What Does Your Back Office Actually Cost?

Your CFO knows revenue per channel by heart. Gross margin. Shipping cost per order. But ask her what it costs to process a supplier invoice, update a product description or handle a return. Silence. Not because she doesn't care, but because the cost is spread across salaries, system licences and time nobody measures.

That is exactly why it costs so much.

The hidden maths

Let's do the numbers. Not with vague percentages but with actual figures.

A manually processed supplier invoice costs on average $10.18. An automated one costs $3.12. The difference per invoice looks small. But an e-commerce company doing 200M SEK in revenue handles thousands of invoices per year. Multiply the difference and you land on hundreds of thousands of kronor. Per year. In a single process.

Add the error rate. Manual data entry has an error rate of 1 to 4 per cent. Each error costs $10 to $40 to fix. That is not just money. It is time, frustration and delayed shipments.

The average employee spends 9+ hours per week on manual data entry. In IT and finance roles, it rises to 20+ hours. Every week. Half the working week moving data between systems instead of creating value.

The worked example: a 200M SEK e-commerce company

Let's take a concrete case. A Nordic e-commerce company doing 200M SEK (~$19M) in revenue, 50+ orders per day, 2,000 active SKUs and a mix of B2C and B2B.

Supplier invoices. 3,000 invoices per year. Manual processing: $30,500. Automated: $9,400. Saving: $21,100.

Product data management. Two people spend a combined 30 hours per week updating product information across three systems. Fully loaded cost: roughly $85,000 per year. With an automated product data pipeline, that drops by 60 to 80 per cent.

Returns. 8 per cent return rate, 12 minutes of manual handling per return. 2,000+ returns per year. Total: 400+ person-hours just on return administration. That is a quarter of a full-time employee doing nothing else.

Error costs. At 50 orders per day and a 2 per cent error rate in manual processing, you get 365 errors per year. Cost to fix: $7,000 to $14,600. Plus the soft costs: customer complaints, credit notes and lost trust.

Total automatable admin in this example: $140,000 to $190,000 per year. In a company where net margin is 5 to 8 per cent, that is money that moves the needle.

Research from Deloitte shows mid-to-large companies implementing AI-driven back-office automation save $2M to $10M annually. Average ROI of 240 per cent with payback in 6 to 9 months. A fashion retailer automated its invoice processing and saved $36,500 in year one with payback under one month. AP automation alone delivers 700 per cent ROI in the first year for mid-sized businesses.

Why "we'll automate" is not the answer

Here is the uncomfortable truth. Most companies that invest in automation get it wrong.

Deloitte measured it: 59 per cent of organisations treat AI as a technology upgrade. They layer a new tool on top of their existing processes. Same steps, same handoffs, same approval flows. Just with a bot sitting on top.

That gives you marginal improvement. Maybe 10 to 15 per cent. Well below the potential.

Only 16 per cent of organisations have actually redesigned their processes. They asked themselves: if we were building this from scratch today, with AI as the starting point, what would the flow look like?

The answer is never "the same as before but faster".

Automating a bad process gives you a fast-running bad process. The real value comes when you do three things simultaneously.

1. Map the process. Not as it reads in the handbook, but as it actually works today. With every exception, workaround and unofficial step.
2. Redesign the flow. Remove steps that exist only for historical reasons. Eliminate handoffs that create delay. Build quality checks into the process, not after it.
3. Automate the new flow. With AI agents that handle the entire chain, not just individual steps.

That requires experience with both processes and systems. It requires someone who has seen the inside of enough e-commerce companies to know where the skeletons are.

Lights Out: fixed cost, SLA guarantee, no hiring

That is what we do with Lights Out. Not automation as a project. Not a platform you buy and then configure yourself with your already overloaded IT team.

Lights Out takes operational responsibility for entire processes. Fixed monthly cost. SLA on quality and throughput. You pay for completed transactions, not for occupied desks.

Here is how it works:

1. We map the process. Not all at once, but those that cost the most per completed step.
2. We build a digital twin, a copy of your flow where the AI agent can work without touching real data. We run until quality is proven.
3. Production with SLA. The agent works in your real systems. Every transaction is verified before it is executed.

What you do not need to do: recruit, train, manage or replace staff for these processes. No three-to-six-month onboarding every time someone leaves. No stress before Black Friday or peak season. Scaling without hiring.

Salary costs rise every year. The cost of executing a task with AI drops every month. Lights Out puts you on the right side of that curve.

The question is not whether your processes can be automated. The question is how long you can afford to keep them manual.