Guides
ATM route downtime: time to know, time to restore, and the formula
Defines U, w, c, P, V, and A, gives the downtime formula, splits outage hours into time to know and time to restore, and covers fault classes and spare parts.
What to take away
- The cost of a fault is mostly the withdrawals that did not happen, and that part never appears on an invoice.
- Downtime is worth different amounts at different sites and at different hours. An outage over a busy weekend is not the same event as the same outage on a Tuesday.
- Work in hours unavailable rather than in percentages. A percentage flatters a route that had one terrible weekend.
- The cheapest reduction available to most operators is not faster repair. It is finding out sooner.
- Once you can price an hour of downtime at a site, the decision to hold a spare part answers itself.
Operators price a repair by adding up the part and the trip, conclude it was cheap, and never notice the larger number sitting beside it. This page sets out how to calculate the real cost, in named terms, and what to do with the answer.
The terms
For one machine and one outage:
- U
- is the hours the machine could not complete a withdrawal, from the moment it stopped to the moment it worked again.
- w
- is the withdrawals that machine would have completed per hour, during the hours the outage actually covered.
- c
- is your contribution per withdrawal, after the host share and the variable costs.
- P
- is the parts cost.
- V
- is the cost of the visit or visits the fault required: travel, your time, and any technician charge.
- A
- is anything you had to write off or refund because of the fault.
The formula
The full cost of the outage is:
The term people leave out
U x w x c
usually larger than P + V + A together
at a busy weekend site it can dwarf them
Cost = (U x w x c) + P + V + A
The first term is the one people leave out. At a placement that runs steadily, it is usually larger than the other three together, and at a busy weekend site it can dwarf them.
Getting w right matters more than getting anything else right. Do not use a monthly average divided by hours. Take the withdrawals that machine produces during the specific hours the outage covered, from the processor's records over several comparable weeks. A machine that is dead from Friday night to Monday morning lost its best hours, not its average ones.
Splitting U into two parts
U is not one number, and separating it is where the improvement lives.
U splits into two halves
- machine stops
- time to know: until you are aware
- time to restore: until it works again
- shorten time to know first
U = time to know + time to restore
Time to know is the gap between the machine stopping and you being aware of it. Time to restore is what happens after.
Most operators try to shorten the second and ignore the first, which is backwards for two reasons. Time to know is often the larger half, especially for outages that begin outside working hours. And it is far cheaper to shorten, because it is a matter of monitoring and attention rather than parts and travel.
If you regularly learn about faults from the host, your time to know is your whole problem, and no amount of technician speed will fix it. That measurement sits alongside the others in the material on preventing losses at the machine.
What each fault class costs
Not all downtime is the same shape.
Fault class cost shapes
Fault
- Cash-out
- long if unnoticed
- Note path jam
- short if nearby
- Card reader failure
- longest lead time
- Communications fault
- days, intermittent
- Printer fault
- no withdrawal stop
U length
- Cash-out
- lost withdrawals
- Note path jam
- travel and time
- Card reader failure
- parts and visit
- Communications fault
- underestimated U
- Printer fault
- host relationship
Main cost driver
- Cash-out
- Note path jam
- Card reader failure
- Communications fault
- Printer fault
A cash-out has a long U if nobody notices, a P of zero and a V you were probably making anyway. It is pure lost withdrawals, and it is entirely a planning failure rather than a hardware one.
A note path jam usually has a short U if you are nearby and a long one if you are not. A card reader failure stops everything and often carries the longest lead time.
A communications fault can run for days while looking intermittent. It is the one where U is most often underestimated because the machine appeared to work between failures.
A printer fault does not stop withdrawals at all, so its cost is a host relationship rather than an arithmetic one.
The spare part decision, answered
Hold a spare when the expected cost of not holding it exceeds the cost of holding it.
Hold a spare part?
Does expected loss without the spare exceed cost of holding it?
hold the spare
do not hold it
Expected loss without the spare is the extra hours you would wait for the part, times w, times c, times how often you expect that fault. Cost of holding is the money tied up plus the risk that the part becomes obsolete before you use it.
Do this per fault class and per site, not for the route as a whole. It is what turns a stocking question into an answerable one, and it feeds directly into the classes set out in the inventory and sourcing guide.
Where to spend to reduce the total
In roughly this order of value for money.
Where to spend first
- Monitoring you actually read
- Load sizing that stops cash-outs
- Spares that stop everything
- Better hardware or better site
- Monitoring you actually read, so time to know falls toward zero.
- Load sizing that stops cash-outs, which removes the most common outage entirely.
- A spare unit or the parts that stop everything, so time to restore falls at your best sites.
- Better hardware or a better site, which is the expensive answer and sometimes the only one.
Security work belongs here too: an attacked or tampered machine is an outage with legal and reporting consequences. Protect the machine and its data with the PCI ATM security guidelines supplement. The systems carrying your monitoring and records deserve the same attention, so a sensible start is the NIST small business quick-start guides.
The full picture of what breaks and how to respond is in the repair and security guide.
Keeping the numbers
One line per fault: machine, start, end, cause, how you found out, parts, travel time. That is enough to produce every figure above.
Read it monthly by cause, not by machine. Then read it by machine quarterly because the same fault twice at one site is a hardware decision, and the same fault across four sites is a process one.
The routine that generates these entries is the preventive maintenance checklist. What changes about all of it as the route grows is covered in the expansion material.
Common questions
What does an hour of downtime cost?
It depends entirely on the machine and the hour. That is not evasion, it is the whole point: an average across a route hides the sites where downtime is expensive, which are exactly the sites where you should be spending to avoid it.
Should I offer hosts a downtime guarantee?
Only one you can meet at your worst site on your worst week. A missed guarantee costs more goodwill than the promise ever bought.
Is a service contract worth it?
Compare its cost against the hours it would remove from U, priced at w times c, for the sites it actually covers. A contract with a response window longer than your own is not buying you anything.
How do I estimate w for a new machine?
You cannot, so use the first few months to measure it and treat early downtime costs as unknown rather than as zero.
