Ask a business owner how much their manual processes cost and the answer is usually a shrug and an estimate. Ask them how much their software costs and they can tell you to the naira. That asymmetry is why so many teams keep paying for expensive manual work while agonising over a modest subscription. The manual cost is real, it is simply spread across payroll, rework, delay and opportunity in a way that no single report captures.
Automation reduces cost in four distinct places. Each one is measurable, and only the first is obvious. This guide walks through all four, shows how to put a number on each, and gives you a five-step payback model that works with figures you already have. The point is not to justify a purchase; it is to work out whether one is warranted, and if so, where to point it first.
Cost centre one: labour spent on repetitive admin
This is the saving everybody thinks of first. Copying data between systems, chasing approvals, retyping invoices, updating the CRM after every call, assembling the same weekly report from the same three exports. It is visible, easy to measure, and rarely the biggest number on the list.
To measure it properly, do not ask people to estimate. Estimates of routine admin are consistently wrong in both directions: individual tasks feel shorter than they are, and the switching cost between tasks is invisible. Instead, have the people doing the work log the task for two weeks, or count the transactions and time a representative sample. Ten minutes per order across four hundred orders a month is sixty-six hours, which is most of a working fortnight.
Multiply those hours by a fully loaded hourly cost rather than the salary divided by hours. Fully loaded means gross pay plus employer contributions plus the share of software, equipment and overhead attributable to that person, typically somewhere between 1.2 and 1.4 times base pay. If you use the raw salary figure you will understate the saving by a quarter, which is often enough to make a good project look marginal.
One nuance matters here. Automating a task that a junior person does at a low hourly cost saves less than automating the same task when a senior person does it. In small teams the second case is extremely common, because the founder or the best salesperson is often the one reconciling records at nine in the evening. Price the work at the cost of whoever actually does it.
Cost centre two: errors and the rework they cause
Manual data entry carries an error rate. Studies of routine transcription work put it somewhere around one percent of keystrokes, and the rate rises sharply under time pressure or at the end of a shift. What matters is not the error itself but the cost of discovering and correcting it.
A mistyped invoice amount is not a thirty second fix. It is a customer query, a conversation with accounts, a credit note, a revised invoice, a delay in payment and, occasionally, a damaged relationship. A mis-keyed email address in the CRM means a proposal that never arrives, followed by a lost deal that nobody ever attributes to the typo. A wrong delivery address means a second shipment paid for twice.
Measure this by counting incidents rather than theorising about rates. Go through the last three months of credit notes, refunds, re-sends, complaint emails and correction entries. Attach a realistic handling cost to each, including the time of everyone who touched it, and where a customer was lost, include the value of that relationship. Most teams find the number considerably larger than expected, because the individual incidents are small enough to be forgotten and frequent enough to add up.
Automation does not eliminate errors. It changes their character. A rule encoded once produces consistent output, so when it is wrong it is wrong systematically and visibly, which means it gets found and fixed once rather than recurring quietly forever.
Cost centre three: delay, the most expensive and least tracked
This is where the real money sits, and almost nobody has it on a report. Delay costs are the revenue consequences of things happening slowly.
Start with lead response. The research on this has been consistent for years: the odds of qualifying a lead fall off dramatically after the first hour, and a response within five minutes converts at a multiple of one sent the next morning. If you take a hundred enquiries a month, close ten percent of them at an average value you can calculate, and your average response time is six hours, the gap between that and a five minute automated acknowledgement plus routing is worth more than every hour of admin in the business.
Then look at cash. Invoices that go out at month-end instead of on completion delay payment by an average of two to three weeks. Multiply your monthly invoiced value by the delay and you have the working capital sitting idle. If you use an overdraft or any form of financing, that idle capital has a direct interest cost. If you do not, it has an opportunity cost that shows up as a purchase you deferred.
Finally, look at handoffs. Every point where work waits for a human to notice it is a queue, and queues have a length. A quote that needs approval before it goes out, a support ticket that needs assigning, an onboarding pack that needs a signature chased. Time the queues. They are usually longer than anyone in management believes, because the person waiting rarely escalates.
Cost centre four: software and outsourcing you stop needing
The smallest of the four, but the easiest to bank. When a workflow is automated properly, some of the tools that existed to manage the manual version become redundant: the standalone scheduling app, the duplicate project tracker somebody bought during a busy quarter, the per-seat licences for people who only ever opened the tool to copy a number out of it.
Outsourced data entry and virtual assistant hours often fall into the same category. Not always, and not entirely, because the judgement work usually remains. But if you are paying an external provider by the hour to process documents, and document parsing is exactly the thing that has become reliable and cheap, that line should shrink.
Audit your subscriptions annually with the workflow map in front of you. Tools accumulate. The saving here is modest but it is recurring and it requires no build work beyond cancelling.
The five-step payback model
With the four cost centres measured, the model is simple arithmetic. Run it on one workflow at a time.
One: total the current monthly cost. Labour hours at fully loaded rates, plus error and rework costs, plus delay costs, plus any software or outsourcing attached to this specific process.
Two: estimate the residual. Automation rarely removes one hundred percent of a process. Exceptions still need judgement, and someone still monitors the runs. Assume you remove seventy to eighty-five percent of the labour on a well-suited workflow, and be honest about which exceptions remain manual.
Three: price the build and the running cost. Implementation fee or internal build time, platform subscription at your projected volume, and an allowance for maintenance. Budget maintenance at roughly ten to fifteen percent of the build cost per year; connected systems change and workflows need updating.
Four: calculate the monthly saving. Current monthly cost minus residual monthly cost minus running cost. If this number is negative, stop. The workflow is either too small or too irregular to be worth automating, and that is a legitimate outcome of the exercise.
Five: divide build cost by monthly saving. That is your payback period in months. Under three months is strong and should proceed immediately. Three to six months is sound. Beyond twelve months, either the workflow is the wrong candidate or the build is over-specified for what it needs to do.
A worked example
Take a services business processing three hundred supplier invoices a month. Each takes about eight minutes to open, key in, route for approval and file: forty hours a month at a fully loaded rate of, say, four thousand naira an hour, so one hundred and sixty thousand naira in labour. Errors run at roughly six corrections a month, each costing about an hour of combined time across two people, adding another forty-eight thousand. Late approvals cost two early-payment discounts a month worth thirty thousand. Total: two hundred and thirty-eight thousand naira a month.
Automated extraction with rule-based routing and a human approval step for anything over a threshold removes about eighty percent of the keying, most of the transcription errors, and effectively all of the approval delay. Residual labour plus platform costs come to around sixty thousand. Monthly saving: roughly one hundred and seventy-eight thousand. Against a build cost of six hundred thousand, the payback lands a little over three months, and every month after that is margin.
The numbers in your business will differ, but the shape rarely does. The labour line is what gets attention; the delay line is what usually makes the case.
Where the savings usually hide
Two patterns come up in almost every audit we run. The first is the report nobody questions. Someone spends half a day each month assembling a document from three exports, and it circulates to a list of recipients who mostly skim it. The saving is not only the half day; it is the fact that once the report is generated automatically it can run weekly instead of monthly, which changes how quickly the business notices problems.
The second is the informal handoff. Work passes between two people by message rather than by system, so there is no record of when it was sent, no reminder when it stalls, and no way to see how long the queue is. These handoffs are invisible to every report you have, which is precisely why they are expensive. Instrumenting them, even before automating them, usually reveals a delay cost larger than the labour cost sitting on either side of it.
A third, smaller one worth checking: duplicated data entry across systems that were never connected. The same customer details keyed into the CRM, the accounting package and the delivery portal. It feels trivial per transaction and it is rarely trivial per month.
What automation does not save
Be equally rigorous about the limits. Automation does not save money on processes that run a handful of times a year, no matter how tedious they are. It does not save money on work that requires genuine judgement at every step, because you end up building an elaborate wrapper around a human decision. And it does not save money on a process that is still changing every month, because you will rebuild it three times before it settles.
It also does not, in most small and mid-sized businesses, reduce headcount. The saving shows up as avoided hiring and increased throughput: the same team handling twice the volume without the quality falling over. That is a real financial outcome, and it is worth stating plainly to a team who may otherwise assume the project is aimed at their jobs.
Frequently asked questions
Start with one number
You do not need a full audit to begin. Pick the process that irritates your team most, count how many times it runs each month, and time it honestly. Then add the delay cost. If the total surprises you, that is your first automation, and the payback model above will tell you whether the build is worth commissioning.