Montag, 20. Mai 2019

Make Hypotheses, not Assumptions

Assumptions are a necessary evil in the working world. To move forward on decisions we often need assumptions in project planning, business cases, specifying requirements, estimations of any kind and surely many other cases. We assume in order to confront uncertainty. But, confrontation is not management. The old joke reminds us of the danger: “to assume” means to make an “ass out you and me”. Fortunately, there’s a better way: work with hypotheses instead.

Assumptions vs. Hypotheses

What’s the difference between an assumption and an hypothesis? An assumption accepts some idea as true, at least temporarily. As we face uncertainty in a decision, we resolve it with a best guess based on what we know now. Investing time in finding the truth would be too much work. So, we accept our best guess. For example, in planning a new product, we might assume our customers will continue to buy at similar rates as the past. As the real rate lies in the future, no amount of effort can tell us what it is during the planning stage. So, instead we suspend uncertainty with an assumption, then decide, and get on with the project. There’s nothing wrong assumptions per se. We need them to resolve otherwise intractable problems. But, assumptions have a nasty habit of being regarded as the truth. Once accepted, no one questions them.

An hypothesis, on the other hand, implies that we are still testing the truth value. We’re open to confirming or rejecting it, just as scientists do in proper experiments. As paradoxical as it sounds, scientists seek The Truth™by falsifying hypotheses. As Karl Popper showed, we can never prove a proposition to be absolutely true, but we can show it to be definitely false. As in Popper’s famous example, the statement “all swans are white” can never be proven, whereas one black swan suffices to negate the claim. Thus, while assumptions suspend uncertainty by imagining truth, a hypothesis-driven approach reduces uncertainty by repeatedly testing.

Applied to a project plan or a business case, hypotheses remind us to revise them as more information becomes available. If the hypothesis has a major effect on the decision, we’ll be sure to test it rigously. We can popper through the problem by setting up various scenarios or conditions that we seek to falsify. After testing and examination, only those scenarios still standing remain plausible.

Many business cases are based on detailed examination of the costs and on dubious assumptions about the potential benefits. If a high-placed manager exclaims, “I believe the market is ripe for such a cool product, we can assume we'll sell a lot”, he silences any objections, sometimes with disastrous consequences. Such a claim, despite the high salary and the Burberry scarf, should be evaluated and tested just as any other.

Using Hypotheses

So, how do we proceed with an hypothesis instead. Let’s suppose we need to sell 60,000 widgets to break even on the investment costs. We know we sold 75,356 of the old model last year. The steps are quite easy:
  1. Set up the hypothesis: customers will purchase at least 60,000 units per year. 
  2. Test the hypothesis: Note our goal is to show that the hypothesis is false! Only that way can we say that it has withstood our test.
    • Have we ever sold fewer than 60,000?  If yes, how often?
    • What happened the last time we introduced a new product?  Did the sales numbers dip first?
    • Is our nearest competitor also still selling lots of widgets?
    • How many existing customers bought the new version last time? Are we dependent on upgrading for the sales numbers?
    • How many new customers are we gaining per year?
    • How is our market share developing?
    • How is the general economy, i.e. the purchasing power of our customers?
    Any question that challenges the 60,000 figure is welcome. Modeling exercises such as Monte Carlo simulations are a valuable tool, too./1/
  3. Accept or reject the hypothesis
  4. Revise the hypothesis and repeat, possibly using some new questions to challenge it further.

Wordchoice matters

Of course, we can do all of this for assumptions, too. But, wordchoice matters. Words take their meaning from what linguists call semantic fields, i.e. the set of associations bound to a word. The semantic fields surrounding “assumption” center on “temporary acceptance of something unknown” and the dangers of doing so. While the word does suggest “temporary” and even “risk”, it does not conjure up the notion of testing. “Hypothesis”, by contrast, immediately evokes images of da Vinci and Galileo or putting on safety glasses in the chemistry lab. We associate it with an open-ended procedure of examination.

If you use the word “hypotheses” in a business case, the decision makers will read the business case differently. First, they may be curious about the choice of word. With that, you can explain the value in testing and being open to changes. Secord, they may ask how you will test the hypotheses, as opposed to just asking, “What makes you so sure about your assumptions?” The testing question fuels a dialogue about the uncertainties involved that itself helps evaluate the hypothesis. Finally, as you test and revise your hypotheses, they will be more open to changes as your knowledge develops. You may even find them actively asking whether the hypotheses have changed.

Conversely, once you satisfactorily answer the assumptions question, it is unlikely that anyone will ask you about it again. It will be accepted as truth, at least, until the sales figures come in, and have proved the assumption wrong. At that point, you’ve sold only 32,398 widgets, and the company is writing the project off as a loss.

Hypothesis testing for advanced practitioners: As Eric Ries has suggested in “Lean Startup”/2/, effective startup organizations will build hypothesis testing into the products themselves. By setting up feedback loops and evaluating the customers’ real use of the product, startups can “pivot” their products in response to those results. As Ries writes, “Those who learn the fastest win.”

It’s all about learning, not about truth. Learning through hypothesis testing manages uncertainty. Assuming the truth just gets you into trouble.



Notes

  • /1/ The best introduction to Monte Carlo remains: Douglas W. Hubbard, How to Measure Anything: Finding the Value of Intangibles in Business, Third edition (Hoboken, New Jersey: John Wiley & Sons, Inc, 2014), p. 127ff.
  • /2/ Eric Ries, The Lean Startup: How Today’s Entrepreneurs Use Continuous Innovation to Create Radically Successful Businesses (Crown Business, 2011).

Keine Kommentare:

Kommentar veröffentlichen