Using Cloud Commitments Effectively

Making effective use of cloud commitment products, such as reserved instances and savings plans, to access discounts, isn’t simple. This article explains some of the principles and approaches behind building your commitment strategy in a way that allows you to balance risk and reward.

Highlights within this article include:

• Lessons learned from years in both the cloud and industries with traded markets
• The importance of historical analysis and the pitfalls of relying on it as a sole predictor of the future
• The dangers of fixating on discount percentage rather than overall value
• The value of building future flexibility through commitment strategies involving regular smaller purchases to look a lot like “staggered commitments”

Clear reward, unclear risk

Many of the off-the-shelf tools to support commitment decision-making tend to focus on marrying up usage with commitments in order to capture a discount. These tools do not yet support the right level of discussion about the future risks in that commitment decision. This problem is particularly acute for commitments that cannot be later resold or changed.

All too often cloud customers make multi-year commitments to a high level of their historical usage, with those commitments later getting wasted when cloud usage changes only a short way into the commitment. Similarly, the fear of wastage can lead to an overly conservative commitment approach, which itself is wasteful in not accessing more of the savings potential. Both of these situations require assessment of and then holding of financial risks, whether it is risk of not capturing higher discounts or risk of not incorporating sufficient flexibility. Risk cannot be avoided completely and it is not inherently a bad thing. However, taking risk without understanding, measuring and taking steps to mitigate it when required, definitely is. The result is that cloud users often do not realize the premium they’re paying for flexibility they won’t use, or who are sitting on a risk time bomb waiting to go off.

The trick to rate optimization is to get the right balance for you (commitment term, size, discount, flexibility and risk) so that you achieve a good discount without over-committing or giving up too much flexibility into the future. Building and maintaining a portfolio of commitments to achieve this requires a cloud user to come up with a commitment strategy for their cloud usage, much like the mature hedging strategies built in electricity (and other commodity) markets. 

What does a good commitment strategy look like?

A good strategy depends on some user-specific and market-specific factors. User specific factors relate to forecasts and risk appetite, which are covered below. General market-specific factors include: cloud vendor, commitment types and availability (e.g. AWS has a far wider range of commitment products available to cover EC2 services than for RDS), fungibility, ability to buy and sell with adequate market liquidity, counterparty credit arrangements, product cashflow timings, and stability of the market (eg risk of product withdrawal or change, risk of regulatory change).

Some more detailed market-specific factors relate to the commitments themselves:

• Commitments offered by cloud vendors are typically 1 year or 3 years in duration, so in order to commit with traditional approaches, you need to have a view of your usage into the future over the same time horizon.
• Commitments are easy to buy but often hard or impossible to sell (some can be sold, some cannot, and with a lot of variation between the major cloud vendors).
• Once bought, customers have to pay for any unused commitment in their portfolio (leading to “commitment wastage” when over-committed).

It is also important to appreciate the financial asymmetry between savings achieved with a utilized commitment, and losses realized on any wasted portion of that commitment. For example, with a commitment that attracts a 20% discount to on-demand pricing, the customer either saves 20% or wastes 80% of the on-demand price in a given hour, depending on whether the commitment is utilized or not. Ratios as high as this, of 4 (80% / 20% = 4), between potential losses and potential savings are not unusual, although the discount is usually higher on longer (e.g. 3 year) commitments and so the ratio and financial payoff asymmetry is usually lower on such commitments, and over a longer period of time. This asymmetry means cloud users need to buy commitments prudently to lower the risk of over-committing versus current usage or forecasted usage, especially for commitments they cannot resell, as it doesn’t take much wastage before some commitments become loss making.
 
It’s important you incorporate these kinds of factors, alongside the specific details of your current and forecasted cloud usage into your commitment strategy.

Cloud usage analysis, forecasts and the challenges

Forecasting cloud usage into the future is a vast topic and this article will only briefly address a few aspects of it.

Forecasting models can be built in a number of ways, the most traditional of which is to use historical usage to predict the future. The history may well be useful, for example in identifying levels of persistent usage relative to non-persistent usage, or in identifying intraday, intraweek, monthly, quarterly or seasonal usage patterns that can inform your commitment decision making.

There are limitations with this approach however, because history has no way of telling you about a scheduled ramp down of 30% of your cloud workload in 2 months time, for example. A model trained on history will always have an unavoidable lag (and so error) when usage instantaneously changes as in the situation described. Artificial Intelligence (AI) and machine learning (ML) models would suffer from this same issue if used in isolation. Buying incorrectly from an erroneous forecast might not be a huge problem in some markets where over-purchases can be resold more easily with (hopefully) limited financial impact. However in this respect the cloud market is relatively immature and the ability to reverse a purchasing decision is very limited. Even when it is possible the model incurs a lag while it identifies the usage change is not temporary and it decides a commitment should be sold. The lack of liquidity in the resale market means the sale process itself could come with further costly delays. All this is a best case scenario, as many of the commitment instruments cannot be resold, so a lot of care must be taken in interpreting forecasts and factoring in anticipated changes that will occur within the window of the hedging term.

Your cloud usage persistence is important. Commitments are applied continuously through their term, so if usage turns off overnight, a commitment may waste more overnight than it saves during the day when usage turns back on. This kind of day-night usage profile therefore needs careful analysis relative to the commitment discount in order to work out whether there is a net saving or a net cost of the commitment. It is helpful to calculate the “breakeven percentage utilization” for each resource in order to identify “persistence risks” in the portfolio, and it will be important to understand both historical and anticipated future usage intraday patterns. Equivalent persistence analyses will be useful over longer “seasonality” cycles too (e.g. weekly, monthly, annual).

If you forecast persistent usage to grow into the future, you may well wish to commit to high percentages of your current usage for long commitment terms, and your commitment strategy might be to frequently commit additionally as usage grows.

If you forecast persistent usage will remain stable into the future, you may also wish to commit to fairly high percentages of your current usage, and commit for long and medium terms. You may perhaps wish to have a slightly larger buffer of on-demand usage than in a forecasted growth scenario, as there is higher risk of wastage when closer to 100% committed.

If you forecast persistent usage will reduce in the future, you will likely commit lower percentages of your usage, and for shorter terms, to lower the risk of wasted commitments and the financial loss you would incur.

The good news is that there are commitment strategies that you can steadily step into that will work in all of these situations, and can manage down the risk attached to the irreversibility of commitments, once bought. 

An on-demand buffer

A high percentage of coverage in a commitment strategy generally results in a high overall savings rate (with a few caveats), but the closer one gets to 100% coverage, the higher the probability of incurring some wastage. Due to the asymmetry in savings versus losses, a given percentage of losses has greater impact on financial outcome than that percentage of coverage. With this in mind, cloud consumers often choose to leave a percentage of their cloud usage uncommitted, as a “buffer layer” that attracts zero discount. This reduces the risk of incurring up to 80% losses (in the case of a commitment with 20% discount) on a tranche of their portfolio should usage decline a little. The size of the buffer layer can be tuned, and should depend on your risk appetite, based on your current best view of future plans in the cloud.

With stable usage you might decide that being well committed is to cover 95% of your persistent committable on-demand spend, leaving just a 5% buffer at on-demand prices to protect against risk of wastage of commitments. If you have a low confidence in your future usage or you have known scheduled reductions in usage you may only be comfortable committing up to 50% of your persistent committable on-demand spend, leaving a 50% buffer at on-demand prices as you ramp down usage. Most cloud users will find themselves wanting to be committed somewhere within this large range!

A buffer layer is useful as it lowers the risk of losses in situations where cloud usage declines. You do forego discounts on the buffer layer, but you avoid the risk of immediately incurring losses in that top end of the coverage range. As can be seen, you should tune the size of your uncommitted buffer layer to your needs and re-calibrate as usage evolves. Mistakes here can be costly.

The advantage of declining commitments

In addition to incorporating a buffer layer into your commitment strategy, the concept of a “declining commitment” can further reduce commitment risks. Consider two strategies: (i) a customer purchases 95% of their persistent usage via one single 3 year commitment, and (ii) a customer purchases 95% of their persistent usage via a series of 12 smaller 3 year commitments, one purchased each quarter for 12 quarters. In situation (i), the customer commitment has a single flat profile, for a 3 year period, until it finally drops to zero commitment on a single expiry date. If usage reduces at any point during the 3 year period the remainder of the commitment is at risk of wastage. At the end of the 3 year lock-in the customer faces a single cliff edge decision: commit it all again, or commit differently and accept some increase in price.
 

In contrast, situation (ii) has been created with an inbuilt natural linear decline. There is an 8.33% step down in commitment size every 3 months. This declining commitment can be left to slowly reduce if the cloud user has falling usage. This acts to cap the risk of financial losses – which will only be incurred if usage reductions are larger than the buffer layer and happen faster than 8.33% per quarter. If usage remains steady-state, the commitment in strategy (ii) can be maintained via a new purchase every quarter to replace the one that expires. Should usage increase, this can be folded into the strategy by increasing the size of the quarterly purchase. Similarly, the buffer layer size can be adjusted in the same way.

The declining hedge reduces the size of a cloud user’s total financial commitment to a cloud provider, approximately halving it versus the strategy of buying one single large commitment. One challenge with strategy (ii) as set out here is that coverage remains low as the commitment is slowly built up. However, adjustments can be made to this strategy to fill some of the intermediate commitment gaps with shorter term commitments so that better coverage levels and discounts are achieved early, and which continue to grow over time.

The declining commitment provides a great way to access the larger discounts of both the more specific resource based and the more general spend based commitment types in a way that supports decreasing, stable or increasing usage levels.

Depending on how specific a commitment is, another benefit of the declining commitment strategy is to increase flexibility to evaluate technical changes like chip migrations. For example, at the time of writing (June 2023), a migration from Intel-based AWS EC2 instances to AMD-based EC2 instances delivers greater performance, cheaper pricing and a lower carbon footprint. That has not always been the case in the past, and presumably Intel will respond in the future. The declining commitment strategy allows this kind of technical change to be continually evaluated, and taken advantage of, whenever appropriate.

When available (as not all cloud vendors offer these), you can further fine tune your strategy through the use of convertible commitment types. Although these might come with lower discount rates they do allow you to adjust your coverage to access those discounts with far less term and product risk. The result is that you can control the size of your on-demand buffer far more precisely giving you confidence to minimize it while taking manageable, shorter term risk.

Together these approaches allow you to use all the available commitment types to take advantage of their relative pros and cons in terms of discount rate, term and flexibility to give you a balanced, sustainable approach to cost optimization.

Summary

At a time of high inflation it is understandable that companies seek to procure cloud efficiently to lower costs and remain competitive. However, while a knee-jerk decision to commit an entire cloud portfolio 3 years into the future might initially provide good savings, it also introduces significant risk, which is why we recommend building a balanced commitment strategy that fits with the products available from your cloud vendor.

A balanced approach might incorporate automation of analysis and execution but allow room for human insight and intervention. This allows for appropriate consideration and management of the consequences of commitments on discounts achieved, flexibility, cashflow and risk.

At Strategic Blue we have the added advantage that we can use our purpose built tooling and portfolio of usage under management (the “portfolio effect”) to open up wider options for discount capture and risk mitigation for each company than any one might achieve on their own. We work alongside our customers so their technicians and FinOps teams can concentrate on the other aspects of cloud cost optimization that need their specific internal knowledge. They put this knowledge of their technology needs and organizational strategy to work in managing what they use in the cloud. We bring our expertise to bear on the rates they pay for that usage so that, working together, we derive the best value from their cloud budget.

Get in touch if you would like to learn more or see how we might help.