Teradata exec. tackles consolidation conundrum

Information is the life blood of many corporations, but its importance dwindles to almost nothing if it can’t be accessed easily.

NCR Corp.’s data warehousing division, Dayton, Ohio-based Teradata, launched a program late last

year to help companies consolidate disparate data onto a single warehouse. Rob Armstrong, Teradata’s director of technical marketing, was in Canada earlier this month to help promote the program and find some new customer prospects. “”They’re asking questions around the feasibility of bringing systems together, because they have tried to do it in the past with other databases and they have failed,”” he said.

Computing Canada spoke with Armstrong about these and other customer questions, and about the paradox around CRM initiatives: the desire for detailed customer information can quickly outpace a warehouse’s ability to deliver it.

Computing Canada: Are customers looking to consolidate their databases?

Rob Armstrong: The idea is that people typically have data marts that are application-specific sets of data. They may have a multitude of those. When somebody wants to look at the data across the applications, it’s a very significant chore to go find the data, format it so it’s consistent and then bring it together to a single user. One of our clients said that their process is at least three weeks. The Teradata solution is that first you need to consolidate the data together. You bring that data together, you make it consistent, and then you allow the usage against it. You don’t do it at query time, you do it before hand, so you migrate the data marts into a true data warehouse. What most people try to do is skip the hard part of data consolidation.

CC: How do you make sure that all the data is scrubbed before it’s brought together?

RA: Clearly all of the data cleansing and transformation goes back to the business rules and business definition of the data. The business owns the data. IT’s just there to make sure the data is there.

The business has to define what the data elements are. Is the UPC on a product the same thing as an item number to a retailer? Well, sometimes it is, sometimes it’s not. The big chore is getting the business involved to define what the data is to begin with. Once that happens, (the next step) is where do I have it? When I bring them all together, they will all be item numbers. In my model, I am transforming the data definitions. That’s the hardest part of data warehouses — understanding what the data is.

One customer I worked with had four different definitions of what net sales is. Everybody wanted to look for net sales, so we needed to get a common definition of how that’s calculated. If I can’t define net sales consistently, then I’ll never get my business metric to be consistent.

CC: How long does it usually take to consolidate all this data in one warehouse?

RA: That’s the $64,000 question. You do this cyclically. The first cycle is probably a three to six month cycle. Let’s say you have 15 data marts. You may start with three that are closely related, but they each have some distinguishing characteristics. I may take the marketing and promotions data marts and the inventory data mart, because they all centre around items at locations.

After that three to six months when I’ve got this core set of data, I look at the other data marts and say, ‘now which data mart brings me the most value and has the most leveragability?’ It has a lot of the core data — maybe 70 per cent is the same — so I need to bring in the other 30 per cent which is unique data. Now I have absorbed the data mart.

CC: So the warehouse is operational within six months?

RA: Yes, but with one caveat. What’s the usage interface? You shouldn’t have a nine-month long project to build an interface to get to the data. There’s lots of third party tools available — like Cognos, Micro Strategies — that should be able to go after your data warehouse very quickly. Within a few months I should be able to ask questions directly of the data — not just get predetermined reports, but they should be able to ask questions about the answers they’re getting. If I get an answer that says sales were down 10 per cent, I can ask, ‘Why sales were down 10 per cent?’

CC: When will the customer see some return on investment on their warehouse?

RA: If I can have the system up and running in six months, the customer should be able to see return within another three to six months. Some actions I take can have returns very quickly, like changing inventory patterns, changing a national advertising campaign may not happen very quickly.

If you’re not getting a return on investment within a nine-month window, you did it wrong. Either you set up the warehouse wrong or you didn’t provide the capabilities the users’ needed to a great enough degree.

CC: There have been some fairly high-profile data warehouse installations that have failed. What are you telling customers so they don’t make the same mistakes?

RA: The problem is, people are using technology that was not designed for cross-functional, ad hoc or unknown analysis. What they’re choosing is technology that they already have in-house that is for their online transactional processing systems which are well-known, well-understood queries. Now when they go to an environment where it’s unknown queries with unknown data demographics with unknown complexity, they don’t know how to tune for it. In order to get the query to run I have to tune for it. But I have not just one group but three groups with conflicting sets of requirements. What typically happens is, the project fails. It fails because the time it takes between when I have an idea and taking action on that idea is so long that the action is no longer valuable. If I can’t make an inventory decision while I’m running out of inventory, it’s a moot point.

CC: What has the growth of CRM in the last couple of years meant to the data warehousing industry?

RA: The growth of CRM has spurred the failure of a lot of data warehouses. When most people think of CRM they think of call centre management, outbound marketing and contact management. When you start getting into CRM you also understand there’s so much more to relating to a customer that is outside of those three aspects.

CRM is requiring data more frequently and at a (deeper level) than they currently have available. When my customer contacts me, I want to be able to contact him the next day with a follow-up, not the next week. If the customer called the call centre, went to the Web site and came into the store on the same day, I should know all three of those things by the next morning at least.

The systems that were defined around functional reports cannot drive that. People are starting to look around saying, ‘My CRM tool needs to deliver data to me daily, but my data warehouse only gets it weekly.’

It either fails because they push it to hard, or they accept the compromise of the data warehouse and start going further outside of it. Now the value of the data warehouse is driven down to zero because nobody uses it.

CC: Now that Teradata is its own division within NCR, has that given you a better focus on pursuing the data warehousing market?

RA: Well, we have more autonomy over what we do. We still have the retail and financial groups of NCR but now that Teradata is on its own, we focus more on the data warehousing aspects of a business as opposed to trying to tie it to a financial or retail-specific solution. The other thing it gave us was our own support infrastructure, like the warrantee environments and maintenance processes. We weren’t just another hardware platform — this is a Teradata data warehouse platform. It gave us more autonomy and more control over data warehouse-specific issues.

CC: What are your biggest verticals right now?

RA: Retail would have to be one of them. The second one is telecom. If you look at some of the largest telecom companies in the world on Teradata, they’re wireless, land line, long distance, ISP providers. Once you get past retail and telecom, then you’re into what we call national accounts, which is manufacturing, transportation, utilities, health care and things like that. While they all are a significant piece of our business, I think if you looked at the verticals people would recognize they would be retail and telecom. I have yet to see a vertical where data warehousing doesn’t apply. If data warehousing applies, we’re probably in there.

Comment: info@plesman.com

Share on LinkedIn Share with Google+