Price-at-risk: a methodology for pricing utility computing services.

Pricing is a crucial business decision in the life of a product. A minor adjustment in price can dramatically affect the profitability of the product, its diffusion in the market, and its ultimate success. Like many corporate decision processes, pricing is driven partly by rational reasoning, partly by established practice, and partly by "black magic" (not necessarily in this order). In the information technology (IT) sector, pricing falls mostly in two classes. For an IT product, such as a hardware device or a software license upgrade, the development costs are small compared to the high initial sunk costs. For example, the costs associated with the production of a new CPU are small compared to the cost of a chip manufacturing plant. The pricing of IT services has strong similarities with instances of pricing in the retail industry. Hardware equipment is sold on a per-unit basis, but this simple unit pricing is supplemented by a variety of price schedule modifications, such as quantity discounts, bundling, and market skimming (gradual price reduction) and dealing (temporary price cutting). Conversely, for IT services, such as services in outsourcing contracts, a fixed-price contract is dominant.

In recent years, IBM has promoted a third way to provision information services (1) (other companies have made similar proposals). Utility computing services deliver information services when needed, in such a way that customers neither incur the high fixed costs of purchasing hardware and software, nor commit to long-term fixed-price outsourcing contracts. Instead, they receive the service they need and pay only for what they use. Utility computing services represent a departure from the current ways of doing business. On one hand, they feature attributes that appeal to customers: short lead times in service provisioning, high reliability and survivability, customized service level agreements, a reduced learning curve in the adoption of a new service, and easy access to new technology. On the other hand, utility computing services have direct financial benefits for the customer. These benefits come about in two distinct ways. First, utility computing services reduce the risk faced by the customer because the costs to the customer are proportional to the volume of transactions performed during a certain time interval (say, a quarter). These transactions are usually correlated with the number of financial transactions performed during the same interval, and therefore with the revenue stream of the customer. Therefore, the cost structure is tied to the revenue. This reduces the downside risk faced by a customer when the revenue falls below target. In this respect, utility computing services represent a risk management instrument for the customer, similar to insurance.

A second financial advantage of utility computing services comes from economies of scale. Utility computing services are designed to run on a shared infrastructure, in which resources can be dynamically shared among customers. As the number of customers grows, the average resource utilization grows because of the statistical multiplexing of customer demand. As a consequence, hardware costs are sublinear in the total volume of transactions. Similarly, labor and software costs do not increase linearly with the size of the infrastructure. The increase in operational efficiency can be translated to lower prices to customers.

While utility computing services deliver distinct benefits to customers of IT services, they pose new challenges for providers:

* Reduced contract duration. Contracts for on demand IT services have a minimum duration of one year, and this term could be further reduced in the future. This is in stark contrast with the currently typical terms of five to seven years for outsourcing contracts. Previously, monitoring revenues and ensuring they are in line with forecasts (revenue assurance) was handled within each individual contract. In the new model, the challenge is associated with a portfolio of contracts, and one uncertainty faced by the provider lies in the variable duration of these contracts.

* Reduced switching costs and customer lock-in. Although set-up fees and fixed recurring lees are also part of utility-computing service contracts, they constitute a smaller percentage of the cumulative revenue. In turn, this facilitates the migration of customers among providers.

* Uncertain customer demand. The core of the realized revenue is variable; that is, it is proportional to customer demand. With a small customer base consisting of few customers, the provider faces the risk associated with fluctuations in demand. If the customer base is sufficiently large, the fluctuations have less impact on the profitability of the offering, and the provider only faces the risks associated with the industry sector in which the customers operate. In either case, the risk faced by the provider is higher than in the previous outsourcing environment.

* Short life cycles and high sunk costs. Durations of utility-computing service contracts are already shorter than the life cycles of hardware and software products. Together with low switching costs, the short contract duration allows customers to switch to the newest available technology at little or no cost. As a result, the life cycle of utility-computing service offerings will be short and tightly correlated to technological cycles. Within the cost structure of utility-computing service offerings, sunk costs are much larger than the variable costs. Sunk costs include development costs for instrumentation, provisioning, and monitoring of new services.

Thus, new utility-computing services require significant ex ante development and start-up costs in the face of uncertain demand. Compared to the existing pricing practices for IT products and outsourcing services, this is the worst of both worlds. We compare the three sectors in Table 1.

How is pricing affected by the features of an on demand offering? We should point out that three types of decisions are involved: what to price, how to price, and when to price. "What to price" pertains to the set of attributes associated with the service. Should a content delivery service include guarantees on performance, such as maximum packet loss or maximum latency? Should the service offer HTTP (HyperText Transfer Protocol) and SSL (Secure Sockets Layer) transactions separately, or should these two types of transactions be bundled as a single service? Notice that even a homogenous product can be differentiated by posting prices that depend on volume. This is a special form of bundling, in which multiple units of the same product are bundled together. Therefore, pricing is strongly related to the choice of a product (or service) line. However, assigning prices to each item in the product line ("how to price") is perhaps the most important task in the pricing process and is the only one that is performed by the "pricer" alone.

In the following, we address the problem of pricing a service with on demand attributes. We focus on how to price a single unit of a service delivered to multiple customers by a shared infrastructure. We leave the attributes of specific utility-computing service offerings in the background; that is, we take on the problem of pricing with a given set of attributes. Whereas the attributes are important and affect the pricing decision, their impact is indirect and is captured in the price elasticity. There are additional features of a complete pricing strategy that are missing in our analysis. Most importantly, we consider neither nonlinear-pricing nor dynamic-pricing strategies. Although these aspects are important, we believe that their impact on the pricing decision is secondary when compared to price point setting for a unit of service.

First, we observe that nonlinear pricing approaches (also known as second-degree price discrimination), such as bundling and quantity discounts, are not allowed when service resale is permitted (2)--a very real possibility in the case of IT services. Second, there is circumstantial evidence that the demand level of an individual customer is not sensitive to price. For example, the traffic to a popular Web site is independent of how much the Web site owner is paying to the content provider. Similarly, the load on a corporate database system is generated by the company employees, and is insensitive to the price paid by the company to the IT provider. Given that demand is mostly exogenous, the impact of quantity discounts on the pricing strategy is likely to be less important than the selection of the unit price. Moreover, it should be noted that one of the most distinctive attributes of on demand services is the high level of contractual standardization: prices are publicly available to customers, thus ruling out first- and third-degree price discrimination that posits different prices for different customers.