Although I haven’t looked at this deeply I know that AWS, and it appears Azure as well, only offer subscription-based support models.  On-premises Microsoft offers both subscription-based models and per-incident support.  So does per-incident support make sense in the cloud?  I really want to hear from customers across cloud platforms, although my musings are clearly AWS-focused.

AWS does not currently offer a per-incident support option, and spending a lot of time in the Amazon RDS Forum one sees the potential for them to offer one. Keep in mind that the forums are targeted at community discussions and not individualized technical support from AWS. If you check the comparison of AWS Support plans you see that Basic Support, the free option, does not include Technical Support. So customers post technical support requests in the forum and, most of the time, no one from AWS responds.  On top of that, whereas there is a strong culture of community participation in (both company and third-party) forums on Microsoft offerings, that culture is not as well developed around AWS.

AWS does offer an inexpensive support plan for Developers, though it can be used to obtain support for production systems as long as you can live within its limits (e.g., only email, respond only during business hours, 12 hour response time on impaired systems). What you can do if you have that once/year kind of need for technical support is sign up for developer support, open a case via email, then cancel support before the next billing cycle. If your monthly spend with AWS is under $1000/month ($966 actually) then you effectively have a per-incident support price of $29. If you are spending over $1000 your effective per-incident price is 3% of your total AWS bill for the month using this “subscribe only for a month” technique.

For more critical support incidents you could sign up for AWS’ Business support for a month then cancel that.  If you do this for a single incident it effectively costs you $100/incident if your spend is under $1000 a month, but escalates rather quickly as it is 10% of the monthly bill.  Actually the 10% applies to the first $10K, with higher spend levels getting volume discounts, but we will ignore that for this discussion.  Using the current Business support subscription pricing in the “subscribe only for a month” technique to obtain per-incident support only makes sense for smaller customers.

Figuring out if “subscribe only for a month” pricing is an adequate substitute for a true per-incident offering could use a benchmark, and we have one.  Microsoft charges $499 per incident for its on-premises (non-consumer) products.  If AWS were to price per incident support comparably you would need to be spending over $16,700/month for that to be cheaper than just signing up for Developer Support for a month then cancelling.  The Microsoft per-incident offering is closer to AWS’ Business Support, but even there the “subscribe only for a month” technique is cheaper as long as your monthly bill is under $5000.

To put this further in perspective, how many organizations are spending more than $200K/year (the breakeven point for our per-incident benchmark vs one month of Developer Support) on a cloud provider?  Or $60K/year (the breakeven for Business Support)?  I’m going to speculate that for 99.9% of customers the “subscribe only for a month” technique is a better option than per-incident, based on the chosen benchmark.  So if this is true in the cloud, why isn’t it also true on-premises?  On-premises support subscriptions are generally annual (or longer) contracts, so you can’t subscribe for a month then cancel.

Is Microsoft’s $499 benchmark the right comparison?  On the surface it seems expensive, but that is because it is one size fits all.  From things that can wait a day and then resolve with 15 minutes of effort to production system down situations that need fast response to complex issues that take person-days or weeks of effort across multiple expertise domains. That’s also why support subscriptions are priced as a % of total spend, because the more complex the environment the more the requests about product/service X turn into debugging a problem across product/service X, Y, Z, and the customer’s app that requires multiple engineers to be engaged.   Services pricing reflects the overall cost structure of delivering services, not the cost of your (potentially really simple) request.

Vendors really hate the idea of unbundling support because it can create tremendous customer dissatisfaction.  You take your dog in to have its teeth cleaned for a simple published price.  Then they tell you it is highly recommended that for a dog of its age they do bloodwork and an EKG before administering anesthesia.  You choke on the price, but you’d hate for your dog to die just because its teeth needed cleaning, so you say ok.  Then they find something in the bloodwork that indicates there could be a problem with the liver.  You are already paying for the anesthesia, so do you want them to x-ray the liver as well.  There in the x-ray is a growth, should they biopsy it?  Etc.   Your $100 teeth cleaning is turning into a multi-thousand $ bill.   The veterinarian you love has just given you the choice of your kids hating you because you didn’t try to save their dog, or being unable to afford their college education.  You can’t blame the veterinarian for the cancer, but you love them less because of the process they put you through.  So it is really hard for a vendor to say something like we will sell you support for your database server, even if you don’t want support for everything else, because they don’t want to come back to you in the heat of a situation and say the problem spills over to the application server and you don’t have support for that.  It will cost you $X for us to continue the effort over there.  Yeah, right.

Low cost vs. reality of support vs. customer satisfaction is a conundrum.

So is there a per-incident model that would work for customers and fit in what I called the “reality of support”?  One rational per-incident offering to explore would look something like current monthly business support pricing with a cap. So it would be greater of $100 or 10% of total spend up to a maximum of some amount. Using Microsoft’s per incident price as a proxy for not having AWS Support cost data to work from, let’s say that cap is $500/incident. Is that the kind of per-incident support that would be attractive to AWS customers?  How about customers of other cloud providers?