Guides
The monthly numbers an ATM route business owner should watch
Eight monthly ATM route KPIs with a definition, a source and what each one hides, why every figure is per machine, and a twenty-minute review that works.
What to take away
- Eight measures are enough for a route. Each one below has a definition, a source, and a note on what it hides.
- Every measure is per machine. Route averages are how a weak placement stays invisible for a year.
- A measure with no defined source produces arguments rather than decisions. Write down where the number comes from before you start tracking it.
- Two of the eight are about how fast you find out rather than about how well you performed, and they are usually the cheapest to improve.
- Review monthly against your own previous months. There is no published benchmark for this trade worth comparing yourself to.
Most operators track revenue and nothing else, then wonder why a disappointing month has no explanation. The list below is small on purpose: each measure has to be capable of changing something you do, or it does not belong.
The eight
| Measure | Definition | Source | What it hides |
|---|---|---|---|
| Completed withdrawals | Cash withdrawals that finished, per machine, per month | Processor report, checked against the machine | Says nothing about value or about who walked away |
| Value dispensed | Total cash given out at that machine | Processor and your own counts | A rise can mean bigger withdrawals rather than more customers |
| Hours unavailable | Hours the machine could not complete a withdrawal, any cause | Your fault log and visit records | Nothing, if the cause column is filled in; everything, if it is not |
| Time to know | Hours from a machine stopping to you learning of it | Fault log, with how you found out | Hides badly when averaged; look at the worst cases |
| Time to restore | Hours from knowing to working again | Fault ticket opened and closed | Averages flatter you; the host remembers the longest one |
| Contribution per withdrawal | What is left after the host share and variable costs | Your own arithmetic | Changes when the host share or the surcharge moves |
| Cash committed | Load plus dispensed but unsettled, per machine | Your records and settlement timing | The number that quietly limits how fast you can grow |
| Open differences | Count and value of unexplained differences | Reconciliation log | A low number can mean good control or lazy logging |
How to define each one before you start
Two operators can track the same eight measures and disagree about every figure. Settle these first.
Settle definitions before tracking
Does a withdrawal mean completed cash dispense only?
count only completed dispenses
decide treatment of inquiries, declines, reversals
Whether a withdrawal means a completed cash dispense only, and what you do with balance inquiries, declines and reversals. Whether unavailable hours count the whole night at a venue that was closed anyway. Which record wins when the machine and the processor disagree. And whether contribution is calculated before or after your own time.
Write the definitions down once. They matter more than the values, because a definition that drifts makes a year of history worthless.
Read them per machine, always
The single most common mistake in this trade is reporting the route rather than the machine.
Route average vs per machine
Route average
- Hidden weak site
- Concealed
- Hidden strong site
- Concealed
- Where insight sits
- Middle
- Action
- None
Per machine
- Hidden weak site
- Visible
- Hidden strong site
- Visible
- Where insight sits
- Ends of list
- Action
- Repeat conditions
A route average conceals a placement earning nothing, because a strong site carries it. It also conceals a strong site, which is worse in a way, because that is the site whose conditions you should be trying to repeat.
Rank machines by each measure monthly. The interesting information is always at the ends of the list, not in the middle.
What each pair tells you together
Withdrawals falling while unavailable hours are flat means the site has changed, not the service. Withdrawals falling while unavailable hours rise means you caused it. Contribution falling while withdrawals hold means the host share or a cost line moved. Cash committed rising faster than withdrawals means you are funding volume rather than earning from it.
What each pair tells you
Pattern
- Withdrawals fall, hours flat
- Site changed
- Withdrawals fall, hours rise
- You caused it
- Contribution falls, withdrawals hold
- Host share or cost moved
- Cash committed rises faster
- Funding volume, not earning
Diagnosis
- Withdrawals fall, hours flat
- Withdrawals fall, hours rise
- Contribution falls, withdrawals hold
- Cash committed rises faster
That is most of the diagnosis, and none of it requires software. It requires that the measures sit beside each other on one page, which is what the reporting side of the software and metrics guide is for.
The two measures about speed
Time to know and time to restore are the pair operators track least and could improve most cheaply.
Time to know vs time to restore
Time to know
- Usually larger
- Yes
- Improve first
- Yes
- Driven by
- Monitoring, attention
- Cost to improve
- Cheap
Time to restore
- Usually larger
- No
- Improve first
- Later
- Driven by
- Parts, distance, spare
- Cost to improve
- Expensive
Time to know is usually the larger of the two, especially for faults that begin outside working hours, and it is a matter of monitoring and attention rather than of parts. If a meaningful share of your outages were reported by the host, that is the finding of the month.
Time to restore is about parts, distance and whether you hold a spare. It is more expensive to improve, so improve the first one first.
The monthly review, in twenty minutes
Pull the eight measures per machine. Rank them. Look at the top and bottom three of each. Write one sentence per outlier explaining what you think happened, and one action with a date if any is warranted.
The twenty-minute monthly review
- Pull the eight measures per machine
- Rank them
- Look at top and bottom three
- Write one sentence per outlier
- Add one action with a date if warranted
- Check differences log is closed out
- Confirm supporting records are in order
Then check the differences log is closed out and the records that support all of it are in order, which the IRS guidance on business records describes, alongside whatever retention your processor and adviser require, per the compliance overview.
Twenty minutes done every month beats a thorough review done twice a year, because the point is catching a trend while it is still a trend.
What not to track
Uptime as a percentage across the route. Revenue without contribution. Anything reported quarterly that changes weekly. And any benchmark from outside your own business, because the venue mix, host splits and market vary too much for a comparison to mean anything.
The systems holding these numbers hold site and transaction data too, so treat their security as part of the job; the NIST small business quick-start guides are a reasonable place to start.
When the measures change
Growth changes what you can watch. Past a certain number of machines, you cannot read eight measures across every site every month. You sample: all machines quarterly, the worst quartile monthly, and anything that just moved sharply immediately.
That shift is one of several covered in the material on growing a route. What you may say publicly about results is constrained by the claims discipline in the marketing and growth guide.
Common questions
Why not track more?
Because measures that never change an action crowd out the ones that do, and a monthly review that takes two hours will not happen every month.
What is a good withdrawal count?
Whatever makes that placement clear its own costs, which depends on your surcharge, your host share and your visit cost. Any published figure describes somebody else's market.
Should I share these with hosts?
Share the count you pay them on, always. The rest is your business, and sharing costs or contribution invites a negotiation you did not plan.
How long before a trend is real?
Three periods, unless there is an obvious cause. Two points is a line through noise, and acting on it is how operators talk themselves into changes the data does not support.
