Startups, Secrets, and Abductive Reasoning

Guest post by Joseph Kelly.

But we must conquer the truth by guessing, or not at all.  CS Peirce

An early episode at at my last company demonstrated one of the paradoxes of startup product development.  At this time our product was still early and undefined.  I had spoken with a potential client about their goals for a project and was trying to create a sales proposal with the engineering team.

Pretty quickly I grew frustrated.  When I’d ask the engineers what we could do for a particular feature, every answer was “well, how does the client want it?”  I wanted to present the client something concrete, but being capable engineers, my team believed they could build anything.

This went on for several minutes before I broke the cycle and said: “If you’re a contractor and the client asks you to build them a gazebo, you don’t ask them everything from what roof pitch angle they want to what kind of screws to use.  You pitch one gazebo design, or a few, then you work together to reach a final version.”  That clicked instantly and we were able to move forward.  My anecdote forced us to adopt a lesser-known mode of reasoning that I’ll explore in this essay, called abduction, which is critical to developing your product strategy.

Product strategy is hard.  We have to make leaps of judgement with incomplete information about the customer and their preferences.  In the last decade we’ve learned new frameworks for how to do this, including the lean startup movement.  However, these models are missing some key ingredients that go into making great product decisions.

The Lean Startup

Lean startup theory says that a startup must define their initial product by some minimum viable criteria (a Minimum Viable Product or MVP), then work through Build-Measure-Learn cycles with the customer. Both the MVP and BML concepts are important ideas that should be in any product manager’s toolkit.  Yet when taken to the extreme, you get stuck just like my engineering team.

 

image07

The BML cycle from The Lean Startup.  Note that the frame is already bounded and determined: there is a clear box for the Target Market and Company.

In this extreme case there’s a recursive paradox in the process of developing your initial design or MVP:

What do you base your MVP on?  Customer feedback from a BML cycle.  

What is the BML cycle built on?  Your MVP.

Oh snap…

image03

How do we get out of here?

Reasoning in a Closed System

An analogy for the BML cycle comes from the task of early analog computers.  These were deployed first in ballistics, where anti-missile defenses needed to intercept enemy missiles in air.

In a simple case, the analog computer would guide the intercepting missile into the path of the enemy missile, so it would explode safely away from ground targets.  The exact trajectory of the enemy missile could not be known up front, so it was up to the intercepting missile to collect data and constantly adjust.

image01

Example of an analog compute cycle, or negative-feedback.  Note the parallels with the BML cycle.

This would also be called a cybernetic or a negative-feedback cycle.  Such cycles operate within a closed system, meaning it only operates inside some pre-defined elements.  In this case, just the two missiles.

By contrast, startups operate in a much more complex and open system.  Target customers and the product vision are often unknown, and that exact “enemy missile” to strike is vague or ill-defined.  Once picked, we can use the BML-cycle to narrow in on the exact target.  But picking any random target is not enough.  We need a framework to winnow down the potential targets to something that’s actually worth betting your company on.

Secrets in Open Systems

A great method for this can be sourced from Peter Thiel’s book, Zero to One.  In the book, Thiel describes great companies that start with a secret, defined as “something which you believe is true but nobody else agrees with.”  Thiel encourages us to look for secrets in the natural and social worlds and considers them essential to a company’s early strategy.

image02

Aiming for your own secrets, versus that which is already common knowledge.

“So when thinking about what kind of company to build, there are two distinct questions to ask: What secrets is nature not telling you?  What secrets are people not telling you?” -Peter Thiel, Zero to One

Founding your startup on a secret that emerges from this type of intuitive knowledge can be risky.  By their nature, secrets aren’t going to have validating infrastructure behind them to make their truth apparent.  The choice to believe in a secret is not a black-and-white ordeal, and feels more like an oscillation between believing in a conspiracy theory and doubt.  While secrets give us a much better chance at starting a successful company, they require a leap of faith and are not guaranteed.

The Three Kinds of Logical Reasoning

We can give this leap of faith a much stronger chance by supporting it with established techniques in logical reasoning.

Most everyone has heard of deduction and induction.  Simply: deduction is when you take a valid theory and apply it to a specific case.  Induction is when you take a specific case and generate possible theories.

image05

A “closed-loop” reasoning system, based on common knowledge (deduction) or incremental knowledge from experiments (induction).

When it comes to startups, deductive reasoning is not going to be effective at generating secrets, since it’s formed on the basis of some already accepted theory.  You’ll build the same thing as everyone else based on common knowledge about a market, product, or set of features.

Induction, by contrast, is most like the Learning step in the BML cycle: it takes measured results and generates possible explanations, theories, or prescriptions.  It is amplifying in the sense that it can add knowledge that wasn’t there before (insights into customer behavior for instance), but in the context of product development it’s limited to bounded domains where you can run experiments.

Yet there is a third, less-well known form of reasoning called abductive reasoning.  In abductive reasoning, you go directly from facts to a theory which accounts for the observation, also known as “inference to the best possible explanation.”  It was originally coined by Charles Sanders Peirce around 1900, but has grown in relevance, especially in the field of AI. The form of reasoning portrayed by the character of Sherlock Holmes is actually abductive, rather than deductive.

“Whereas deduction probes the boundaries of thought within a closed system, and induction structures evidence to support the formation of opinions, the abductive process involves the imaginative creation of explanatory hypotheses, generating alternative ‘may-bes’ in response to ‘what if’ inquiries.” –Elkjaer and Simpson, 2011

Abductive reasoning is critical to making progress as an early stage startup.  It’s your only hope for escaping the BML closed-loop cycle and finding significant secrets to build your company on.

Forming Abductions

A great example of abductive reasoning is what a doctor does when making a medical diagnosis.  Under perfect conditions, a doctor runs every test on their patient and consults the entire medical literature to compare results with established theories.  But when a patient is near-death, they can’t keep collecting data.  They must put a diagnosis forward or the patient dies.

Abductions have two components: an insight and a bet.  In the doctor’s case, their domain insight comes from years of training in diseases, injuries and their treatments.  They use this background knowledge to make educated guesses as to the right diagnosis.  Then they put their best diagnosis into practice, making a two-part bet: 1) that the patient’s symptoms match this particular disease from the medical literature, and 2) that the proposed treatment is the best one for this actual case.

image04

Insight + Bet are the key ingredients for making an abduction that leads to a foundational secret for your company.

The same ingredients are at work in developing a startup’s strategy, except the output is a secret.  The secret can be in the form of a unique customer niche (Tesla’s high end electric vehicle buyers), a novel product implementation (Google’s PageRank), or a go-to-market strategy (Facebook’s initial university focus).  

image06

Comparing different kinds of secrets in either a target market or product implementation.

According to Thiel a good company only needs a handful of secrets.  Let’s look at the building blocks more closely to see how they help us find them.

Insights

Insights are built on both deductive and inductive reasoning.  Consider these typical sources of insight for company building:

Watching the Majority

This kind of insight comes from observing the existing market and competitors.  You’re watching everyone zig one way, while you suspect it might be better to zag in the other direction.

Perhaps the market has decided to wash all their marketing materials with terms like cloud computing and big data, while you decide to focus on specific industry solutions.  You might decide to be different just because, or due to some other insight that tells you why the majority is wrong.

Domain Experience

Here you’ve gained insights as a result of experience in a company or industry that allowed you to observe a lot of BML cycles, yielding knowledge that is not yet common.  Maybe you worked at Google and decide to start your next company based on some machine learning or data management techniques you learned there.

VC’s talk about how they love funding entrepreneurs with experience in the domain they’re going after.  They’re trusting the founder’s experience to give them a better perspective on how to solve problems in the area.

Exclusive Access

In this case, you’ve developed insights due to your access to exclusive resources.  This might be through a personal network, which you can use to call up the CEOs of big banks to get their list of technology priorities.  Or maybe you have access to a laboratory that does one-of-a-kind research in a specific area.

Ultimately, all of these insights come from a mix of deductive and inductive reasoning.  Most often they involve simply paying attention and giving yourself time to think or gain experience.  What’s important is to use time and any other advantage to define the insight concretely and make sure it’s a real one.

Once you are confident your insight is real and has applications, you need to decide whether it’s something worth building your company on.  This judgement looks most closely like a bet.

Bets

Your startup is already a bet, whether you think about it that way or not.  Now that you have a collection of insights that might lead to secrets, you can evaluate them based on their likelihood to succeed and the magnitude of their potential success.

In addition to this analysis, there are great models to borrow from value investors like Warren Buffett, Howard Marks, and Seth Klarman.  Their ideas give us even more clues to look for when trying to make the right bet:

The Mispriced Bet

Great investors look for opportunities where the real risk is poorly understood and people think something is riskier than it really is.  Buffett and his partner Charlie Munger often compare stock picking to betting on horses.  Great horse bettors make gambles only when they see the market’s odds for a given horse diverge from the real odds they calculate themselves.

The example in entrepreneurship would be when a market opportunity is considered too hard or inaccessible by the majority, so they avoid it completely.  If you have some insight that tells you it’s not actually that hard, this makes for great betting criteria.

Contrarianism

If you want to get better than average market returns, you have to do something different than the rest of the market.  This is a fundamental truth in business, investing, and life.  If you want to have a better outcome than the average of a competitor group in a market, you need to be doing something different than them.

As Howard Marks describes, most great investors spend a lot of their time being lonely.  You will need to embrace and get comfortable with believing different things than the rest.  This can be especially hard when you are working through cycles that have a strong follow-the-herd mentality, such as VC fundraising.

Margin of Safety

Startups involve lots of risk, no question.  But there are plenty of risks that you don’t necessarily have to take.  If you are risk conscious up front and guard against extreme downsides, you can build a safety net for the business when it hits a wall in sales/hiring/fundraising/etc.

Great investors don’t just gamble on any mispriced opportunity, as Seth Klarman writes in his book.  They do it when the gap is huge, giving them a safety cushion when either their calculations turn out to be wrong, or market conditions take an unexpected turn.

Likewise an entrepreneur shouldn’t bet a whole company on any hint of a possible secret.  Good entrepreneurs act only when their secret, or ideally collection of secrets, is big enough to allow them to make a few mistakes along the way and still come out on top.

Building a Company

We make abductions all the time.  We’re just not aware of them.  When you walk into your kitchen and find water spilled on the floor, you immediately generate possible explanations: was it the sink?  Ice maker?  Five year olds?

The same thing happens in startups, except in a much higher-stakes scenario: bad guesses cost you and the company time and money.  If you pick the wrong initial market and the wrong MVP, you will just build a BML money pit for VC money.  Make the wrong abduction and the patient dies.

image00

Ongoing abduction and BML cycle inside a company.

If instead you step back and evaluate your abductive reasoning process, you might find that you are using assumptions and making guesses that may or may not be sourced from domain insight.  You can decide whether or not your company has any real secrets.

Great companies use this process continually.  They are constantly hunting for, validating and hoarding secrets.  And when things aren’t going well they trace back for where they made a wrong bet.

Because nobody wants to build the wrong gazebo!

Thanks to David Manheim, Daniel Schmidt, and Royal Frasier for their helpful feedback on this post.
Joseph Kelly is co-founder and CEO of Unchained Capital, a wealth management solution for bitcoin holders. Previously, he was co-founder and COO of Infochimps. Follow him on Twitter.

Get Ribbonfarm in your inbox

Get new post updates by email

New post updates are sent out once a week

About Guest

Ribbonfarm is a longform blog featuring a variety of themes and perspectives. The common factor is the general idea of "refactored perception." If you are interested in writing for us as a guest contributor, check out this page.

Comments

  1. That last diagram reminds me of the simplified OODA loop from “Certain To Win”:

    • Shlok Vaidya says:

      Yeah. Lean, as shorthanded by Reis et al, basically cribs from Steven Blank, who in turn admittedly cribs form Toyota and, in turn, Boyd.

      • Except Lean drops some detail that OP brings back (at least). Lean is the single level feedback loop Joseph starts with. I *think* the two-level loop Joseph shows unrolled at the end is the same as Boyd, but maybe I’m missing additional nuance.

        • I agree, that was my point. Joe adds back nuance that Lean canon dumps.

          I’d also add that Boyd cracked open the second O, Orientation, and that’s the key step, the rest is just execution/breathing.

        • Yep, there’s something there to how this rhymes with OODA. Though the last diagram is also meant to show how secrets accumulate (or are discarded for another), rather than a purely circular process.

          • Just to close the loop (heh!) on our offline conversation, I wanted to add the two ways we came up with to map your diagram to OODA loops. The first is compressed inside the Orient step:

            The second is stretched out over the whole loop:

            That’s a very interesting Necker-cube-like effect. Also reminds me of DNA coiled tightly inside the nucleus sometimes and stretched out all over the cell at other times.

  2. All inductions are always-already abductions . A lot of this turns on the Hermeneutic circle in Heideggerian terms (in b4 he’s a Nazi). There is never a period where we are making inferential judgements without some kind of pre-understanding: what you would call an insight. The difficulty is that pre-understanding is mostly given to us by convention (“The They”). This is the “*I* think this is a cool problem” problem.

    The gulf in the in the completion of your first MVP is this kind of hermeneutic problem. The real difficulty is getting out from your own assumptions, your pre-understanding, as to what you think is a problem to be solved, towards an understanding of a users actual problem.

    For this I really like the addition of Design Thinking into the Lean Startup equation. “Get Out of The Building” and beginning with proper ethnographic research that aims at the 5 Whys of a problem is a lot cheaper than a misguided MVP. Low-fidelity prototypes are not “minimally viable” but still give a lot of cheap insight.

    There is a lot of questioning of your abductive reasoning you can do before you spend money on a MVP. But it is qualitative data, the way people feel about and how they interpret their own problems. That can make some quantitative types feel woozy.

    • Thanks – I’ll check out some Heidegger and hermeneutics. I agree, there’s definitely a spectrum of inference where induction involves a much smaller step and abduction is on the other extreme. Still, it seems philosophers smarter than me find it worth drawing a distinction between the two.

      Another take on abduction calls it a “world creating device.” I think this parallels very well with what Design Thinking and your comment gets at. You’re trying to imagine the world your customers inhabit, and that might mean physically or mentally Getting Out of the Building.

  3. I agree with the design thinking suggestion, in the context of successfully pairing it with the BML/MVP phase. In fact one way I like to phrase it, because I’ve long been bothered by the emphasis on “build first” that comes with BML, is to add another L so that it’s LBML, adding a learning stage at the hypothesis or design thinking or abductive insight beginning. Using diverge and converge passes with desk research and field research helps my teams better understand the domain and the demand-side space, prior to decision making on what to build.

    I’ve also been having success lately adding in some Presumptive Design, using an artifact early on to provoke responses and improvements – to follow the example in the article, an early drawing of the gazebo, to accelerate breaking the chicken-and-egg cycle. It’s familiar paper prototype thinking, just moved as far upstream as possible, to day one ideally.

    And to also add a thumb’s up to the focus on “secret” insights, another way I’ve seen this put is something like “the secret of business is to know something that nobody else knows”. Those secrets are the gold nuggets that indicate a deeper vein to bet on, and you’re spot on, at some point there has to be a leap-of-faith bet, as informed and structured as possible, but still a bet!

    Great read, thank you!

  4. The opening gazebo anecdote reminds me of the Seinfeld episode with the kitchen cabinets contractor.

    Slightly related is customer-driven design, as in the car designed by Homer Simpson. It’s abduction, but from the “wrong” perspective — of the user rather than the designer/engineer. So it behaves more like inductive (one customer thinking their n=1 needs are the right needs for all car owners, and also operating blind with respect to engineering possibilities and constraints…the literal faster horse, except it’s not even that).