5 ways to forecast your cloud spend
Why do you need to forecast?
In every company there is a need to plan for future flows of money, and IT is no exception. Before the Cloud, IT spend was driven by people costs, hardware costs, and maintenance costs, each of which is stable under change management (i.e. you need authorisations to change things).
Then, one day cloud is used and the costs are less predictable. What was a server, with a buying and maintenance cost, becomes a dynamic set of virtual machines, with different network, storage, size that can change on an hourly basis. While resulting in better price and performance, the cloud complicates forecasting a lot. There is no one-size-fits-all approach and so IT departments are left to become creative.
In this article I will investigate five cloud spend forecast methods with their pros and cons to help you.
Disclaimer: Unless you are in a very boring business where nothing changes a forecast will always be wrong! Business life will happen, and no model can account for every aspect of reality.
1. My usage is constant and planned to stay constant
This approach is useful when the forecast timeframe is short (e.g. one week) or when IT is not involved in the operational aspects of the business. For example, when IT only provides internal services like email and a website, and not used to deliver to customers. Another example is automated manufacturing, where IT is the brain of the machinery and unless new machines are bought (or removed) nothing will change. -Summary: Take the last period consumption and assume the same for the future.
- Example report: Last month, we spent $10,000 on the cloud. There are no changes planned next month, so we expect to spend another $10,000.
- Best suited for: Predictable workloads over the selected period.
- Effort level: Almost null. Look at the last number and use that.
| Pro | Con |
|---|---|
| Immediately available | Extremely sensitive to change |
| No effort needed | |
| Very accurate when nothing changes |
2. Take the last period and add/remove X%
Many companies have asked me for help with their yearly cloud forecast. When I’ve asked them how they were doing today, the answer was we add 20% to last year’s spend. I guess this is part of a ritual, where each department asks for more than last year and finance provides a subset of the request each year.
- Summary: Take the last period number, add X% (or any other value).
- Example report: Previous year cloud costs were $100,000, based on a business growth of 20% we estimate next year to be around $120,000.
- Best suited for: Providing a single number to feed a financial model where the cloud costs are directly in correlation with business growth.
- Effort level: Low as the % to add (or remove) should be provided by a financial model.
| Pro | Con |
|---|---|
| Very fast | Very generic |
| Very easy | Very static number (when the number is not reviewed often) |
| Backed by business data (kind of) |
3. Using AWS Cost Explorer Forecast (or other cloud vendor tools)
When accessing the AWS cost explorer, one of the first pieces of information presented is a forecast for next year, based on last year data and elaborated by a huge-sized company. The graph is very good-looking, and is easy to add into a document, but I would not recommend you to use it to make decisions.
- Summary: AWS and other cloud vendor forecasting tools.
- Example report: Based on our past data and using the algorithm provided by AWS, we can estimate our cloud costs to be between $x and $y.
- Best suited for: A fast, backed-by-others forecast easily presentable to non-IT.
- Effort level: Low and usually self-service.
| Pro | Con |
|---|---|
| Available immediately | Based only on data available to the cloud vendor, so no business connection |
| An innovative company offers the results | Future scenario modelling requires the use of additional tools (e.g. Excel) |
| The graphs look great | The methodology is unknown, therefore difficult to justify |
| Provides minimum, maximum and expected values | The accuracy drops fast |
4. End-of-month forecast
Like in a car, sometimes what you need is the current speed and an idea of where that speed will get you over the next few hours. For cloud spend, the equivalent is the end-of-month forecast.
- Summary: Create an End-of-month forecast based on the already known monthly consumption and projecting the last 24hours until the end of the month.
- Example report: The most recent optimisations, introduced on the 14th, have reduced the monthly projected spend by 15%.
- Best suited for: Proving the value of individual changes, and to keep an eye on the impact of daily activities on the monthly budget.
- Effort level: Significant as the methodology is not available by default and requires extracting data from the cloud vendors and analysing it.
| Pro | Con |
|---|---|
| Available in the Strategic Blue dashboard | Needs to be implemented |
| Great to show the impact of cloud initiatives | Data management and ETL knowledge is a requirement |
| Capabilities to follow the monthly budget on a day-to-day basis | Understanding the cloud vendor’s financial reporting files is a must (e.g. AWS Cost and Usage Report) |
5. Use business data (sales projections)
When there is a culture of sales forecast in the company, IT should leverage the good work of other groups and identify a connection between the business element of income (e.g. customers or transactions) and cloud costs. For example, identifying that each transaction costs 1 cent allows you to use the sales forecast to build the cloud cost one.
- Summary: Identify a cloud-cost per unit of work (e.g. $/user or $/transaction) and map on the financial model used in the business.
- Example report: Our cloud infrastructure has a fixed cost of $100,000 per month. Each customer costs 12 dollars per month. By using the planned growth in the sales forecast, we can project the following monthly cloud-cost .
- Best suited for: A dynamic representation of the cloud costs mapped on the company’s units of business measure with forecasting based on a well known financial model.
- Effort level: Important, requiring expert consultants.
| Pro | Con |
|---|---|
| Converts IT slang into money, making it easy to understand for other departments | Complex to implement |
| Easy to recalculate, even daily | Requires bespoke modelling |
| By mapping cloud costs to the company units of goods sold, the results are more comprehensible to non-IT | A change in the business model requires an update in the algorithm |
| Any IT effort to reduce the cost per customer is made visible |