The Ops Audit: Find Every Manual Process in Two Hours

You know too much runs by hand. You just do not know exactly what, and what should go first.

That is what the ops audit is for. You track for a week, then you need two hours for the analysis. At the end you have a sorted list instead of a gut feeling.

I analysed processes at PwC for twelve years. Back then something like this took six weeks and cost six figures. For an ecommerce brand with two to ten people it can be shorter. Here is the condensed version we used for our own brands nano, mate and MUSTAX.

Why most process analyses fail

The usual mistake: you sit down and write from memory what you do all day. The result is always the same list. Orders, support, reporting. All correct, all useless.

From memory you write down what you consider important. You do not write down what actually costs you time. The twenty minutes in which you retype numbers between two tabs show up in no recollection. They only show up in the week in which you track them.

The second mistake is altitude. Many people note "customer service". That is an entire area. A process, by contrast, has a trigger, a few steps and an outcome. "Return request comes in, create label, email the customer" is a process. You can measure that, you can automate that.

The third mistake is timing. Whoever measures in the week before Black Friday measures a state of emergency. Take a normal week.

What manual processes cost you in total is something I worked through in The true cost of manual processes. The ops audit is the step before that: first know what is there.

Step 1: Track for a week

Five working days. One document, a notes app, a sheet of paper next to the laptop if you like. The only thing that matters is that it is always the same place.

Every time you finish a task, you write down four things:

• What you did. One sentence, concrete: "answered six WISMO emails". "Support" is not enough as an entry.
• How long it took. An estimate is fine. Accurate to five minutes is accurate enough.
• What triggered it. An email, a calendar entry, an order, a customer, your own head.
• Whether you have done it before. A tick. That tick is the most important part later.

Two rules so the week is worth something. First: note it immediately, not in the evening from memory. Second: include the five-minute things. The small tasks are the reason the day is full even though nothing big happened.

If there are two or three of you, everyone writes their own list. Merging comes later. Writing into a shared document means influencing each other.

Step 2: Sort by frequency and duration

Now the two hours. Take your list and merge identical entries. Six rows of "answered WISMO emails" become one row with a frequency.

For every row you need three numbers:

• Frequency: How often per week.
• Duration: Minutes per run.
• Hours per month: Frequency times duration times 4.3, divided by 60.

The third number is the only one that counts. It makes things comparable that otherwise cannot be compared. A task that takes two minutes and happens thirty times a week beats a task that takes two hours and comes up once a month. Almost always, something ends up at the top that you would never have considered a problem.

For us that was order processing at mate. A single run was short. In total it was four hours every day. Today it is 15 minutes a day.

Step 3: The table that comes out of it

This is what the result looks like. The numbers are an example, not a promise.

Two things stand out immediately. First: the top three rows together are more than a hundred hours a month. Second: the bottom two rows are the work you actually became a founder for, and they get six hours.

That distribution is the real finding. Not any single row.

Step 4: Effort against benefit

Now comes the second column you have to fill in yourself: how hard is this to automate. I use three levels, no more are needed.

• Easy. Clear rule, clean interface, no judgement required. One to two days to build. Example: enter the tracking number and send an email.
• Medium. Several systems, a few special cases, maybe a language model for the unclear ones. One week. Example: pre-sorting support emails and preparing standard replies.
• Hard. No clean access to the data, many exceptions, or the process still changes every four weeks. Example: a fulfilment partner without an API.

The hard rule behind it: if you could hand the steps to a new intern as a checklist, a machine can execute them. If you would say "it depends" on every third case, it cannot.

Then you do rough maths. A process with 17 hours a month and effort "medium" pays for itself in a few months. A process with 3 hours a month and effort "hard" never does. What the build realistically costs is in What automation really costs.

Step 5: Set the order

Now you sort one last time, this time by hours divided by effort. At the top is whatever eats a lot of time and is easy to build.

Then you delete everything below third place. Really. Three processes, no more.

The reason is uncomfortable, and I have seen it often enough: whoever starts five automations at once has five half-finished ones after eight weeks. Half-finished is worse than not at all, because then you maintain both routes, the old one and the new one.

For the three winners you write two sentences each: what triggers it, what should happen at the end. That is already most of the briefing. How a project like that then runs is described in How an automation project works. Which station makes sense in which order is in Automating ecommerce processes.

What the audit did for us

We ran this for our own brands, twice in fact. The first time order processing was at the top, which we had suspected. The second time something was at the top that nobody had expected: reporting at MUSTAX.

Hunting down numbers, building tables, checking numbers. In total two days per cycle. It never felt like a problem, because it only came up once. Counted in hours per month it was in second place. Today reporting at MUSTAX takes two hours instead of two days.

Across all processes we save 33 to 46 hours a week today, at three brands with two people. It started with tally marks on a sheet of paper.

What an ops audit cannot do

So you have no false expectations. Four things this method does not find:

• Processes that do not exist. You measure what you do. If nobody at your company analyses return reasons, the missing analysis shows up nowhere. The audit finds waste, not gaps.
• Other people's work. What your fulfilment partner, your agency or your tax advisor does by hand is on none of your lists. It still costs you money and lead time.
• Whether the process is right at all. A process can be efficient and still be nonsense. A weekly report nobody reads should be abolished. For every row, ask once: what happens if we drop this entirely.
• The cost of errors. The audit measures minutes, not damage. A task that comes up rarely but regularly goes wrong looks harmless and is not.

And there is a question of size too. Under about ten orders a day the effort usually does not pay off yet. Your time is better spent on growth there. What you need at which volume is in the roadmap from 0 to 100 orders.

Conclusion

The ops audit needs no consulting project. One week of tracking and two hours of sorting are enough.

Five steps: track, sort by hours per month, estimate effort, weigh benefit against it, pick three processes. The rest waits.

The most valuable part, by the way, is the moment you see how little time per month is left for the work that moves your business forward. That number changes decisions faster than any tool.

Do the week. Afterwards you know more than any vendor trying to sell you a solution.

If you have the list and need someone for the next step who builds this every day: we at Flowhouse are happy to look at your top three rows. No pitch, just a first look.