When Choice Is A Bad Thing – The Marginal Utility of Choice

Being able to choose between different options is a good thing for the user! right? but when you can choose between 65 different kinds of blue, 1122 different fonts and whether a display should only work on Sundays between 11 and 12 for a special group, giving MORE choices to users start to be not so good or, to put it bluntly: bad!

In general most are brought up in a democratic state, where expecting to have a choice is as basic as eating or breathing. This is why choice in all its guises has a positive ring to it. But there are actually situations where limiting your choice is the best strategy. It has worked for artists and musicians to enhance creativity, but it also works for ordinary people. This is important to consider when you are designing and building new products.


The marginal utility of choice

In order to understand why and when more choice stops being a good thing I will introduce the concept The Marginal Utility of Choice: “The marginal utility of choice is the perceived benefit that the option of one more choice will offer”.

Let us look at an example. If your product is a car rental service, then increasing the choice of color of cars from 1 to 2 may be a significant increase in utility for the user. Now you can suddenly choose whether you want the car in black or white. Adding blue will also offer great utility. Continuing to do this will continually add some utility to the car rental service as a whole. But when you already have, say, 35 different colors to choose from how much will adding color number 36 improve the utility of the service as a whole?

This example shows that it is not a linear function, that is, adding one more choice will not indefinitely result in the same increase in utility. In the following I will offer a posible explanation based on human psychology that could help explain when having more choice stops is a bad thing.


The optimal number of choices

In general the value of an extra choice increases sharply in the beginning and then quickly drops off. Given the choice of apples, oranges, pears, carrots and bananas are great, but when you can also choose between 3 different sorts of each the value of being offered year another type of apple may even be negative. The reason for this phenomenon has to do with the limits of human consciousness.


In order to make a conscious choice there is one fact we know with Cartesian certainty: you need to be conscious about it. From half a century of psychological research starting with the seminal article by George Miller from 1956 “The magical number 7, plus or minus 2” we know that our consciousness has some severe constraints on how many things it can work with. It seems to be able to maximally hold 4 to 7 items at the same time (depending on the type of test and training). When the number of choices exceeds 4 to 7 items you can’t hold it in your consciousness anymore and the choices can’t be evaluated against each other. Therefore the marginal utility of choice quickly stops at the other side of 7.

I once got into a discussion with Chris Anderson (author of “The Long Tail”) about this observation. I argued that companies that offered only a very limited number of choices of products and functions of their products would be more successful, contrary to Anderson’s argument that the long tail and infinite choice was the way to go. We never really reached consensus on that though, but consider the following:

How many different phones does the worlds leading phone manufacturer, Apple, produce? 4, iPhone 6, iPhone 6 Plus, iPhone 5s, iPhone 5c

How many different choices does the worlds best restaurant offer their customers? 3 choices: 1 menu, a wine menu and a juice menu

How many car models does the worlds most hyped autoproducer, Tesla Motors, offer to their customers? 2, Model S and Model X

All of these leave the consumer with a number of choices that is below the threshold of our consciousness.

Expressed more formally we can stipulate that “the marginal utility of choice rises sharply between 1 and 7 choices and then decays”


Now the question is “what happens after that?” When choice cannot be held inside consciousness, it will try to group them into different segments. That could work well to some point probably around 50 (7 chunks time 7 choices). In this interval a new choice can still be grouped with others with some effort, but it will take mental effort to understand and compare it to all the other choices. This is why adding a new choice will be a bad thing, since it is cognitively costly.

In this range the marginal utility of choice is slightly negative, because the added mental burden of yet another choice detracts more than the utility of it.

More than 50

After about 50 everything will just be a blur because the possibility of comparing it to all the other choices has broken down in our consciousness.  So, adding another choice will not make any difference. The marginal utility of choice is zero.

This means that the aggregate utility might even fall below zero some and then stabilize. This means that if you keep adding choices there might come a time where the utility is lower than having no choice.


The marginal utility of choice curve

Please keep in mind that this is a hypothesis, that is based on theoretical observation and some, albeit anecdotal, empirical evidence. I think it aligns well with a lot of observations, but it should be tested more rigorously.

We can now draw a function for the marginal utility of choice. It looks something like the above

We see that the total utility rises sharply in the beginning, we can call this “The climb to enlightenment”. This is the section where adding another choice gives the most marginal utility until the it approaches zero. The curve then evens out and decays which we could call “The slope of attrition” where adding another choice reduces the perceived benefit. The marginal utility of choice is below zero, which means that every new choice added decreases the aggregate utility of the service. The reason is that each new choice adds cognitive friction. Finally it will stabilize “This we can call the plateau of indifference”, this could be above or below zero, because having a lot of choices could very well be more frustrating than having none at all.


Case study – SaaS pricing models

It would be interesting to see if we could see this in real life. Let’s examine the case of Software as a Service pricing. SaaS companies live from selling subscriptions. Usually the user can choose between several different plans. This is obviously a case where the number of choices is important. If it were true that there is an optimum of choice between 1 and 7 we should be able to see this in how many tiers are actually offered.

In an excellent report by Price Intelligently “The SaaS Pricing Page Blueprint”, the authors studied 270 SaaS companies’ pricing page. If we look at how many plans the companies have it is overwhelmingly evident that most companies (88%) only let their customers choose between less than 7 options. About half of all companies offer three or four choices (55%). So, clearly SaaS companies have most success with offering less than 7 choices.

This seems to be an indication that there could be evidence for the stipulated rule that the marginal benefit is postitive between 1 and 7 .

Limiting cases

Now, this curve holds under the assumption that the user is doing his choice unaided by anything other than his or her own consciousness. It is important to note that this assumption doesn’t hold in all cases. Today we can often use AI to help us cut down the number of choices. When we look at a book on amazon a number of books are presented below. It is not a list of all books amazon has, but a subset based on what the algorithm thinks is relevant. The utility in this case depends on the precision of the algorithm, which is a completely different problem.

Another condition that should hold is that the choice should be comparative, that is, it should be a choice where the alternatives are compared rather than evaluated one by one. An example is finding a movie on Netflix or iTunes. you may go through a long list but you are not usually comparing every single movie to each other. Either you will create a short list (which will probably not be much longer than 4 to 7 movies) or you will just choose movie by movie: “do I want to see this now?” (a binary choice).


So, if you have a situation where the user should be given a comparative choice and there is no way to make AI support for that choice, then the marginal utility curve stipulates that about 4-7 choices is the best from a general point of view.



A Practical Guide To Doing Cost Of Delay Based Prioritisation

It is often very difficult to prioritise what to build and when. One of the most efficient methods of prioritising features is prioritising according to cost of delay.

Originally invented by Don Reinertsen in “Managing the Design Factory” as a new way of looking at how to build stuff, it has inspired many agile teams to apply this thinking in their sprint planning efforts.

The problem is always how exactly to find out what the cost of delay is. It is notoriously difficult to put a price tag on a feature. This is probably what has discouraged most people from doing it. But the fact that you can’t put a precise price tag on the cost of delaying a feature shouldn’t keep you from applying this kind of thinking.

One very good solution to how you might do this is provided by Dean Leffingwell in his book: “Agile Software Requirements”. He argues that cost of delay can be broken down into three components: User value, Time value and Risk reduction.

How To Break Down Cost of Delay

User value is the potential value of the feature in the eyes of the user. Product managers or product owners usually have a good feeling for this, but other parts of the company like consultants or sales people who spend a lot of time will also have a pretty good understanding this as well. One should not forget that asking the user him or herself is the most obvious solution. The reason why most people don’t so this is probably that it is relatively cumbersome at scale. At Sensor Six however we have seen customers use our product with great success to engage directly with customers to get input on the user value. Often a company will have a customer panel or a mailing list, where it is natural to ask your customers. It doesn’t have to be real value, but something simple as a 10 point scale could easily be used. More sophisticated uses we have seen with our customers is the use of forced ranking, which is an excellent way of measuring if you don’t have too many features to rank.

Time value is based on how user value decays over time. Many features are more valuable if they are delivered to the marketplace early, so they can provide differentiation. This depends very much on an analysis of competitors current state and what is assumed they are working on. This is why it is usually business analysts or product managers who will be able to rate this. Again it is possible to just use a 10 point scale to measure it.

Risk reduction/opportunity enablement describes the degree to which a feature helps us mitigate risks or exploit new opportunities. We live in a world with many unknowns and therfore it is important to be able to guard yourself against the unknowns that are threatening (risks), but also to remain open to the ones that could help us (opportunities). This evaluation will always be very subjective and dependent on the person doing the rating. Since people have very different perceptions of risk it is a good thing to invite several people to ascertain risks.

I would say that these are really good suggestions, but there could be other approaches as well. Leffingwell argues for rating thes on a scale from 1 to 10, but that can also depend on the context. If you have very few features and want a very precise measurement you should use a relative method such as ranking or pairwise comparisons. But in the end it is more important to have some sort of indication, so an easy method like a 10 point scale will be a good place to start.

How to Use the Cost of Delay Calculation For Prioritisation

The cost of delay of a feature is the same as the value it has if it is not delayed, ie. produced. Now that you have this value the next thing you would want to do is hold it up against the effort it would take to produce the feature. Effort can be estimated in the same way with a 10 point scale. If you work in an agile context you may as well use story point if that is possible. Here is where you can get your engineers to look at the features and give some sort of estimate.

Once you have som data on the effort there are several ways you could attack the problem. According to Reinertsen in The Principles of Product Development Flow there are two ways:

Shortest job first is the method to use if you want to look at minimising effort. You simply start with the features that are smallest and work your way through them regardless of the cost of delay. If features all have the same cost of delay, this is the best way to do it.

High Delay cost first is where you simply start with the features that have the highest cost of delay regardless of the effort. This is efficient in producing a high economic output. If features have the same effort this is the best way to do it.

Weighted shortest job first is where you weigh the value against the effort. If features have different effort and value, which is most often the case, this is the most efficient method to use.

How To Do It In Practise

Strangely, given the popularity of this thinking, no tools seem to support this particular way of prioritising, so it is always something that needs to be done in spreadsheets. To our knowledge Sensor Six is the only product management tool that does all of the above out of the box and even makes it possible to engage different stakeholders directly. In the following I will show how you can do exactly what Leffingwell recommends in Sensor Six.

You can see how the above would be set up in Sensor Six by going to our website and logging in with the credentials below or you can follow the step by step guide to prioritise your own features.


Doing Cost of Delay Prioritisation In Sensor Six

First you set up the different criteria: User Value, Time and Risk Reduction. You configure them as benefits and a 10 point scale. Supply them with a description.

Now you can rate them solo and evaluate everything by yourself. If this is fine and sufficient for your needs you may skip the next section.

If you want the product manager, sales rep or any other stakeholder group who are fit to act as a proxy for the user to evaluate user value simply go to the collaborate section. Here you can configure a workspace that will allow the user to give input to the user value. Simply copy the link to the workspace and send it to a list of people who you think could give you this input. Have the competitive intelligence function whether it is located in marketing or some other division rate the time value to get input on the time cost of delay per feature. Let the business analyst give input on the risk reduction opportunity dimension. Finally you need to know something about the effort. This you will invite your engineering department to work on.

The actual persons or roles can be cut differently, so maybe you will ask the same persons to rate the time and risk reduction domains (typically business analysts), engineering will estimate the effort while sales or support may evaluate user value.

When you have input on the cost of delay, you can plot the total cot of delay on the y axis and the effort on the x axis.  You should then choose those with the best cost of delay/effort tradeoffs first in order to arrive at the most efficient development plan, if your features have varying effort and cost of delay.

So this simple process can save you hundreds of hours and deliver more value to the market place quicker.