Every platform fails sometimes
Software breaks. Networks fail, a data feed stops, an update goes wrong. A provider that tells you its platform never goes down is describing a hope.
So the sensible question is not whether a platform will fail. It is how often, at what hour, for how long, and what the people behind it do when it happens. Those answers separate providers far better than a feature list.
Turn the percentage into hours
Uptime is quoted as a percentage because percentages look reassuring. Convert them.
The Wikipedia entry on high availability carries the standard table. At 99 percent availability a service can be down about 3.7 days in a year. At 99.9 percent it is about nine hours a year, or roughly ten minutes a week. At 99.99 percent it is under an hour a year.
The gap between those rows is the gap between a platform you forget about and one your players complain about. Three and a half days of downtime in a year is a lot of Sundays.
You can do the sum yourself for any figure. A year has 8,760 hours. Multiply by the percentage that is not uptime. For 99.7 percent, that is 0.3 percent of 8,760, or about 26 hours across a year.
Ask how the figure was measured
Three questions turn an uptime number from marketing into information.
Over what period? A figure for last month says little. Twelve months is a reasonable minimum.
Measured by whom? A provider's own monitoring and an independent monitor can disagree. Neither is wrong, but you should know which you are reading.
Is it a record or a promise? A statement of what happened is different from a commitment about what will. A formal commitment usually sits in a service-level agreement, which Wikipedia describes as an agreement between a service provider and a customer that defines particular aspects of the service, often including uptime and what follows if the provider misses it.
For our own part, this site states 99.7 percent uptime over the last 12 months, on our own monitoring. That is a record, measured by us. It is not a promise about next year, and we would rather say so than have you read it as one.
When matters more than how much
Thirty minutes of downtime at four on a Wednesday morning costs a book almost nothing. The same thirty minutes in the hour before the Sunday games costs bets that never get placed and players who text you asking what is going on.
An annual percentage hides this completely. Two providers with the same figure can have had very different years.
So ask when the downtime happened. A provider that plans its maintenance for the quietest hours of the week and can tell you so is showing you how it thinks.
What happens to the bets
This is the part that affects your money, and it is the part agents forget to ask about.
Bets already accepted. If a player placed a bet and received a confirmation before the outage, does it stand? It should, and the provider should be able to show it.
Bets in the middle of being placed. A player pressed the button and saw nothing. Is there a record either way? The rule matters, because that player will be sure he bet if his side wins and sure he did not if it loses.
Live bets. In-game betting depends on prices that move by the second. Ask what happens to live markets when the feed behind them stops.
Grading. If results arrive while the platform is down, when and how do the bets get graded, and can you see that it was done correctly?
Get these answers before you need them. For the wider list, how to compare providers covers the rest of the bill and the platform, and see what a provider should put in writing.
What a good provider does during an outage
You are judging behavior more than engineering.
- It tells agents quickly, without waiting to be asked.
- It says plainly what is affected and what is not.
- It gives a time for the next update and keeps to it.
- It has a person on the phone. This is when a ticket queue is useless.
- Afterwards, it explains what failed and how accepted bets were treated.
Silence is the worst sign. A provider that goes quiet during a failure leaves you to face your players with nothing to tell them.
What you can do on your side
You cannot keep someone else's servers running. You can make an outage smaller.
Know the rules in advance. Ask the questions above and write the answers down.
Keep your own records. Know each player's limits and where his weekly figure stood at the last report you read. If the platform is unavailable when the week closes, you can still tell a player where he stands.
Tell players your policy. One sentence is enough: if the site is unavailable, bets already confirmed stand and nothing new is taken until it is back. Say it before it happens.
Do not take bets by text during an outage. It is tempting, and it creates bets with no grading, no limit check and no record both sides agree on. That is how a thirty-minute problem becomes a three-week argument. How a week closes depends on every bet being in one place.
Questions to ask before you sign
- What was your uptime over the last 12 months, and who measured it?
- When did the downtime happen, by day and hour?
- When do you schedule maintenance?
- What happens to accepted, pending and live bets during an outage?
- How will I be told, and how quickly?
- Who answers the phone when it happens, and in which languages?
If you are moving from another provider, add one more: how you would move your players across without losing a week.
An example
Take an agent, call her Noemí, with fifty players in Punta Arenal. Her platform is unavailable for forty minutes on a Sunday, starting twenty minutes before the early games.
Because she asked beforehand, she knows that confirmed bets stand. She sends one message to her players saying so, and that nothing new can be placed until the site is back. Two players ask her to take a bet by text. She declines both.
When the site returns she checks three accounts against the confirmations those players sent her. They match. On Monday the provider tells its agents what failed. Her week closes on the usual day.
The outage cost her some bets that were never placed. It did not cost her a dispute, because the rules were known before anything broke.
BookiePerHead is software and a support line. It takes no wagers and holds no player money, so during any interruption the book, its limits and its players remain the agent's.