Business intelligence -- the collection and analysis of a company's most valuable data -- seems an unlikely task to farm out to contractors. But some companies are doing exactly that. Why? Because they lack the in-house skills needed to perform statistical analysis or maintain a data warehouse. And if you want to turn BI over to the experts, there are more options than ever -- along with the same big worry: losing control of your data quality.
At Canon Information Technology Services in the US, the technical support subsidiary of Canon, the decision to outsource data mining and analytics was simply a case of finding people with the right expertise at the right cost.
"I had been doing (basic) analyses on Excel spreadsheets of our customer feedback data," says Mike Larson, assistant director of product and process quality at Canon ITS. But he wanted more power to do full-fledged data mining and the ability to give support managers online access to reports.
"Our analytic expertise, when it comes to slicing and dicing information, is just not that deep," Larson says. So that left him with two options: buy packaged analytic software or outsource.
After looking at what it would cost to buy the applications and infrastructure and hire additional staff, Larson explains, Canon ITS concluded that it would cost at least twice as much to do the work in-house as it would to hire CustomerSat, a US application service provider. The ASP's clients pay annual hosting fees of about US$100,000 on average, according to a CustomerSat spokesman. Canon ITS signed a one-year contract in April, and the service went live in June.
Each night, CustomerSat uploads data from Canon ITS's two CRM systems, which handle phone and e-mail support. A typical data set includes items such as customer name, description of problem, product purchased and type of interface employed.
"Data quality is a very hot issue for us," says Larson. "The data we export to CustomerSat is just part of a bigger initiative" that involves taking responsibility for data quality across the multiple in-house customer databases. "We screen for things such as duplicate names and bad e-mail addresses," he says. And Larson does some tweaking on his own, such as setting up business rules to ensure data quality.
CustomerSat generates daily, inñdepth questionnaires that go out to a random sampling of Canon ITS customers, and Larson keeps an eye on the process and sample sizes. The responses are loaded into the hosted database and later viewed as reports and graphs via a standard browser.
"A few years ago, you really couldn't do any customization on your application if you chose to go with an ASP," says Guy Creese, an analyst at Aberdeen Group in Boston. "ASPs now give you far more options. It is no longer take it or leave it."
Although Larson is happy with his service provider, he says Canon ITS might eventually bring some of the analytics back in-house. He says the company is "looking at some bigger projects involving more demographic analyses" and is considering buying analytic software from a vendor like Cognos or Business Objects for those efforts.
This hybrid approach to BI is becoming a trend, says Creese. "I think we will see more of this: You outsource where it makes sense and keep the rest in-house."
ETL Is Heavy Lifting
Canon ITS, in fact, keeps the primary customer data source in-house in a PeopleSoft CRM system that feeds the CustomerSat database. Larson says the next step will be to complete the circle and allow CustomerSat to repopulate the PeopleSoft system with data mined from the hosted database.
Even in this distributed architecture, Canon ITS keeps the central data warehouse on-site and maintains primary control over data quality. But that's still a lot of work. Most companies that take full responsibility for their data warehouses need staffs trained in BI and extract, transform and load (ETL) tools, as well as a realistic perspective of the effect that a new project will have on employees and the business.
However, Creese says that it often makes sense to outsource the entire data warehouse. "Maintaining a data warehouse with all the ETL functions is heavy lifting," says Creese. "But often it can be treated as a standard, utility service. So why not outsource that piece and keep the front-end analytics, the stuff that encapsulates your business, in-house?"
That's what Synovus Financial did. "We wanted to mine our bank customer and product data more deeply," says Jeff Kennedy, director of information systems at Synovus, a $19 billion financial services holding company in the US. But, he adds, "supporting a data warehouse, scrubbing and loading all that information, is not something we have ever done in the past, and we had no reason to think it was something we wanted to take on in the future."
In June 1998, Synovus outsourced an operational data warehouse to Metavante, a US provider of technology products and services for the financial services industry. The database supported standard functions like balance reports and audit-letter generation, says Kennedy.
"But we knew we would need a new database design to do more advanced analyses on things such as behavior within a portfolio or loan origination," he says. "And we wanted this to support all 40 of our banks."
In October 2001, Kennedy and his team fleshed out the specs to build a new analytic data warehouse. They met with some resistance from Metavante over the need to replace the data warehouse, Kennedy says, but eventually collaborated on the design and testing. "The result was a database built in a star schema structure -- and still hosted by Metavante," he says. Synovus, however, kept the data mining applications in-house and is using Business Objects software with a browser interface on the front end.
Trust and Partnership
Synovus has moved five of its 40 banks onto the new system and expects to migrate the remainder by next summer. Kennedy says the new data warehouse more than pays for itself.
"We estimate that the analytics we run save us $250,000 over and above what we pay Metavante for the service," he says. But, savings aside, this process involves turning sensitive data over to someone outside the company. Kennedy says it's OK to do that, provided you trust the outsourcer.
"With our collaboration on the analytic data warehouse," he says, "Metavante proved to us that we can work through disagreements together. We built a partnership. Their real value is in updating the data warehouse on a daily basis, and we are confident that our data is in good hands."
One of the things all of these deals have in common is flexibility. For example, in its hybrid arrangement, Synovus could have chosen to outsource the entire BI front end.
"Outsourcing BI, or anything else, is far richer than it used to be," says Aberdeen's Creese. "It wasn't that long ago when your options were to choose between IBM or EDS to come in and run your entire IT shop for $40 million."
Canon ITS chose to keep its primary data store in-house, and the company still keeps close tabs on the quality of data being outsourced. Synovus turned over the maintenance of its data warehouse, but only after the outsourcer went the extra mile to prove itself worthy of the trust.
And neither company has relinquished the ability to monitor data quality. "If you outsource your ability to do this, you have gone too far," says Mark Sullivan, managing director at BearingPoint (formerly KPMG Consulting). "At the end of the day, it is still all about the data. If the data is flawed, no provider can give you good BI or satisfy your SLA (service-level agreement)."
Building Trust
A spirit of partnership pervades an outsourced business intelligence deal at Volvo Cars of North America.
It all started in 1999, when Volvo turned a marketing data warehouse over to Harte-Hanks.
"This is one of a limited number of partnership arrangements we have," says Phil Bienert, manager of CRM and e-business at Volvo. "By this I mean we commit to a multiyear contract."
He says it also means that there's a built-in understanding that goes beyond anything in writing. "We state publicly that Harte-Hanks is our agent of record for customer analytics, and we expect that Harte-Hanks will give us additional support on-site," says Bienert. "We expect that they will supply us with a steady stream of good ideas and that they won't rip us off. You can't put all this in a contract."
Analyzing customer data wasn't Volvo's core competency, so it made sense to find a partner. The automaker is very good at keeping track of its cars, but it needed to "build a customer-centric view of the world," says Bienert. And the company needed someone who could build and host all the pieces, the data warehouse and the applications, mining and analytics.
Bienert and his staff run simple queries and reports on demographics and sales using a dashboard developed at Harte-Hanks. For more sophisticated analyses, such as predictive modeling to support new-vehicle launches, Volvo turns it all over to Harte-Hanks.
But Harte-Hanks' project managers keep the communication pipe open with Bienert and his database manager. It's critical, says Bienert, to make sure that the Harte-Hanks team knows what Volvo's priorities are. "Trends change quickly in the auto business," he says. "What is hot today may not be tomorrow."
Bienert wouldn't talk about the value of the Harte-Hanks contract, but he says Volvo conducted an extensive study before signing the deal. "We determined that it would be significantly cheaper and more effective to go outside," he says.
Harte-Hanks' fees vary, ranging from US$5,000 to US$100,000 per assignment, according to a company spokesman. Assignments vary in duration and scope, but they typically involve things like customer segmentation analysis. Multiyear contracts with clients such as Volvo will involve several assignments.
Dan Rubin, vice president of analytics at Harte-Hanks, says the Volvo project is a popular one at his company. "Our people really want to work on this account. In fact, we have people here who have been on it longer than anyone at Volvo."