Manual work rarely announces itself as a problem. It shows up as a team that is always slightly behind, a month end that takes four days, and a person everyone has to wait for.
This guide lists the signs that are worth counting, explains how to put a number on each, and says which ones justify automation and which point at a process that should be simplified instead.
Key takeaways
- The signs are countable: Keystrokes, chase messages, corrections and days to a trusted figure all produce numbers you can gather this week.
- Count an ordinary week: A figure taken from your worst week gets dismissed, and the ordinary week is usually bad enough.
- Price the errors separately: The expensive part of manual work is the mistake that reaches a customer, and no time study shows it.
- Re-count after the change: If the hours moved to another desk rather than disappearing, the same tally sheet will say so.
Look for the business signal behind the technology
The signs that manual work has become a constraint are countable rather than atmospheric. The same record typed into a second system, a daily round of messages asking where something is, a month-end that takes four days to trust: each of these produces a number, and the numbers are what make the case.
Some manual steps are doing real work. A person checking an unusual order is applying judgement that would take a long rulebook to replicate, and automating it badly means automating the mistakes. Separate the steps that decide something from the steps that move information around.
Follow one process for a week and write down what happens between the steps. The gaps are where the cost sits: a form waiting in an inbox, a question waiting for an answer, a file waiting for somebody to come back from lunch.
Use four lenses to understand the opportunity
Duplicate Entry
Follow one record through every place it is typed and count the keystrokes. Duplicate entry is the easiest sign to quantify and the hardest to argue with, because the number comes from watching rather than from an opinion about how the process should work.
Status Chasing
Count the messages that exist only to ask where something is. If a person spends an hour a day answering that question, the answer belongs on a screen, and publishing it is usually cheaper than automating anything behind it.
Error Recovery
Work out what a typical mistake costs to notice, trace and undo. Separate the ones caught internally from the ones a customer found, because the second group carries a cost in goodwill that never appears in a time study and is usually the larger number.
Reporting Delay
Record how long after month end the figures are trusted and which decisions wait on them. A delay here is not a reporting problem on its own, it is the accumulation of the other three areas arriving late and needing to be reconciled by hand.
Assess the current state without buying software first
Count on an ordinary week rather than the worst one you remember. A figure taken from a crisis gets dismissed as unrepresentative, and the ordinary week is usually bad enough to make the point. Keep the tally sheet, because you will want to run the same count again afterwards.
Decide what useful measurement looks like
Convert the counts into hours and the hours into a cost somebody in finance would recognise. Keystrokes and chase messages are vivid but unarguable only when they have a number attached, and the conversion is what moves the conversation from annoyance to budget.
Measure the errors separately from the effort. Time saved is the headline, but the expensive part of manual work is usually the mistake that reaches a customer, and that cost does not appear anywhere in a time study.
When software is likely to help
Automation pays where the volume is high, the rule is stable and the current failure has a price. One of those three on its own is rarely enough, and a low-volume process with a complicated rule is usually better fixed by simplifying the rule.
Hold off when the process changes every quarter or when two teams still disagree about how it should run. Automating a disputed process freezes the dispute into software, and unpicking it later costs more than the original manual work ever did.
For what the digitised version of this work looks like in practice, see Digitalization in Business: Benefits and Roadmap.
Start with a bounded improvement
Take the single largest number from your count and change only that step, by hand if necessary. Re-count at the end of the month. If the saving is real it shows in the same tally, and if the work simply moved somewhere else you have found that out for the price of four weeks.
Separate a structural constraint from a temporary problem
Test whether the pattern survives a change in conditions before it becomes a project. Does the duplicate entry still happen when the team is fully staffed, in a quiet month, with the experienced person on shift? A problem that disappears under those conditions is a resourcing problem wearing a technology costume.
Put your cases in a grid: routine and exceptional on one axis, busy and quiet on the other. Check how duplicate entry, status chasing, error recovery and reporting delay behave in each square. Most businesses find that one square accounts for the bulk of the cost.
Where the constraint is structural, state the improvement as a claim that could turn out to be false: this step will take under an hour, for this team, by the end of the quarter. A claim that cannot fail will be declared a success regardless of what happened.
Risks to make visible
| Lens | Common risk | Early response |
|---|---|---|
| Duplicate Entry | Only the easy entry is automated | Run the new path against last month's five worst exceptions before going live. |
| Status Chasing | Chasing moves to another channel | Count the status messages in the receiving team as well as in the sending one. |
| Error Recovery | Error cost is never priced | Take twenty corrections and record what each took to find, trace and undo. |
| Reporting Delay | The saving is never checked | Repeat the original count ninety days later using the same tally sheet. |
A 30-day validation plan: Business process automation
Days 1 to 5: count the duplicate entry. Sit with the person who handles the highest volume and count how many times one record is typed into a second place. Count the status messages they send and receive in those five days as well. You are after two numbers, keystrokes and interruptions, because those are the ones that convert into hours and that a supplier cannot argue with. Count on an ordinary week rather than a peak, because a number taken from the worst week of the quarter will be challenged and lost.
Days 6 to 15: find out what the errors cost. Pull the last twenty corrections and work out what each one took to notice, trace and undo. Separate the ones caught internally from the ones a customer found, because the second group carries a cost the hours do not show. If reporting delay is on your list, time how long after month end the figures were actually trusted. Ask the customer-facing team as well, since the errors they absorb quietly never reach the correction log at all.
Days 16 to 30: automate one step and check. Pick the single largest number from the two weeks before and change only that step, by hand if necessary. Re-count at the end of the month. If the saving is real it will show in the same count, and if it moved somewhere else you have learned that before committing a budget rather than after. Keep the original tally sheet, because the comparison is only convincing if both counts were made the same way by the same people.
A practical worksheet: Business process automation
The signs of an overloaded manual process are specific and countable. Take one process you already suspect, and put a number against each row from a single ordinary week rather than from an unusually bad one.
| Decision area | What to record |
|---|---|
| Duplicate Entry | Every place the same fact is typed again, and the keystroke count a person performs to move one record from the first system to the last |
| Status Chasing | How many messages per week exist only to ask where something is, and who sends them |
| Error Recovery | What a typical mistake costs to find and undo, counted in hours and in the apology it requires |
| Reporting Delay | How long after month end the numbers are trusted, and what decisions wait on them |
| Measurement | The hours per week the four rows add up to, converted into a cost you would be willing to show a finance director |
A process that scores low on all four rows is working, and automating it buys very little. The case for change comes from one row being large enough to fund the work on its own, not from an even spread of mild annoyance.
Related guides
- There is a longer treatment of the same subject in Digital Transformation Roadmap: A 7-Step Business Guide.
- Where this gets expensive is covered in Paperless Workflow: How to Digitize Paper-Based Processes.
Should Business Process Automation: 7 Signs Manual Work Is Holding You Back turn into a plan with owners and dates attached, Discuss Your Digital Roadmap.
Ali Boran Gazel