Analytical Engine

Interesting things are taking place in pharmaceutical industry, which I believe has the potential to change pricing landscape of high cost drugs.

In June’07, the National Institute for Health and Clinical Excellence (NICE) of UK decided that Johnson and Johnson’s multiple myeloma drug, the one year treatment of which stands at around USD 48,000 a year, can be defrayed for if the company is willing to refund the cost in case of not showing a pre-decided level of cure.

This move towards sharing the risk of treatment outcome did not go unnoticed. Providers are all the more willing to share the risk of no-outcome with the drugs manufacturers. Biogen Idec, maker of Avonex for Multiple sclerosis drug that costs around $18,000 a year has made a dynamic pricing agreement with UK’s NICE. Under the plan, 5.000 multiple sclerosis patients are being followed for 10 years to see the effect of the drugs in slowing the progression of the disease. The prices of the drugs will be adjusted based on improvement of performance. Merck-Serono has offered to refund the primary care cost if its drug Erbitux if the patient does not respond within 6 weeks.

Although Europe seems to be leader in this development, but US companies are definitely taking note. United Healthcare have made an agreement with Genomic Health with United paying for the $3,460 genetic test to determine whether a woman with early-stage breast cancer would benefit from chemotherapy. If tests don’t have the intended impact of lowering chemotherapy United Healthcare will negotiate a lower price for the tests.

Essentially this development adds a new dimension to the pharmaceutical pricing, i.e, risk. Before pricing a drug, the point of paramount interest to both manufacturer and providers is what drugs are the ideal candidates for outcome-based pricing? The high cost drugs which are new in the market or have relatively low success rate seem to be the ideal target from the developments so far. The principal risk that companies should take into account is financial risk, which in turn will depend upon probability of no- outcome and amount of dollar exposure. Apart from regular variables such as physician’s ability, patient’s specific variables etc. there may be exogenous factors such as pollutions level for certain treatment which will affect the probability of outcome.

There are several interesting questions already. Will the companies have different refund policies for different locations? Will there be selection criteria? How will regulatory environment affect the risk-sharing?

We are at an interesting corner here. I will wait and watch how the pricing territory shapes up in the near future.

How many times one has deleted that email offering that free gold credit card or a pre-approved loan? How many times one has curtly disconnected the phone call from another tele-caller with a pre-approved check for that awesome holiday in Hawaii? In a market, where retail banks have exhausted almost all the channels to reach the target customers, ATM machines can be a powerful alternative for cross-selling.  ATM machines also provide the opportunity to reach out to the otherwise difficult to reach customers such as expats. 

So, how would it work? Almost every bank has pre-approved offers for certain customers.  The bank can load the offers for some of the pre-approved customers to the server that interacts with the ATM machines.  The moment the ATM recognizes a preapproved customer, it can show her the pre-approved offer through a blinker or an additional screen at the end of the transaction. If the customer is willing to go ahead with the offer, all she needs to do is to press “Y”.  The moment she presses “Y”, the bank gets notified and can track the lead through different mechanisms. For example, an SMS can go to the sales manager nearest to that ATM with the customer detail, who can take up the matter further.

Surely, it’s not as easy as it sounds. While the technology for such a process is available in the market, it’s still a challenge to implement it. Prediction accuracy is another big challenge.

·         It is not feasible to upload the entire set of pre-approved offers to the server.  And hence the need for a model, which can forecast with high accuracy the probability of a customer visiting a cash machine.

·         Each ATM screen on an average takes 5-7 seconds, which includes the time taken to interact with the server.  So, the time to pitch the offers must be chosen judiciously, especially for locations with long customer queues.  But restriction of timing will in turn affect the prediction accuracy.

·         The ATM cannot close the sell. The completion of the process depends on further integrations of channels.

Nevertheless, ATMs can still be used as an excellent channel and in the near future itself, with the advent of technology, it can be expected to be one of the primary channels.

�

Societe Generale has been in the news recently and not quite for reasons they would like to be in the news for . In one of the biggest cases of Insider Trading Fraud, an SG employee managed to cause a dent of over $7bn. And this is hardly the first ever case of such use of material non-public information. In one of our earlier posts, we talked about a project where we had built filters to tag suspect insider trades to help our client with identifying potentially suspect trades for manual evaluation. The system has been used retroactively so far, and is being considered for proactive tracking as well.

The efficacy of what we proposed was based on the simplicity of filters that we built. Using tribal knowledge for identifying suspect transactions, and augmenting it with statistical rigor and tests, we were able to generate efficiencies in the compliance/audit process. These same rules can be implemented in a real time fashion to flag suspicious activities for monitoring.

Update: Julie wrote a note to us and pointed out that SoGen case is not a case of insider trading but one of outright fraud. Thank you for catching it and pointing it out to us. Nonetheless, a set of logical and statistics based filters can be a useful part of effective control environment which may have been able to detect such fraudulent activity.

What once was a good ole’ boy sport that relied on tribal knowledge buried deep inside a crew chiefs’ head is becoming a hi-tech, data rich, and analysis driven sport. More and more teams, crew chiefs, engineers, and even pit crews rely heavily on granular data elements that are used to feed complex models to extract the slightest bit of knowledge that can be used to save thousandths of a second on the track or in the pit. The sport of NASCAR is no longer defined by car lengths that won the race rather it’s becoming a sport defined by hundredths and in some cases thousandths of a second that decides between the winner and the first loser, a.k.a. “second place”.

After having an interesting and thought provoking conversation with Ken Brown, former engineering director at Dale Earnhardt Inc. and lead engineer for GM’s Corvette Racing program, I found myself drawing parallelisms between what I do in marketing science and statistics at Diamond for our clients and what NASCAR teams are doing to extract value from information. Not in the ROI sense but in the thousandths of a second sense. However, this in the end does translate into ROI for the thousands of sponsors that make the sport go round and round, no pun intended!

According to Ken, granular data elements such as wedge, nose weight, spring rates, tire pressures, camber, and gear ratios (just to name a few), all play a critical role in determining the how good a car will perform at a particular race. However, the key decisions are often made long before the car arrives at the race track. Essentially through track testing, wind tunnels, shaker rigs, and simulation exercises, the engineer and crew chief can decide which and how much to change each variable to make their car the best on the track. In marketing science I call these things covariates!

So how are teams doing this? Although the details of the process are unique to each team, the general process that goes into setting building and setting up a stock car is as follows:

1. Know the track (flat track, road course, banked)

2. Define/identify the significant set up parameters (or covariates) relevant to that track

2. Gather test data (prior race data, testing, wind tunnel, etc) and validate its quality

3. Analyze preliminary data via data visualization techniques (histograms, pareto charts, and cross-tabs)

4. Build models (e.g., DOE, Pareto Analysis)

5. Analyze results of simulation techniques

6. Set-up car for the race

7. Test and refine!

This process that is followed by the top teams is not unlike the process we follow in our information and analytics practice at Diamond when trying to identify the key variables that will attract and retain the customers or segments a business is going after. Moreover, the quality of the testing data is one of the most critical components to the entire process.

So the next time you see a professional stock car race, it’s important to know that before there was the driver in victory lane, there was a data geek in the background making it all possible.

The Net Promoter Score (NPS) methodology uses a 10 point Likert scale to calculate the all important metric. I am not sure whether any research was done before a 10 point scale was chosen, because the scale certainly can influence the metric. In general, this caused me to wonder how to decide whether a Likert scale survey question should have a 5 or 10 or any other point scale.

A quick google search pointed to some academic research. While there are no rules of thumb, here are a few guidelines to consider when you are deciding on the scale.

1. Respondent knowledge of subject matter:

1. If the respondents are not very familiar with the subject matter – they tend to abuse the endpoints of a longer scale.

1. More response options introduce error when respondent group has very different frames of reference.

So if you are asking about a tax filing software and the respondents have just used the product to do their taxes, then their knowledge of the software and their frame of reference will be relatively similar – so a 10 point scale can provide more nuanced results.

However, if you are doing a survey to determine whether the respondents will sign up for a new wireless phone service – where wireless usage and needs of the group might significantly differ and also their knowledge about various services – a shorter 5 point scale might be the better option

Update: NPS does have a whitepaper on the topic where they justify their scale

A New York Times article about mobile marketing mentions two consumer survey based studies, and highlights the wide discrepancies in the results to seemingly similar questions. For example, one study quoted that 54% of population in Britain used mobile web compared to 14% quoted in the second study.

As pointed out by moconews, there are subtle differences in the way questions have been asked in the two studies, which result in these huge variances in responses, emphasizing the importance of proper survey questionnaire design.

Developing a hypothesis based design, using the right targeting filters and writing an appropriate questionnaire is a mixture of art and science. We have put together a short training manual which is a good and practical resource if you use surveys to understand your customers and make decisions based on the responses. You can download it from here.

Defining the problem and understanding the underlying data is necessary prior to selecting the statistical measure to employ in any data analytics exercise. Failing to do so will result in faulty policies, assertions and decisions as you see in the examples below:

Virginia newspaper reported that teachers were refusing to give students zeros for work not turned in because the zero had a devastating impact on the students’ average score (Via Huffington Post). They were using mean instead of median as the measure of central tendency. (Updated per Zach’s comments)

American Farm Bureau Federation added up all personal food expenses and divided by personal income to come up with the Feb. 6 date as National Food Check-Out Day by which, “the average American earns enough disposable income to pay for his or her food supply for the entire year” (Via WSJ online- subscription required). Again used mean instead of median as the measure of central tendency.(Updated per Zach’s comments)

A creationist publishes a book under the pen-name of John Woodmorappe explaining how Noah’s ark could have actually held all of the animals. He did this by computing the median size of a species, and then multiplying that by the number of species (Via Good Math, Bad Math). Used median instead of mean to define the population and extrapolate.

And finally a relevant quote:

“When she told me I was average, she was just being mean”. (credits – Gary Ramseyer)