How AI finds the money quietly leaking out of your business, and takes over up to 70% of the repetitive work.
The premiseLosses leak, they do not explode
The methodMeasure, then automate
The outcomeCost, speed, accuracy
Ebhota Jonathan
Co-Founder & Partner, Technology, Data & AI, Serge Ltd
The quiet leak
Nobody budgets for the money they never notice leaving.
Most of the money a business loses is not stolen, and it is not lost in one bad decision. It goes out in small amounts, over and over, and it never appears as a line in your accounts.
0%
of the average working day goes on tasks that today’s technology could already do without a person.
McKinsey, 2023
$0M
about ₦18 billion
what messy and wrong records cost a typical large company every year, in wasted work and missed chances.
Gartner
0%
of the working week goes on repetitive work done by hand: email, chasing information, typing the same thing twice.
Smartsheet / Workfellow research
You cannot fix what you cannot see. The first job is measuring, not buying software. Converted at ₦1,400 to the dollar (July 2026).
Diagnostic
Four places the money goes, before it reaches profit.
Doing it by hand, again
Typing the same information over and over, and copying it between systems that do not talk to each other.
1,000+ copy-pastes per week
EXAMPLEThe same order is typed into your main business system, then again into a delivery sheet, then again onto the invoice.
Waiting around
Work that sits still while it waits for someone to approve it or pass it along.
Days, not minutes
EXAMPLEA purchase order waits four days for a signature because nobody knows it is sitting in an inbox.
Mistakes and doing it twice
Mistakes found later, when fixing them costs far more than getting it right the first time.
Up to $53, about ₦74,000, per fix
EXAMPLEA supplier is paid twice because the same invoice arrived by email and by post under two references.
Not knowing what is going on
Decisions made on numbers that are out of date, stuck in another system, or quietly incomplete.
60% never measure this cost
EXAMPLEStock is reordered on a month-old count, so the fast movers run dry and the slow movers tie up cash.
Figures: Workfellow research on repetitive work; IOFM on the cost of fixing one invoice error; Gartner on companies that never measure it. Converted at ₦1,400 to the dollar (July 2026). The pipe narrows at every drain, which is the point.
Case in point
One process. One department.
Paying one supplier invoiceBy handAutomated
What it costs to handle one
$15 to $16₦21,000 to ₦22,400
$3 to $5₦4,200 to ₦7,000
Time from arrival to paid
14.6 days
Hours
Invoices with a mistake in them
39%
Near zero
Typed into the books by hand
68%
Only the odd one
Now count every other job in the business that looks like this one. Most companies have a dozen versions of the same story.
Accounts payable, manual path
Published industry figures: IOFM, APQC (2024 to 2025), Ardent Partners (2025). Ranges vary with how many you handle and how complicated they are. Converted at ₦1,400 to the dollar (July 2026).
Method
How AI finds what nobody wrote down.
Mapping how the work really flows
Works out how a job actually moves through your business, using the dates and times your systems already record. Not the version drawn on the whiteboard. For example, it can show that 30% of orders quietly go through a second approval nobody ever wrote down. (The industry calls this process mining.)
Spotting the odd one out
Finds the strange cases nobody would catch while skimming a report: the same bill paid twice, spending that does not fit the usual pattern, a step that suddenly takes three times as long. For example, one supplier charging 8% above the agreed rate for eleven months.
Joining the dots between systems
Puts your accounts, your operations and your customer records side by side. Normally they sit in separate systems, so a cost in one place can never be traced back to its cause in another. For example, linking a jump in refunds back to one step in the warehouse.
The records your systems already keep
How it works
01 The records your systems already keep
02 AI works out how the job really gets done
03 A ranked list of what costs you most
The headline number
0%
Seventy percent of employees' tasks today could be automated.
Alexander Sukharevsky, Senior Partner, McKinsey / QuantumBlack, 2023
Read this carefully. It is about tasks, not jobs. It means seven in ten of the small jobs people do each day could be handled by a machine. It does not mean seven in ten people are not needed. It also varies by industry: McKinsey puts banking and finance nearer 43%.
100 everyday tasks, shaded by what a machine could do
Automatable
Needs judgement
You do this one task at a time, not one department at a time. That is what makes it manageable.
Where to start
What typically goes first.
Invoice matching and approvalsRead the PDF, match it to the purchase order, route anything over the limit
Finance
Sorting customer messages and first repliesTell "where is my order" apart from a real complaint, and draft the easy reply
Support
Matching records and monthly reportsMatch the bank statement to the books, and flag only the lines that disagree
Finance
New staff and new customer paperworkCollect the documents, check them, open the accounts, chase what is missing
HR and admin
Orders and stock that go wrongCatch the short delivery and the wrong price, and let the clean 95% through
Operations
The time you get back
Choose work that happens often, follows clear rules, can be measured, and does little damage if it gets one wrong. Not the job with the highest profile.
The delta
What changes, measured.
Manual today
After automation
Cost to handle one invoiceDown roughly 80%
₦21,000 to ₦22,400
₦4,200 to ₦7,000
Time from start to finishDays become hours
14.6 days
Under 1 day
How often it goes wrongMistakes become rare
39%
Near zero
Share of the week spent working by handTime given back to your team
25%
Checking only
Bars are drawn against the by-hand figure. Source figures are in dollars: $15 to $16 by hand and $3 to $5 automated, per invoice (IOFM, APQC), plus Smartsheet research on repetitive work. Converted at ₦1,400 to the dollar (July 2026).
Worked examples
What the numbers look like in practice.
Three ordinary businesses, using the published figures above against real volumes. Put your own numbers in and the method still works.
Wholesale distributor
1,000 supplier invoices a month, three people in accounts payable.
LeaksEvery invoice typed in by hand at about $15.50, roughly ₦21,700, and 39% carry a mistake somebody has to fix later.
ChangesThe system reads the invoice, matches it against the order, and sends it for approval on its own. People only handle the ones that do not match.
₦193 million
a year, saving ₦16,100 (about $11.50) per invoice
Professional services firm
12 fee earners, timesheets and client reporting done by hand.
LeaksA quarter of the working week goes on admin: chasing timesheets, rebuilding the same monthly report.
ChangesThe monthly report builds itself, and reminders go out without anyone chasing them.
2 full roles
of staff time given back, out of 120 admin hours a week
Support desk
900 tickets a month, four agents, one shared inbox.
LeaksStaff answer the same handful of questions over and over, and at busy times the first reply takes more than a day.
ChangesMessages are sorted, sent to the right person, and a first reply is drafted for the common questions. Staff check and send instead of writing from scratch.
Minutes
to a first reply on common questions. The tricky ones still go to a person
Worked examples, not real client results: the published IOFM, APQC, Ardent Partners and Smartsheet figures applied to the volumes shown. Converted at ₦1,400 to the dollar (July 2026). Put your own volumes in to get a number you can defend.
Every sector
The same four leaks, wherever you work.
The examples so far came from a distributor, a services firm and a support desk. The pattern does not change from one sector to the next. Only the paperwork does.
SectorWhere it leaksWhat a machine can take over
Non-profit or NGO
Donor and grant reports rebuilt by hand every quarter. Field data collected on paper, then typed up again.
Pulling the same figures into the same report format. Turning collected forms into clean records.
Matching payments to pupils, flagging who has not paid, drafting the routine reminder.
Clinic or hospital
Patient records kept in more than one place. HMO claims prepared by hand, then chased for months.
Assembling the claim, checking it is complete before it goes out, tracking what is still unpaid.
HR team
New starter paperwork, leave tracking, the first sift through a pile of CVs.
Collecting and checking documents, opening accounts, sorting applications against clear criteria.
Startup or product team
The same investor and metrics update rebuilt every month. User feedback read one message at a time.
Assembling the recurring update. Grouping feedback into themes so you see the pattern.
Different paperwork, the same four leaks: doing it by hand, waiting, mistakes, and not knowing what is going on. If your sector is not on this list, the test on the next slide still applies.
Try it now
Would it make a good first project? Score it.
0/5
Tick what applies
Then add it to your list
Your shortlist (0)
Nothing added yet. Score a job on the left, then press Add to list.
Build up a few and print the ranked shortlist to take away.
Held on this page only, nothing is sent anywhere. Print produces a clean ranked sheet. This is the same test used to pick the candidates two slides back.
Sequence
Four phases, in this order.
The order matters more than the software you buy. Most projects that fail jumped straight to phase three.
PHASE 01
Measure what you do now
Before you change anything, write down four numbers: how long one item takes, how often it goes wrong, what it really costs you including salaries, and how many you handle a month. Every claim you make later rests on these four.
PHASE 02
Choose honestly
Rank the options by how often the work happens, how clear the rules are, what mistakes cost, and how hard it is to connect to the systems you already have. Pick what fits, not what looks impressive. The wrong first choice can cost you a year.
PHASE 03
Try it small and measure
Run one job on a fair sample of the work for four weeks. Real results from a small trial beat any forecast, and they settle arguments inside the business quickly.
PHASE 04
Grow it and keep watching
Move on to the jobs next door, keep people on the unusual cases, and watch for it slipping over time. This is something you look after, not a project you finish.
Return
The honest picture on getting your money back.
How quickly you get your money back depends almost entirely on how much you take on at once. Both of these are true at the same time.
061218243036
Months to get your money back
One job, tightly defined
2 to 6 months
Large, joined up across many systems
18 to 36 months
0%
85% of company leaders spent more on automation in 2025. Only 6% got their money back inside a year.
The reason is almost never the technology. Deloitte found the main causes were staff not actually using it, the project quietly growing beyond the original plan, and picking the wrong job to start with.
Sources: Deloitte automation research (2025); published payback figures for business automation (2026). Figures paid for by software suppliers are left out.
What it needs
What has to be true for any of this to work.
Tidy records that talk to each other
This is where most projects stall, and you find out when you start connecting systems, not when you plan. Budget for it on purpose instead of being caught out.
Bringing your people with you
Your team has to trust what comes out, and know what to do with the time it frees up. Leaving this until the end is the most common and most expensive mistake.
People on the unusual cases
Let the machine do the routine work and keep the judgement calls with people. Checking is not wasted effort, it is what makes the automated route worth trusting.
Keep it small to begin with
One job, proven, then the next. “Automate everything” is not a plan, and it is the quickest way to a stalled trial and a board that stops believing you.
Deloitte (2025) found that companies missing their targets blamed people not adopting the change far more often than they blamed the technology.
The takeaway
The businesses that win are not the ones with the most data. They are the ones who know where it is leaking.
Start by measuring one job.
Pick one you run often. For four weeks, record how long it takes, what it costs, how often it goes wrong, and how many you handle. Those four numbers are worth more than any supplier’s forecast.
Where is your money going?
Sources: McKinsey (2023, 2025); Gartner; Deloitte (2025); IOFM, APQC and Ardent Partners invoice figures; Smartsheet and Workfellow research on repetitive work. Converted at ₦1,400 to the dollar (July 2026). Photography under CC0.
Over to you
Questions.
Three worth putting to your own operation on the way out:
01Which job do you do most often that needs the least judgement?
02Do you actually know what one of them costs you, or are you guessing?
03Once a machine is doing it, who is responsible when it goes wrong?
Start with one job and four weeks of measuring. Those numbers are worth more than any supplier’s forecast.
