Cost Per Task: The Number That Gets Your 2027 Automation Budget Approved

Cost per task is the fully loaded cost of running one instance of a process, and you calculate it by taking the total annual cost of doing that work (salary plus employer CPF plus bonus plus software plus rework) and dividing it by the number of times the work actually happens in a year. It is the number that makes an automation request defensible, because it compares like with like: $4.45 to confirm one order by hand against $1.30 to confirm one order through a system. A monthly licence fee of $280 tells your financial controller nothing. A unit cost tells them whether to sign.

With budget season closing over the next six weeks, this is the difference between an approved 2027 software line and another year of doing it the slow way.

Why does cost per task beat a monthly subscription figure?

Because a subscription price is an input and a unit cost is an outcome. When you present "the WhatsApp automation costs $280 a month", the only available response is to compare it against other things that cost $280 a month. When you present "each order confirmation costs us $4.45 today and $1.30 after, across 900 orders a month", you have given the decision a denominator. Volume, seasonality and growth all become visible: if December brings 1,400 orders instead of 900, the manual cost rises by 56 per cent and the automated cost barely moves.

It also protects you from the opposite mistake. Plenty of tools look cheap per month and expensive per task because the volume is tiny. A $90 contract-review tool used eleven times a year costs $98 per use. That is not automation, that is a hobby.

How do you work out the cost per task you have today?

Four figures, in this order.

• The fully loaded hourly rate. Take the monthly salary, add 17 per cent employer CPF, then add roughly 8 per cent for the annual wage supplement spread across the year. A $3,300 admin salary becomes about $4,100 a month, or roughly $24 an hour on a 168-hour month. Use the loaded rate, never the bare salary, or finance will correct you in the meeting.
• The honest time per task. Time it, do not estimate it. Preparing a delivery order, checking stock and sending the customer confirmation came to 11 minutes in one wholesale client's case, not the 5 minutes the team believed.
• The volume. Pull it from your system, not from memory. Invoices issued, orders confirmed, quotes sent, claims processed.
• The rework. Count the ones that get redone. Wrong address, wrong price, missed follow-up. Rework is where manual processes hide most of their cost.

Eleven minutes at $24 an hour is $4.45 per order. At 900 orders a month, that is about $4,000 of labour, or $48,000 a year, sitting inside a process nobody has ever costed.

What does the automated version actually cost per task?

Three components, and the third is the one most proposals leave out.

• Software and messaging. Say $280 a month for the platform, plus 1,800 WhatsApp messages at around 8 cents, so $144. Running total $424.
• Human supervision. Someone still checks the queue. Budget 90 minutes a week, which is 6 hours a month at $24, so $146. Running total $570, or 63 cents per order.
• Exceptions. Assume 8 per cent of orders fall out of the flow and take longer to fix by hand than they would have taken to do manually, say $9 each. Blend it: 92 per cent at 63 cents plus 8 per cent at $9 gives a true cost of about $1.30 per order.

$4.45 down to $1.30 is a saving of about $2,800 a month. Against a one-off setup of $6,500, payback lands inside three months. That is a proposal that survives scrutiny, and the exception rate is in it rather than discovered in February.

Which numbers will your financial controller challenge?

Three, every time.

"Is the labour saving real money?" If nobody leaves and nobody is redeployed, you have not saved $2,800, you have released 55 hours. That is only cash if those hours absorb growth you would otherwise have hired for, or replace overtime and temp staff you currently pay. Say which one it is. Claiming a headcount saving you are not going to take is the fastest way to lose credibility on the next request.

"What happens at peak?" Show the number at December volume and at February volume, not just the average. Processes that are marginal at 900 orders are compelling at 1,400 and pointless at 200.

"What is the cost of switching off?" If the tool fails or the price doubles at renewal, can the team go back to doing it by hand? Keep the manual cost per task on the page as your fallback number.

If you are planning to claim a government grant on the setup cost, calculate the case before the grant as well. Support levels change; your process economics should stand on their own.

When does cost per task tell you not to automate?

When the volume is too low for the fixed cost to spread, when the exception rate is above roughly 20 per cent (you are automating a process that is not stable enough to automate yet), or when the task is a judgement call dressed up as admin. Pricing approvals, credit decisions and anything a customer will dispute belong with a person who can be held accountable.

The useful habit is to run the calculation on your five highest-volume repetitive processes before the budget closes, then automate the one with the widest gap, not the one with the newest technology. Low cost per task is not the goal in itself. A defensible number you can repeat in twelve months is.

Frequently asked questions

What is a reasonable cost per task for a small admin process?
There is no benchmark worth quoting, because it depends entirely on volume and complexity. What matters is the gap between your manual and automated figures and how quickly the setup cost is recovered. As a rule of thumb, if payback runs beyond twelve months on an SME process, the case is weak enough that something else should get the money first.

Do I count staff time as a saving if nobody is leaving?
Only if those hours are going somewhere specific. Label it as capacity released, name the work it will absorb, and state the headcount you are avoiding in the next twelve months. An unspecified time saving is treated as zero by most finance people, and they are right to do so.

How often should I recalculate cost per task?
Twice a year is enough for most processes, plus once after any volume shift of more than about 30 per cent. Messaging prices, licence tiers and wage costs all move, and an automation that was cheap per task at 900 orders a month can quietly stop being cheap when volume halves.