Back office automation
Back office automation has a marketing problem: the work is invisible until it fails. Nobody thanks the person who reconciled the accounts, matched the deliveries or kept the customer records straight, so nobody counts the hours either — which makes it the hardest kind of automation to get funded and often the most profitable once it runs.
Finding the hours before arguing about them
A company growing through the middle usually has no idea where its administrative time goes, which is why these projects stall at the business case. Two weeks of looking properly settles it.
- Count the copies. Every time the same fact is typed into a second system, that is an hour a month you cannot see and a disagreement waiting to happen.
- Count the chases. Messages whose entire content is asking somebody to do something they already agreed to do. In most companies this is the single largest category of administrative work.
- Count the reconciliations. Anywhere two lists are compared by a human, monthly, to find the handful of rows that differ.
- Count the assemblies. Reports, packs and summaries built by copying from four places into a fifth. Always on a deadline, always by the same person.
What to build, in order of payback
- Stop the re-typing: one place where a fact is entered and everything else reads from it. Dull, and it removes the disagreements as a side effect.
- Automate the chasing: who owes what, asked on a schedule, in the company's tone, escalating to a person when the answer is not routine.
- Turn reconciliation into an exception list: the machine compares, a person looks only at the rows that differ, with the reason attached.
- Assemble the recurring packs, leaving the commentary to whoever owns it.
- Add something that notices when any of the above stops, because back office automation fails silently by definition.
What to leave alone
- Anything where the exception rate is high and the rules are unwritten. Automating that produces an exception queue as big as the original job.
- Judgement dressed as admin: which customer gets credit, which supplier gets paid first, which invoice to dispute. It can prepare the case; a person decides.
- Processes about to change. Six months before a new system is the wrong time to automate around the old one.
- Work that only looks repetitive. Ask the person who does it what they check that is not written down; the answer is usually the reason it is still manual.
The measure that matters afterwards
Not hours saved, which nobody can verify and everybody disputes. Two better measures: how long a thing takes from start to finish, and how many exceptions arrive per hundred items. Both are countable before and after, and both are hard to argue with.
The same watching, already running
- Sales funnel report — the exception-list principle in production: roughly 750 items read every day and only the ones with a real problem surfaced, so nobody reads 750 rows to find four.
- Personal AI assistant — the small administrative half in daily use — things chased, notes filed where they can be found, research done. Back office work is that, with more names attached.
Questions from finance and operations
What counts as back office automation?
The administrative work that keeps a company running and never appears in a sales figure: re-keying between systems, chasing people, reconciling lists, and assembling recurring packs. It is invisible until it stops, which is exactly why it is under-automated.
How do I build a case for it if nobody counts the hours?
Do not argue about hours. Measure how long a process takes end to end and how many exceptions arrive per hundred items, before and after. Both are countable and neither invites a debate about whose time was really saved.
Where does this usually pay back fastest?
The chasing, followed by reconciliation as an exception list. Both are high volume, both have clear rules, and both are jobs people are relieved to hand over rather than protective of.
Will this reduce headcount?
Usually it absorbs growth instead — the same team handling twice the volume without the administrative load rising with it. That is the honest pattern in companies this size, and it is worth saying plainly rather than implying more.
What is the biggest risk?
Silent failure. Back office automation is defined by nobody watching it, so the monitor is not optional. If a proposal does not include something whose only job is to notice a stoppage, it is incomplete.
Measure it before and after, or do not start
Pick one process. Time it end to end, and count how many of every hundred items come out as exceptions. Those two numbers are the only honest way to know later whether any of this worked, and almost nobody records them beforehand. Do that first. Then, if you want a view on which process to take, the first hour is free.
Book that hour →