The Need for Smart Enough Systems (Part 7) ~ Contributing Value to your ROI Calculation: Cost Reductions
Last time, we covered when to use (and not use) enterprise decision management and took a look at the Return On Investment (ROI) for Enterprise Decision Management. We concluded by promising a look at three layers — cost reduction, revenue growth, and strategic control — that can contribute value to your ROI calculation. In this instalment we explain the first of the three, cost reductions.
Information systems that aren't smart enough to cope with current demands waste an inordinate amount of money for organizations of all types. This waste can come in a variety of forms, including costs for buying unnecessary data, wasteful activities, lost opportunities, fraud, and fines. Which costs matter most to you or for a particular decision vary. Applied correctly, enterprise decision management has the potential to eliminate or reduce many costs.
Time Not Spent
The first element of cost reduction is perhaps the most obvious: time. The amount of staff time required to process transactions in which decisions are automated is reduced dramatically. Many processes have already been automated and streamlined repeatedly but still leave core decisions to be made manually. Typically, leaving some manual decisions means having items queued on work lists; before the process or transaction can continue, there's a wait for staff to access the list and process the item. Many of these decisions can be 'rubber stamped' quickly by someone, but even this process consumes a lot of time if the transaction volume gets high enough. Automating these decisions can eliminate extra staffing costs from most transactions. This cost can be significant, especially for decisions made by professionals, such as pricing or authorization decisions.
Few decisions are amenable to 100 percent automation, so there will always be some referrals for manual decisions. An EDM approach can deliver high percentages of automation, however. Auto insurance renewals, for example, have been automated at rates of 95 or even 99 percent. Even when the remaining manual decisions represent a disproportionate percentage of the workload because of their complexity, organizations still save a lot of time. One group of advertising managers, for instance, freed 30 percent of their time when they used enterprise decision management to automate all but the most complex pricing decisions.
Improved Referrals. The process of automating these decisions could even improve referrals. The system can give you insight as to why the decision is being referred, often more than was previously available. For instance, if a pricing engine refers a price decision for an ad insertion because the customer wants a combination of layout and colors for which no rules exist, the advertising manager who gets the referral can be told immediately what the issue is. Understanding the reasons for a referral makes it easier to process an exception.
You can also eliminate time from a process by removing the need for a 'do-over'. Typically, a process with manual decision making has a delay from data capture to processing. Data can be captured from a Web form or a call center employee and then queued for decision making by someone else. When the more skilled employee reviews the data to make a decision, the data might be incomplete, so the employee needs additional questions answered. A typical process, shown in Figure 1, can lead to constant back-and-forth interaction as the more skilled employee requests more data, which is then gathered, which turns out not to be enough, and so on. In extreme cases, such as commercial underwriting, this process can result in as many as seven attempts to make a decision, with the first six failing because of lack of data. All this rework costs time and money, to say nothing of the negative impact on customer service.
Figure 1. Manual decision-making processes can involve many 'do-overs'
Automating a decision, in contrast, pushes it to the point of contact and allows the people collecting data to make sure they have captured all the data required to decide. The front-line system or worker collects data, and then the decision service tries to make a decision while the initial 'conversation' is still open. If it succeeds in making a decision, including one to refer the decision to an expert, the conversation can close. If the decision service can't make a decision, it can identify why and prompt the system or person to collect additional data. The service can repeat this process, shown in Figure 2, until a decision can be made. The decision service can then make the decision in many cases, and the customer gets the answer without waiting any longer.
Figure 2. Automated decision making increases the odds that the first conversation results in a decision
Some transactions have costly activities, such as property inspections, that are optional. These activities aren't required for every transaction, because they might not make a difference in the actions taken. For instance, if an application for house insurance states that a house is falling down and needs to be completely rebuilt, an inspection isn't likely to change the insurance rate. All it would do is confirm the information on the application. Alternatively, if the application hinges on the truth of a positive comment about the property, an inspection is probably essential. In more complex situations, it may not be immediately obvious whether the inspection is called for. For instance, an unusual combination of property features may meet some legal requirement for an inspection.
In systems using manual decision making, these inspection activities commonly start in parallel to reduce the transaction's total time. This method can result in unnecessary costs when you pay for inspections or other activities that turn out to have no impact on the decision made. Automate the decision, however, and it's easy to order the inspection only if it makes a difference to the decision. The system tries to make a decision without the inspection and then orders one if it can't decide between options.
If more sophisticated decision making is involved, it might even be possible to order inspections only when the difference they might make exceeds the inspection cost. In other words, if the inspection is going to make little or no difference to the price — and, therefore, to your profit on the transaction — perhaps you can't justify its cost and shouldn't perform it.
Not Just Saving Money. You can extend this approach to nonmonetary costs. For example, if you have an overused system with tightly controlled and rationed access, accessing that system only if you must can reduce operational costs dramatically. If a system is working remotely from a data source, perhaps wireless network traffic can be reduced by not requesting information from that data source unless it makes a difference. This practice can also improve performance if slower parts of the decision are invoked only when they must be.
Data Not Purchased
A similar savings can be had with external reports and data, such as government record searches or address validation services. Ordering reports from external services costs money — typically a certain price per transaction. In auto underwriting, for example, more than one motor vehicle report is commonly ordered for the same policy application, because different people in the process order and reorder the report at various steps in the process. In fact, one company with a typical manual decision-making process found that it ordered 1.1 reports per policy. After automating the decision, reports were ordered only when the decision reached the point where a report would make a difference. This practice reduced the use of reports to 70 percent of transactions, eliminating an average of one report for every 2.5 policies underwritten. Again, the system tries to make decisions without additional costly information and requests information only when required.
Fines Not Paid, Errors Not Made
In regulated industries, avoiding fines and penalties by using only manual decision making is becoming more difficult. When a network of agents or third parties is involved, avoiding fines and penalties can be even more complicated. Often the company pays fines even if the misbehavior came from its independent agents. Large sums can be saved by automating regulatory compliance rules to make sure no noncompliant transactions go on the books and by focusing enforcement and audit resources on the highest risk transactions, where there's potential for litigation or fines. For instance, life insurance often has regulations on sources of funds to prevent unscrupulous agents from encouraging customers to 'churn' from one product to another. However, often the insurer must pay the fine. Knowing that transactions are compliant can eliminate or dramatically reduce hard costs, such as these fines, as well as soft costs, such as litigation. Indeed, you can become more sophisticated as to which claims you decide to pay, for instance, by considering the potential cost of litigation part of deciding how to treat a transaction.
Eliminating Bias. Eliminating bias in decisions through automation has a long history. Replacing human judgment, which is subject to conscious and unconscious bias and hard to police, with a verifiable, mathematically sound approach can reduce bias. This practice can eliminate fines and long-term litigation risk and improve an organization's image.
Some people dislike credit scores because they're impersonal, but these scores aren't based on skin color, religion, or any other trait once held against credit applicants. As an early Fair Isaac advertisement for credit scores said, "Good credit doesn't wear a suit and tie."
Cost reduction can also come from eliminating costly errors. One of the attractions of automation is that the decision logic applies to every transaction. In contrast, manual decision-making processes are often applied to only a subset of transactions. For instance, few insurers using manual underwriting decisions review more than 10 to 20 percent of renewals to see whether they should be cancelled or changed. The remainder are renewed automatically without review. An automated underwriting process, in contrast, can process 100 percent of renewals effectively and efficiently and flag those most worth revisiting and repricing.
Some manual decisions are made in high-stress environments, such as health care, and an automated decision can be used as a 'backstop' to catch errors. Eliminating obvious medical errors, such as the wrong dose of a drug for a patient's weight, can reduce medical costs (by getting the treatment right the first time) and potential downstream costs, such as legal bills. Of course, these automated decisions have a clear benefit for patients, although it might be hard for a provider to put a monetary value on this benefit!
Eliminating or drastically reducing fraud is a classic use of enterprise decision management. For instance, a key issue in claims is identifying fraudulent claims quickly but not annoying honest claimants with unnecessary delays and problems. A company must not pay too soon and have to recoup losses later, nor must it delay legitimate payments. This situation is ideal for rules and analytics to work hand in hand. Predictive models can be developed that estimate the likelihood of fraud initially and can be refined as additional data about the claim is collected. Combined with rules to manage acceptable levels of risk for automatic payment or referral to special investigators, these models allow efficient process automation. In the gray area, rules can be used to give advice to adjusters and capture additional data to explain actions taken. When volumes are particularly high, as after a hurricane, the company could soften the rules to refer fewer claims and/or use the models to prioritize the most suspicious claims for investigation.
Note. An EDM approach introduced in the credit card business, for instance, reduced the overall rate of credit card fraud from 18 basis points (1/100th of a percentage point) in 1992 to just 5 in 2004.
Opportunity Costs Reduced
Opportunity costs, lost revenue, or costs incurred by failing to respond to a change in the market can be a big problem. Taking advantage of an opportunity before a competitor does or while it's still relevant can be complicated if your organization must change how it makes pricing or product-offering decisions. Responding to a competitive threat quickly, so as not to lose revenue and customers, can also mean rebalancing risk management decisions or repricing products. In both cases, an EDM approach can reduce the opportunity cost by allowing organizations to change their systems, or at least the decisions in them, more quickly. For instance, in trade credit insurance, an entire industry's creditworthiness can change rapidly. The Enron scandal affected all energy companies' trade credit for a while, so when Enron crashed, any company offering trade credit insurance to energy companies had extra risk exposure for as long as it took to change its system. One insurance company used an EDM approach to reduce the time to make this change from two weeks to just eight hours, which improved response time and helped avoid potentially bad deals.
Similarly, failing to take advantage of market opportunities can lead to lost revenue, especially if a competitor can respond more quickly. An electronics distributor, for instance, found that changing its rebate management system took too long — so long that its customers couldn't take full advantage of manufacturers' rebates. Changes could take six weeks, during which time the distributor seemed more expensive than its competitors because it wasn't passing along manufacturers' rebates. This delay resulted in lost revenue. Taking an EDM approach to automating rebate calculations more flexibly meant the distributor could change its programs immediately to reflect changes from manufacturers. This practice increased revenue and improved billing accuracy.
Next time, we will explain revenue growth, the second of the three layers that can contribute value to your ROI calculation.
|Acknowledgement: This material is from the book, Smart (Enough) Systems, by James Taylor and Neil Raden, published by Prentice Hall (June 2007). ISBN: 0132347962.|
# # #