Council Post: Converting ‘Cloud Credits’ to ‘Currency’ – Estimating Operating Costs for Cloud Based (SaaS) Analytics Technologies

Technology budgets are typically prepared once a year to align with the annual operating plan. However, underestimation in these budgets can lead to significant cost overruns, impacting the availability of project funds and the ability to deliver projects effectively. Cloud computing platforms are increasingly attractive due to their scalability—allowing resources to be scaled up or down as needed. However, this flexibility comes with a challenge: unexpected costs due to idle instances, temporarily spiked usage, or other unforeseen factors can quickly escalate expenses.

One of the primary challenges IT and Analytics leaders face is estimating operating costs for cloud-based technologies. This is particularly difficult due to the ‘credits’-based resource provisioning model used by many technology and platform service providers, which often combine the consumption of various software services such as data ingestion, data engineering, analytics, and data transfers. Unfortunately, cloud service providers may not provide a clear operating cost estimation model to directly translate the consumption of these credits into currency.

Service overages tend to be more expensive than the standard unit prices of the technology SKUs. If these overages are not anticipated and managed carefully, they can cause significant budget management issues. Therefore, developing estimation methodologies that enable IT and Analytics functions to accurately estimate cloud-based technology operating costs is critical.

Ex-Ante Cost Transparency: The Need for Estimation Models

Effective IT management and investment decisions require ex-ante cost transparency, which means anticipating and planning costs before they are incurred, rather than after the fact. This approach necessitates the development of ex-ante cost estimation models that standardize methodologies for the business processes enabled by IT and Analytics. By doing so, organizations can align technology functionality with specific process elements and assess their technology needs accordingly.

The first step in this process is understanding the operating cost units (credits) offered by the cloud technology solutions provider. These units must be aligned with a standardized business process methodology to ensure accurate cost estimation. For example, understanding the cost implications of running a specific analytics workload in the cloud, including the data ingestion, processing, and analysis phases, is essential for precise budgeting.

Existing Cost Estimation Models and Methodologies

Several existing cost estimation models can serve as a foundation for developing a robust methodology for estimating cloud-based technology operating costs. These models typically cover:

Service Usage Data: Measuring costs for infrastructure, scalability efforts, and additional service usage for systems already in operation.

Usage Characteristics: Analyzing requests, interactions, and user amounts to gauge system demand and predict costs.

System Architecture Analysis: Examining planned and existing systems to identify the most comparable systems, which helps in estimating costs by using similar usage and cost indicators.

Data Pattern Analysis: Analyzing historical data patterns to predict future costs.

A simple yet effective approach for estimating cloud analytics software operating costs involves focusing on key factors like CPU, network, and storage usage. 

For a marketing technology, the cost estimation process might involve the following steps:

1. Collect Data: Gather data on historical usage and costs.

2. Blend & Unify Data: Combine data from various sources to create a comprehensive view of resource usage.

3. Analyze: Perform a detailed analysis to identify patterns and trends.

4. Predict & Derive Inferences: Use predictive analytics to estimate future costs based on past trends.
5. Segment Customer Profiles: Identify different customer segments to understand their specific usage patterns.

6. Communicate: Share the insights and cost predictions with stakeholders.

For example, a hypothetical marketing technology use case might estimate credit consumption based on the above methodology as follows, with adjustments made for anticipated changes in usage patterns.

Example 1: Estimating Costs for a Marketing Technology Use Case

Consider a company with two business units (BU1 and BU2) with customers as follows:-

Based on the marketing technology steps provided previously, the operating cost of the SaaS solution may be estimated as illustrated below.

Figure 1: Illustrative operating cost estimation for marketing (SaaS) technologies

This methodology is linear in nature. It may be improved considering business demand, influences of relevant events and seasonality.

Example 2: Estimating Costs for a Generative AI Solution based user-interaction solution

Consider a popular generative AI solution where the ‘credits’-based charge-out rate of a foundational AI model is linked to ‘capacity units (CU)’ of the software SKU. This adds a layer of complexity to the estimation process. Here’s how it might work:

Suppose a generative AI SKU has a cost structure where:

• The input prompt (per 1,000 tokens) consumes 500 CU seconds.
• The output completion (per 1,000 tokens) consumes 1,500 CU seconds.

If each prompt request has 2,000 input tokens and 500 output tokens, the price for one prompt request can be calculated as follows:

Assuming a customer has an SKU costing Rs. 250,000 per year, which provides 16,588,800,000 CU seconds over 30 days, this can support approximately 947,931 prompts per month. With an employee base of 500 interacting with this SKU, the average number of prompts available per employee per month is 1,895. This number can be easily exceeded with increased adoption, leading to cost overages. If multiple use cases rely on the same SKU, cost estimation becomes even more complex. 

Improving Estimation Models with Business Demand and Seasonality

The examples presented above provides a linear and directional estimation method of cloud-based analytics service costs. The following factors may be included in cost considerations depending on the extant situation & nature of analytics operation:-

• Model fine-tuning
• Data pipeline and database operations
• Data integration and APIs
• People costs including architect / designers / analysts and developers
• Training, Development and Support

To improve these estimates, it is also important to account for business demand fluctuations, events, and seasonality. These factors can significantly influence usage patterns and, consequently, costs. By integrating these elements into the estimation model, IT and Analytics leaders can develop more precise budgeting strategies that mitigate the risk of unexpected expenses.

Conclusion

Precise estimation of cloud-based analytics technology operating costs is crucial for aligning technology budgets with business processes and analytical outcomes. The methodologies outlined in this article offer a foundational approach to estimating costs, with a focus on potential overage costs. However, this is just the beginning—continuous refinement and feedback are necessary to improve these models.

Implementing a robust FinOps strategy is expected to further enhance an organization’s ability to estimate, track, and control cloud technology costs effectively. By doing so, businesses can leverage the full potential of cloud-based technologies while maintaining control over their budgets.