Strategic Learning & Experimentation

Foundations of Business Development - NIT Northern Institute of Technology / TUHH, Hamburg · part of my Technology Management MBA · study notes for revision.

---

Every chapter so far assumed you could reason your way to a good plan: spot the opportunity, understand the value, choose a strategy. This last chapter admits an uncomfortable truth - at the start, almost everything in that plan is a guess. You do not yet know who the customer is, whether they feel the pain, whether they will pay, or whether your channel reaches them. So the real skill is not writing a confident plan; it is learning your way to the right one by spending small amounts of money to find out which guesses are wrong before they can hurt you.

Tip

The definition to hold on to: a young venture runs on assumptions, not facts. The biggest risk is not that one assumption is wrong - it is spending your whole budget before discovering it was wrong. Strategy under uncertainty is therefore a process of buying information: run the cheapest experiment that could disprove your riskiest assumption, learn, and only then commit.

1 · Why strategy here is a learning problem

In a stable business you can analyse your way forward, because the facts sit in the past and you can measure them. A new venture has no past to measure. Its plan is a stack of hypotheses - testable beliefs - and the probability that the whole idea is actually valuable is unknown and, worse, cannot be fully judged from the sidelines. Gans and Stern (2017) name four features of the entrepreneurial situation that make this hard:

Freedom many paths

• More than one way to create and capture value from the same idea - no single obvious route

Constraint scarce resources

• You cannot afford to walk every path at once, so you must choose where to spend

Uncertainty unknown value

• Whether the idea is valuable is genuinely unknown, and cannot be fully learned without some commitment

Irreversibility doors close

• Moving down one path with commitment changes the conditions for the others - some become unviable

The entrepreneur’s context (Gans & Stern, 2017): lots of freedom, too few resources, deep uncertainty, and choices that quietly foreclose other choices. That combination is exactly why you learn before you commit.

The phrase to remember is “commitment-free learning is noisy.” You cannot fully evaluate an idea by thinking about it; at some point you have to put a little skin in the game to get a clean signal. The art is doing that in small, cheap, reversible steps first.

2 · Assumptions and the riskiest-assumption test

Start by writing your plan down as a list of the beliefs it secretly depends on. A useful prompt set - from Aulet’s “leap-of-faith” style questions - walks across the whole business model:

• Who is the customer? (which specific segment)
• What value do you bring them? (the problem you solve)
• Will they pay - and how much profit does one customer bring? (willingness to pay, lifetime value)
• How will they get the product? (the channel)
• What does it cost to win a customer? (acquisition cost)
• Can the technology actually deliver it? (feasibility)

You will end up with a dozen assumptions. You cannot test them all, and you should not try to - some are safe and obvious, some are terrifying. The move is to rank them by two things at once: how uncertain the assumption is (how likely it is to be wrong), and how much rides on it (how badly the venture is hurt if it is wrong). This is the HiLo or risk-assessment grid.

↑ High impact if wrong

High impact, low uncertainty watch

• Matters a lot but you are fairly sure - keep an eye on it, do not spend a test on it yet

High impact, high uncertainty test first

• The leap-of-faith assumption - critical and shaky. This is where your first experiment goes

Low impact, low uncertainty ignore

• Safe and minor - do not waste a euro of learning budget here

Low impact, high uncertainty later

• Genuinely unclear but not fatal - park it until it starts to matter

Low uncertainty ← → High uncertainty  ·  ↓ Low impact if wrong

Prioritising assumptions on impact against uncertainty. The top-right cell - most critical and most uncertain - is the one leap-of-faith assumption to test before anything else. Everything survives or dies on getting that one right.

Note

Why “most uncertain and most critical” first, not “easiest first.” It is tempting to test the assumption you can check quickly. But if a different assumption is the one that would sink the venture, every euro you spend before checking it is at risk. Testing the riskiest thing first means that if the idea is doomed, you find out cheaply and early - the single most valuable outcome an experiment can give you.

3 · The build-measure-learn loop

Once you know which assumption to attack, you turn it into an experiment. Lean start-up (Eric Ries) frames this as a fast loop: build the smallest thing that produces a signal, measure what real people do with it, learn whether the assumption held, then go round again. The point of the loop is speed and cheapness - how much you learn per euro spent.

Assumption → Hypothesisa falsifiable belief

→

Buildcheapest experiment that could disprove it

→

Measureagainst a metric set in advance

→

Learnpersevere, pivot, or perish

The build-measure-learn loop. The output is not a product - it is validated learning: evidence that an assumption is true or false. The loop feeds back: what you learn reshapes the next hypothesis.

Step by step

1. Turn the assumption into a falsifiable hypothesis. A good hypothesis says two things: what you believe, and how you will know if you are right. Template: “We believe that people like customer type have a need to do behaviour; we will know we are right when we see this measurable outcome.” Make it sharp enough that a result could clearly fail it - a belief you cannot disprove is not a hypothesis, it is a hope.

2. Design the cheapest experiment that could disprove it. Reach for a stand-in, not the real product: a pretotype (a fake of the thing that tests demand before you build it), a landing page, a concierge test, a fake-door test, or a genuinely minimal MVP. The best experiment is easy to build, looks real to the customer, changes easily, and wastes as little effort as possible for the learning it returns.

3. Set the success and kill metric in advance. Decide before you run it what result counts as a pass and what counts as a fail - a target threshold, written down. Deciding after you see the data is how founders talk themselves into keeping a dead idea alive.

4. Run it, then decide. Compare observed against expected, and act: clearly above target is a green light; a serious shortfall means dig into why; nowhere near target means the hypothesis is invalidated. Design the test to disconfirm, not to flatter - you are hunting for evidence you are wrong, because that is the evidence that actually saves money.

Caution

Confirmation bias is the enemy. The natural instinct is to build a demo that makes people smile and call it validation. A polite “nice idea” is not a signal. A real experiment puts the customer in a position to say no with their behaviour - no click, no deposit, no signature - and takes that no seriously. If nothing could have failed your test, it proved nothing.

A catalogue of cheap experiments

Different assumptions need different tests. The trick is matching the experiment to the question - problem, solution, or willingness to pay - at the lowest cost that still gives a trustworthy answer.

Experiment	What it is	Best for testing
Customer interviews	Structured conversations about a real pain	Is the problem real and painful?
Landing page (“smoke test”)	A page describing the offer with a call to action, before the product exists	Interest and message - do people click and leave an email?
Fake-door test	A button or “buy” link for a feature that is not built yet	Demand for a specific feature, at almost zero cost
Concierge MVP	Deliver the service by hand, one customer at a time	Does the solution actually help? What does delivery involve?
Wizard of Oz	A real-looking front end with humans doing the work behind it	Solution demand without building the automation
Explainer video / crowdfunding	Show the value proposition; ask for money or pre-orders	Willingness to pay - the strongest signal of all
Single-feature MVP	Ship only the one core feature to real users	Whether the core value holds up in real use

The highlighted row is worth calling out: a fake-door test buys a real behavioural answer (“would you use this?”) for the price of a button. That is the spirit of the whole toolkit - the cheapest instrument that still tells the truth. Famous examples: Dropbox tested demand with an explainer video before building sync; Zappos photographed shoes in local stores before holding any inventory; Airbnb ran a concierge-style piecemeal service before automating anything.

Working backwards to a threshold

A metric only means something against a target. The clean way to set one is to work backwards from the business you need. If you eventually need forty paying customers a year and your realistic funnel converts, say, five percent of leads to trials and five percent of trials to purchases, then you can compute how many leads and how much reach an experiment must show to be on track. Set that number first; then the experiment either clears the bar or it does not.

Observed meets or beats targetstrong positive signal - the belief holds

↓

Observed falls well shortassess why - a fixable miss or a real flaw?

↓

Observed nowhere near targethypothesis invalidated - change something

Reading the result against the pre-set threshold. Only one of these is a green light; the other two send you back into the loop with a sharper question.

4 · Validated learning and the pivot

Now the decision that closes each loop - Ries’s blunt trio: persevere, pivot, or perish. Persevere means the evidence backs your current path, so keep going and test the next assumption. Perish means the core idea is dead; stop. The interesting one is the pivot.

A pivot is a structured change to one part of the strategy while keeping one foot on the ground - you hold on to what the evidence validated and change the piece that failed. It is not a random restart and it is not stubbornly carrying on; it is turning a failed experiment into a new, better-aimed iteration.

Persevere evidence supports the path

• The assumption you tested held against its threshold
• Keep the strategy; move on to the next riskiest assumption
• Danger: persevering out of sunk cost, not evidence

Pivot keep one foot grounded, change the other

• The assumption failed, but you learned something usable
• Change one element; keep what the evidence validated
• Danger: pivoting so often you never actually commit

Persevere vs pivot is a decision made against a metric you set beforehand - not a mood. Both have a failure mode: clinging on when you should turn, and turning so restlessly you never build anything.

Pivots come in recognisable types, depending on which single element you change:

Pivot type	What changes	What you keep
Customer segment	Who you serve - a different group has the pain	The problem and roughly the solution
Problem	Which need you solve for the same people	The customer you understand
Solution	How you solve it - a different product form	The customer and the validated problem
Business model	How you make money - pricing or revenue logic	Product and customer
Channel	How you reach and deliver to the customer	Product, customer, and value

The story every founder knows: YouTube began as a video-dating site, learned people only wanted the video sharing, and executed a problem pivot. Instagram was a cluttered check-in app whose one loved feature was photos - a solution pivot down to that single feature. In both cases the pivot was validated learning in action: keep the part the evidence liked, drop the rest.

Note

Lean has real limits - know them. Endless small experiments can trap you in “local search”: you optimise a mediocre idea to a slightly-less-mediocre one and miss the bigger move nearby. Add the sunk-cost fallacy (spending because you already spent) and escalation of commitment (doubling down to justify past bets), and you see why learning alone is not a strategy. Lean is a superb way to optimise within a strategy; it does not, by itself, tell you which strategy to pick.

5 · Commitment vs flexibility - the real tension

Here is the hard part the module builds toward. Experiments are wonderful because they keep options open - you learn without betting everything. But a start-up cannot win on optionality alone. At some point you have to commit: commitment is what attracts investors and talent, creates focus, and builds a defensible position rivals cannot casually copy. A venture that only ever experiments, hedging every bet, drifts - it never accumulates the advantage that comes from going all-in on one path.

Flexibility cheap, reversible, early

• Learn with little or no commitment - options stay open
• Lower accuracy, but negligible sunk cost if you are wrong
• Right when uncertainty is high and doors are still open

Commitment bold, costly, later

• Move with real resources - builds focus and advantage
• Higher fidelity, but sunk, largely irreversible costs
• Right once the critical assumptions are actually validated

The two are not enemies to balance evenly - they are a sequence. Learn cheaply while doors are open; commit boldly once you have earned the confidence to close them.

The resolution is the real-options view: treat the venture as a series of staged investments. Early on you buy cheap, reversible peeks - the small experiments of sections 3 and 4 - because each one reduces risk for very little money. As the riskiest assumptions get validated, you increase commitment: bigger, less reversible investments, now justified because the uncertainty has shrunk. Every stage asks the same question: has the new learning bought enough risk reduction to justify committing more - or should we STOP.

Cheap, reversible tests”take a peek before walking through a one-way door”

↓

Riskiest assumptions validatedthe leap-of-faith beliefs now backed by evidence

↓

Bold commitmentstage up resources - focus, resources, defensibility

Good strategy sequences the two: reversible learning first, irreversible commitment once earned. This is how you keep real options without endlessly drifting.

This is also why the module distinguishes strategic experimentation from plain lean experimentation. Lean, trial-and-error testing optimises within a chosen strategy at almost no opportunity cost. Strategic experimentation is about choosing between strategies when the very act of testing requires partial commitment and can foreclose alternatives. An ideal experiment, in that harder setting, tests a critical assumption with high fidelity (a trustworthy signal) at low opportunity cost - and you sequence experiments so the irreversible commitments come as late, and as well-informed, as possible.

Caution

When you skip the peek: the graveyard. Better Place raised roughly 850 million dollars to build battery-swap networks and sold under a thousand cars before liquidating - a giant irreversible bet placed before the demand was proven. Webvan burned about a billion dollars building integrated grocery warehouses, while Peapod ran the same idea as a lightweight complement to existing supermarkets, stayed small and reversible, and survived. The lesson is not “never commit” - it is “do not commit big through a one-way door before the evidence justifies it.”

6 · Tying the whole module together

Step back and the entire course is one arc. You learned to recognise an opportunity in the changes around you (Part A); to understand the customer value and the market; and to make coherent strategic choices about customers, technology, competition, and identity. This final chapter adds the piece that makes all of it survivable in the real world: none of those choices is a fact when you make it, so you learn your way to the right version - testing the riskiest assumptions cheaply, disconfirming rather than flattering, pivoting on evidence, and committing hard only once you have earned the confidence to close the door behind you.

Recognise the opportunitychanges, sources, ideas

↓

Understand value & marketcustomer value, research

↓

Make coherent strategic choicescustomers, tech, competition, identity

↓

Learn your way to the right versiondisciplined experiments, staged commitment

The whole module in one line: recognise → understand → choose → learn. Strategy is not a plan you write once; it is a set of bets you validate, one cheap experiment at a time, before you commit for real.

Revision summary

Tip

The one-minute recap

• A new venture runs on assumptions, not facts. The fatal risk is spending everything before finding out a key assumption was wrong - so strategy under uncertainty means buying information to reduce the riskiest unknowns first.
• Rank your assumptions on impact against uncertainty. Test the most critical and most uncertain one - the leap-of-faith assumption - before anything else.
• Run the build-measure-learn loop: turn the assumption into a falsifiable hypothesis, build the cheapest test that could disprove it (pretotype, landing page, fake-door, concierge, Wizard of Oz, MVP), set the pass/kill metric in advance, then decide. Design to disconfirm - beware confirmation bias.
• Close each loop with persevere, pivot, or perish. A pivot changes one element (customer, problem, solution, business model, channel) while keeping what the evidence validated. The goal is learning per euro spent.
• Commitment vs flexibility is a sequence, not a balance: cheap reversible tests early, bold irreversible commitment once the critical assumptions are validated. Keep real options without drifting - and never walk through a one-way door before the evidence justifies it (Better Place, Webvan).
• The module in one arc: recognise → understand → choose → learn your way to the right version.

That completes the module. Back to the course overview - from spotting an opportunity to learning your way to a winning strategy.