A model solution

By Steve Towns, Jan. 2006

IT outsourcing may seem like a straightforward way to cut costs and improve efficiencies. But as a growing number of businesses and governments contract out everything from data center operations to networking to help desk support, they're finding that designing these contracts can be incredibly complicated.

“You don’t have to look very far to see massive amounts of public- and private-sector money being put into projects that never deliver the value they promise,” says Richard Taylor, part of an HP Labs research team in Bristol, UK, that’s using sophisticated mathematical models and scenario planning to produce better IT service contracts.

The consequences of project failures can be costly for organizations, which may risk alienating customers, missing deadlines or losing control of projects. The situation isn’t any better for IT service providers, which can take a big financial hit when revenue from these projects is tied to meeting strict performance targets.

Communication problems

Project difficulties usually stem from poor understanding among the diverse stakeholders in large technology undertakings, says Taylor. Agency managers, operations staffs, system engineers, policymakers, budget experts and others may think they’re communicating their requirements — but often they are not.

“When push comes to shove and the system goes live, they find that none of them quite understood what the other was talking about,” he says.

Taylor and his team have developed a concept called “open analytics” to address the shortcomings of the contract design process.

Modeling better contracts

Open analytics integrates input from all stakeholders in a project and uses mathematical modeling techniques to create a clear picture of what the system will do and what the values and costs of those functions will be. These tools help organizations understand the relationships between system performance levels, system capacity, flexibility and cost.

Then, through a process known as Rapid Scenario Planning (RaSP), models are shared among stakeholders to solve design, implementation and management challenges — reducing the communication problems that can doom a project. With RaSP, stakeholders plan their needs into the larger system rather than focusing only on aspects of the project that affect their area.

“We wrapped modeling technologies into a set of processes that bring stakeholders together to explore the technical, social and usage features an organization needs to meet its goals,” Taylor says.

Real requirements for realistic projections

Open analytics remains a work in progress, but HP has used it to craft IT services contracts for more than a dozen projects in the past several years.

The process identifies key bottlenecks, risks and costs. It also generates models that can be used to understand the impact of changes over the system’s lifetime. For example, open analytics accurately predicts the requirements and costs involved in designing a solution that meets everyday processing demands and yet flexibly responds to spikes caused by seasonal deadlines.

The process lets businesses and governments realistically project system capacity and performance demands based on real business requirements. In practical terms, this helps organizations determine if they really need 99 percent reliability for a particular business function or if 95 percent reliability will satisfy requirements at a more appropriate cost.

Rational SLA negotiation

This information is vital for creating service level agreements (SLAs) that form the heart of contracts between government agencies and IT service providers. Historically SLAs have been set somewhat arbitrarily, often because IT benefits such as customer satisfaction and service improvement are hard to quantify. Therefore, SLAs tend to focus on easily measurable factors such as system availability or capacity, regardless of their importance to what the customer actually needs.

This can prompt a vicious scenario that strains or destroys outsourcing relationships: Poorly designed SLAs lead to a service that doesn’t deliver the value a customer wants; therefore, the customer tightens the SLAs even further, which has little effect.

With open analytics, HP can work with government agencies to develop SLAs that reflect real-world requirements and deliver actual business benefits.

“We’re quite happy to perform this modeling activity through what we call rational negotiation of SLAs to develop a solution that is successful for customers and HP,” says Mike Yearworth, another HP Labs researcher involved in the development of open analytics. “There is a very fine balancing act here to make sure we don’t overprovision and come up with an unprofitable solution, or under provision and end up with a solution that doesn’t meet the service level agreements.”

Open solution

Open analytics derives its name from the fact that the modeling process is exposed to everyone involved in a project, which is key to building confidence in the results.

“One of the issues you run up against when you’re using mathematics is the fact that people don’t necessarily understand it, so they don’t necessarily trust what you’ve done,” Taylor says. “You can’t simply say, ‘I’ve got the most fantastic model in the world that will predict everything. Trust me.’ ”

Instead, open analytics produces recommendations based on a better, more comprehensive understanding of customer requirements — and it clearly explains how HP arrived at those conclusions.

“We’re presenting the customer with how HP is going to solve the problem, and why HP believes this is the appropriate route to a solution,” says Yearworth. “We describe the availability and reliability of the infrastructure, and the service organization we’ll put in place. We give customers the models so they can have third parties validate them.”

New approach

Open analytics lets HP capture the human expectations and business requirements involved in IT services contracts, as well as the technology capabilities needed to support them. What’s more, the technique is designed to translate these requirements into terms that make sense to a diverse group of stakeholders.

“ We’ve developed a comprehensive framework,” Taylor explains. “At one end of the spectrum, we have the sorts of models you might see if you went into a social sciences laboratory. And at the other end, we’ve got the hard mathematical facts about availability and reliability of a supercomputing system. We’re integrating across that range.”

The IT industry has been slow to exploit the power of modeling to improve results, notes Taylor. Modeling techniques that do exist tend to focus on narrow aspects of system performance or reliability.

“What we’ve done is different because we’ve used a highly intuitive approach,” he says. “We recognize that understanding reliability in isolation from performance, service structures and customer requirements isn’t terribly useful.”