You have to have an uptime guarantee. How you setup your SLA is what can cause issues though. PLus so many factors such as should you honor their monitor's reading or what.
When I was a newbie host back in the day I made an SLA where if uptime fell below 99% in a month and the customer noticed it of course, I would refund a minimum of 10% of their monthly fees. Now this only was applicable if they had not reached their ddos protection threshold and they had live graphs of that to see.
Boy what a mistake, with apache/webserver restarts, network farts (sorry, no other word I can think to describe when your network or routing takes a dive for a second to a few minutes) and general server reboots and such you will be lucky to keep at 99.99. God forbid someone gets ddosed on your server and its on a shared ip.
Anyway, when I did this, since my average plan was about $90-379 for ddos protected hosting so some customers (not all but some) would setup monitoring on 30 second delays or even less, watch their attack stats like a hawk, The minute they seen the SLA failing they would call demanding refunds or account credits.
Not many abused this but enough did where I had to ask a fellow web host about the issue. He said it wasnt possible with any type hosting and I found that out later.
So I guess the point Im trying to make is, have an uptime guarantee, NEVER a 100%, even if you are running load balanced cloud/cluster or whatever. And try not to promise too much on SLA.
Its really hard to say what to offer on an SLA but now what I do is give small discount on following month depending on how bad the SLA was broke.
Id be interested in seeing some of your guy's SLAs. Since I got out of ddos protected hosting I have only once had to give discount for sla cause the clients ror app kept failing, which it turned out to be their fault in the end lol.