekly hours it burns, at what those hours truly cost you in wages, taxes, and overhead, across all 52 weeks. Ten hours a week of re-keying at a $55 loaded rate is $28,600 a year. That's multiplication, not a forecast — and it's all you need to rank one find against another.
Some waits never show up in payroll. When invoices sit because chasing them is manual work nobody got to, the cash you've already earned ages right along with them. When quotes go out late, some stop turning into orders at all. You won't price those precisely in an afternoon — flag them, and treat the payroll figure as the floor, not the total.
The ROI calculator will hand you the annual figure from three inputs — team size, weekly hours, loaded rate — with no form in the way, and the link keeps your numbers if you want to put them in front of a partner.
Rank the list and read the top three. Those are your bottlenecks, named and priced. And if you'd rather hand the whole diagnosis off, that's the operations audit: $4,500, one week, me walking your flow with the people who run it. You get back the three most expensive waits with hours and dollars attached, and a build list ordered by payback. Hire me to fix one and the $4,500 comes off that invoice; if the week uncovers less than it cost, it's free.
Not every bottleneck is a software problem.
Finding a bottleneck tells you nothing about what kind of fix it needs — and plenty of the waits you'll find dissolve without a line of code. There are three families of fix. Check them in this order, cheapest first, because only one of them involves paying someone like me.
Change a rule
Some queues exist because a decision stopped mattering years ago. Raise the approval threshold, retire the report nobody opens, let routine jobs skip the second signature. Costs nothing and ships the day you decide.
Train a second person
When the vacation test keeps returning one name, the queue is a person. Cross-train a backup, write the steps down, split the role. Software can't drain a queue that exists because only one person knows the steps.
Build a system
When the wait is mechanical — the same fields moved between the same systems, the same chase run on the same schedule — software can hold it permanently, and the queue time goes to zero.
If your top bottleneck dissolves with a rule change, change the rule and pocket the difference — you'll get no argument from me. The walk costs an afternoon precisely so you never buy a system for a problem a policy could kill.
How to find automation opportunities in your business
An automation opportunity is a priced bottleneck that survived the cheaper fixes above and passes four checks. Run each finalist through them before you talk to anyone selling software.
- 01
The input arrives in a known shape. A form, an email with the same fields, a call, a file from the same system. If every case starts differently, a person is still the fastest parser.
- 02
A careful new hire could follow the rule. If you can write the steps on one page, a system can run them. If the steps are “it depends, ask me,” the rule isn't settled yet.
- 03
Done is obvious. The quote is sent, the lead has a reply, the invoice is logged. A task with no clear finish line turns into a system that needs babysitting.
- 04
It happens every week. Monthly chores rarely pay back a build. The work that repeats daily, or arrives at random while you're busy, is where the hours and the lost jobs pile up.
In a small business the bottleneck that passes all four is usually at the edge, where customers reach you: calls that ring out, form leads answered hours late, the same questions asked every day. Those often don't need a custom build at all; a prebuilt agent covers one job for a monthly fee. Deeper in the back office, where several systems meet, a finalist that passes is a candidate for a scoped build.