Generating Data on What Customers Really Want
At a fundamental level, the decisions managers make about revenue and profits fall into two categories—those related to growth and those related to cost reduction. Both types are meant to increase margins. But how data are used in each decision-making process is completely different.
Cost reduction data are precise. Firms know their cost structure very well and can compute with a reasonable level of certainty the savings each alternative being considered will generate. Managers use cost-related data as an objective input for decision making, since they consider it both reliable and reasonably predictable.
Managers do not treat growth-related data in the same way, and for good reason. Growth alternatives usually involve either launching new products or entering new markets, and these are activities where uncertainty is high.
Here data are used mainly as a tool for persuasion among managers. A stereotypical example would go more or less like this: Managers gather the data that reinforce their own point view on what they believe is the right business decision. During a meeting they all explain their various rationales and present their various data points. Usually, they end up reaching a consensus that they all consider makes sense. Then they leave the room still thinking that their own alternative was best but that in life you have to make tradeoffs.
Most growth-related decisions in management are made this way because managers do not have data that reliably predicts how new customers will react to their offerings or how any customer new or old will react to innovative offerings. If only Coca Cola would have predictably known that its customers would not embrace New Coke! But its executives lacked a way to generate predictable information on that score or tell them that the data they’d generated from their taste tests would ultimately not be relevant.
Disruptive innovation practitioners have just such a tool for reliably predicting customers’ behavior. It’s a methodology that uncovers what in disruptive innovation parlance is called a person’s “job to be done.” Briefly, the idea is this: Consumers don’t go to the store to buy products. They go to the store to buy something that will enable them to get some important job done in their lives.
The classic example, attributed to HBS professor Ted Levitt, is that people don’t want to buy a quarter-inch drill; they want something that will make a quarter-inch hole. Making a quarter-inch hole is the job to be done. The product that does that job most reliably, easily, conveniently, and less expensively is the tool they will be most likely to purchase for that job.
Work out your customer’s jobs-to-be-done, disruptive innovation practitioners have found, and you will generate data that more reliably predicts what a customer will buy and why. How do you do that? Traditionally, corporate innovators are told to conduct ethnographic studies, starting with no preconceptions, and to observe customers’ behavior and frustrations with the same open mind that start-ups employ. Human nature being what it is, that’s a hard thing to do.
Here, instead, is a simplified version of a methodology for identifying a customers’ job-to-be-done that starts with information about your own product. Since product information is not part of a job-to-be-done, the information about your product will drop out of the process in step 6, so that it doesn’t distort your results. But in this way, you can make your way to a new insight by starting in familiar territory. (And you’ll not keep the people you’re interviewing wondering what all your questions are about during the entire interview process.)
Step 1. Prepare a list of the key characteristics of your offering.
List at least 10 of them. Your product or service may be faster than your competitor’s. Or cheaper. Or have a better screen resolution. Or have leather seats. Or a battery that lasts for many days. Or connect to the internet and let you play with other people online. Let’s say you sell cars. Your list might include characteristics such as speed, gas consumption, how little it pollutes the environment, number of doors, colors, type of seats, cup holders and amenities inside the car for the driver and the passengers.
Step 2. Interview at least 10 consumers and 10 nonconsumers about the various features connected to your offering.
Nonconsumers are people who are not buying either your offering or your competitors’. So in this case you would be interviewing both people who drive cars and people who could drive cars but chose not to. The interviews can be anonymous, but you need to record the entire conversation. For each of the characteristics listed above ask three questions.
First, “Where are you when you are using this feature?”
Second, “When you use this feature, what are you really trying to do?”
And third, “If this feature were not available, what would you be using instead?”
Now, here’s an important part: for consumers, you need to ask the second two questions without reference to your product. So, to return to the car, let’s say you asked:
“Why do you sit in the passenger’s seat of your own car?” and the answer was: “I am a salesman, and in between meetings I work in the car.”
Then you ask: “When you do that what are you trying to accomplish?” He answers: “I try to have an environment that mimics my office space so I can concentrate and work comfortably for a while.”
Then the third question: “If you could not do that what would you do instead?” He replies: “In the car I use the cup holder for my coffee, and in the passenger’s seat I can work with my laptop and recharge my phone. If I could not do that I would go to a cafeteria, but it is difficult to concentrate there. It’s noisy and I waste time locating one.”
Step 3. Transcribe the recordings.
It is important that you do not miss anything. The transcript must end up looking like an interview, faithfully recording exactly what was said, complete with pauses. Being systematic about when the data stop is important for the statistical analysis used in step 5.
Step 4. Codify the transcripts by tracking all the meaningful nouns, verbs, adjectives, and adverbs.
To continue our example we would extract from the sentence: “I try to have an environment that mimics my office space” will result in the following codes: “try,” “environment,” “mimics,” “office,” and “space.”
With this you create a table in which you count the number of instances of each word (so if in the entire first transcript the word office is only repeated once you would have a 1 in the first row). Complete the table until all the sentences from all the transcripts have been codified. There are software tools available to do that more easily.
Step 5. Group the codes. Using the statistical technique of cluster analysis, group the codes based on their proximity.
That is, it measures how many times each word appears close to one another (that’s why the pauses matter). Let the software that you choose determine the optimal number of clusters. The end result is that all your codes will now appear in groups.
Step 6. Remove descriptive data so you have only prescriptive data left.
To do this, you remove all the groups that contain information about your product, in this case, all the groups that contain the word “car.” The end result is a series of groups that each contains a number of different code words. Within each group, these codes refer to the customer’s way of thinking and the portions of the context that the customer considers relevant to deciding on which product alternative to buy.
If you compare how product performs in relation to the concerns expressed in each group, your next product improvement will become compellingly clear, not only to you, but also to your colleagues. In this example, it would become easy to make the case for focusing an innovation effort on helping salespeople become more productive and work more comfortably in their cars.
Better electrical outlets, perhaps, so people can charge more than one item at a time? Storage for computers or samples? Something that mimics a desk more effectively?
The point here is this is a fruitful avenue for further attention since most car models sell fewer than 100,000 units per year, but millions of people work in their automobiles.
During the 1950s Edward Deming and others developed such tools as statistical process control charts and total quality management techniques that have made the cost-reduction data we use today predictable and reliable. Before that, though, data about cost reduction was as unreliable as growth related data are today.
Now the first tools are starting to emerge that add predictability and reliability to growth related decisions. Jobs-to-be-done is one such tool. Once managers learn how to compute a job-to-be-done by themselves, their growth related decisions will become much more objective and less opinion based.