I have been talking with folks a lot about the issue of return on investment (ROI) when it comes to Business Intelligence (ROI) and, quite timely, I saw this report from Knightsbridge about the ROI of Business Intelligence when I was scanning BusinessWeek magazine. Now I don't want to imply anything about the quality of work Knightsbridge does for its clients as I have absolutely no data but the study left me saying "yes, but..." as it defined success with BI as "getting more of a return that I expected". No-one was asked for hard ROI numbers, just if they had met or exceeded their expectations. Now this is a little self referential - I say it did better than I expected it to - and fails to account for a known issue with BI projects - that no-one defines exactly how an ROI is going to be generated! The report had some good advice on making BI successful but failed, in my mind, to show any kind of real BI.
While I was pondering this I saw a report in ComputerWorld on BI Home Runs. As far as I can tell these "Home Runs" were selecting by the magazine, and presumably by the BI vendors who participated, as exemplars of the BI approach. All I could think of as I read these was "That's it? That's the best you could do?"
These cases were not impressive to me as examples of the best of BI. That's not to say that the companies profiled had not done a good job at what they were attempting - they mostly had - but that the basic BI approach was either flawed or at least very timid. Let's consider each of them in turn.
1-800 Contacts: Fine-tuning a Sales Strategy
A classic example of managing metrics and displaying them. But where are the analytics to derive insight from this information? How could best practices and policies be overlain on top of these analyses? Why would you not focus on the operational decision - next best action for the call center - and give the call center agents the ANSWER not just data about the question? An EDM approach would have focused resources on giving the call center agents better offers when an offer was appropriate, better non-offer actions when that was called for, and would not have required the agent to worry about that thus enabling a focus on the conversation and improving the customer experience.
Alliant Energy: Viewing Fine Details in Financial Data
A good example of a real BI one as this is strategic management - applying insight to a low volume decision
Allstate Insurance: Getting Fast Help to Those in Need
This could be an example of BI but what about using in-line analytics and rules to process the claims at once? Why manage them in a data warehouse first? Why not just focus on the operational decision, pay the claim as an emergency or not, and use location intelligence, rules about what classifies as an emergency and perhaps analytics to do this as the claim arrives?
BNSF Railway: Tracking Finances on the Rails
This sounds like a nice dashboard and is a good performance management example.
But what about improving operations - are there ways to use this information and insight automatically to route trains, schedule maintenance, manage staffing so as to reduce costs? If you applied EDM to this problem you would see a much greater return as it would improve the large number of small decisions a railway takes.
Dreyfus: Testing a Campaign's Success
While I am sure this generated a return, all I could think was "What a basic kind of approach". Finding out that a campaign was a big success among customers who already owned another specific fund is pretty basic. Analytics would give you much greater precision on this (I suspect that there is much more potential granularity in segmenting their customers than "own fund/don't own fund" that would help target the offers. In addition, if one thought of this as an Enterprise Decision Management problem, one would think about how to automate the decision of both which customer to mail and what to mail them so as to personalize and target the offer. Sending the same offer to a simple subset is a start, but only a start. Applying an EDM approach to a similar problem (getting existing customers to buy a product from a category they never bought from before) a retailer got 17% response rates or a 2000% lift from their existing process. There's a bunch of posts on this topic in the Marketing section.
Eastern Mountain Sports: Getting Smarter with Each Sale
Nice example of using BI for strategy assessment and then replicating it. I think they miss something by not thinking about this as an operational cross-sell decision, though. What about the website or the call center? If they took an EDM approach to build the best cross-sell option into a decision engine then they would be able to get the same increase across all channels much more quickly and automatically. Indeed there might not even be a need to train new staff as the till could make the recommendation and explain it.
Emergency Medical Associates: Measuring the Emergency Room's Pulse
Good example of performance management and strategic redesign based on BI. A comment in the story struck me as particularly relevant "Teaching people how to act upon data and analytics is harder than building the tools that report the data". Perhaps this means you should not attempt to teach people how to use the data and analytics but instead focus on automating key operational decisions and using medical rules/analysis of data to alert/warn staff with respect to specific transactions. This is what an EDM focus would do - develop decision automation that helps the staff as they go without requiring them to become analytically sophisticated.
Highmark Inc.: Detecting a Web of Fraud
Detecting fraud is a classic use for EDM - applying rules and analytic insight to transactions to tell which ones are fraudulent. Now BI can help in the research of fraud and perhaps in pattern identification but to catch fraud early enough to do something about it you need an EDM approach. You need to think about applying inline analytic models and expert rules as a transaction comes in, you need to make an operational decision that this transaction is suspicious enough to be treated differently. Indeed this kind of system exists - check out this post on healthcare fraud detection - and the rates of return are much higher than in "pay and chase".
Hillman Group: Making Price Adjustments on the Fly
While the story seems like a fairly classic BI story, I wonder if more could be done. As Dan Everett of Ventana Research says market profiling and segmentation is crucial. An EDM approach would be to think about these kinds of analytics in operational decisions like shipping additional stock to the point of sale based on predictions that the particular store will soon be out and so on.
Humana Inc.: Keeping a Watchful Eye on Patients
In many ways this was a nice example of assisting personnel using analytics and reporting. But what about helping patients self serve? How about influencing the behavior of other providers? If your focus was on improving the operational decisions you could do this. There is some really interesing work being done on EDM in heatlhcare.
Keystone Automotive: Knowing Which Products Are in Demand
When I read this I think to myself "do they really need to ask managers to decide?" Could they just not schedule the work? Why only daily? Why not drive the machines and people scheduling too (so as to manage maintenance down-time, staff overtime etc)? A focus on the operational decisions (which part to manufacture next, which staff to call in to handle overtime) and on EDM would deliver potentially much greater benefits. Plus what about predictions and patterns not just demand? Perhaps the drop off in demand yesterday is a good predictor of increase tomorrow? Unless you focus on predicting the future not just on analyzing the past you will miss a key benefit of analytics. This focus on the future is one of the differences between BI and the predictive analytics in EDM.
PHH Arval: Pinpointing Ways to Save
Another good performance management scenario. Could EDM, though, offer benefits by improving the scheduling of vehicles or of maintenance? Are there operational day-to-day decisions that could be similarly improved?
Schneider National: Aligning Purchases with Solid Data.
A classic strategic decision making problem, BI is the best way to do this for sure.
United Pipe & Supply Co.: Staying Stocked During Disaster
Again a nice tactical, ad-hoc BI project.
Virgin Entertainment Group: Getting Schedules in Sync with Customers
Why make managers do analysis like this? Why not have them focus on their staff and store and customers and automate this? Why not use an EDM approach to bring analytics (of trends, likely future behavior, dominant customer segments using this store) and rules (best practices about how often to change store layout for example) to automate these decisions? Why have a book signer or a manager look at reports when the system could be doing this stuff automatically and letting them focus on their fans and customers. Strikes me that a focus on the actual operational decisions in terms of trying to automate them would have been much more effective for Virgin.
So to summarize
- Some of these were really good examples of applying analytics, not just reporting, to strategic business problems. Classic BI.
- In most of those cases the company could probably get even more value by thinking about applying that insight to operational decisions with EDM.
- Most of the rest of the examples should be using EDM if they really want to get value out of the data - their current focus is leaving a lot on the table.