Affordable Loss

The forecast depends on assumptions the team hasn't tested

A small team believes a compact robot could automate a difficult warehouse task. The commercial story is attractive: many warehouses, expensive manual work and a large potential market.

The spreadsheet gives the idea a 30 percent chance of becoming a £20 million business. Someone changes it to 40 percent, another person doubles the market, and soon the team is arguing over numbers nobody can really defend.

The important uncertainty is much nearer. Can a compact machine cross an uneven loading-dock threshold while carrying its intended payload?

One response is to build the complete product before testing the assumption. Another is to keep researching till the estimates feel respectable. Both focus on the distant forecast instead of the next test the team can control.

Affordable Loss offers a different decision rule:

When you can't say what the upside is or how likely it is, decide what you can afford to lose and keep the next move inside that limit.

Set the first limit low enough that failure still leaves time, money, and trust for the next useful test.

The concept comes from Saras Sarasvathy's research on effectuation, a logic used by expert entrepreneurs under high uncertainty. They don't choose only by forecast return. They cap what they can lose at each step and start with the people, money and knowledge already available.

Set the exposure before you build

The team defines one affordable experiment:

• Four weeks of work from one engineer and one designer
• £12,000 for parts and fabrication
• A test rig, not a production deployment
• No customer inventory inside the trial area
• Stop if a plausible failure could injure a person or damage the warehouse
• A review before committing another block of time or money

They build only enough machine to test the threshold.

The prototype tips.

That's disappointing, and still far from a company-ending surprise. The test shows that the machine's weight and turning force at low speed caused the fall, so the next test can focus there before the team spends money on navigation. They learn this without interrupting warehouse work or putting employees and customers inside the test, and they still have the time and money for another try.

The team may redesign the drive system, choose a flatter environment, or stop. It cannot treat the £12,000 as a reason to spend another £120,000 simply because it has already been spent. Each new block of spending needs its own affordable-loss decision.

Cash is only one part of the loss

Money is easy to count, but the experiment can also consume time, attention, trust and safety.

Before acting, set boundaries in at least four dimensions.

Money and material

How much money, stock, or equipment could disappear without putting basic obligations or the next experiment at risk?

Time and attention

A cheap project can take the month needed for something more valuable. Name the people, the time, and what they are giving up. An experiment that fills everyone's head is not really part-time.

Trust and reputation

A failed internal prototype may be affordable. Testing an unreliable product on customers without consent can damage trust for years even when the financial loss is small.

Safety and reversibility

Ask whether harm can be contained, reversed or compensated. A loss affordable to the budget may be unacceptable to the person injured by it.

That produces a more honest question than “Can we afford the invoice?”

If this fails, who loses money, time, trust or safety, and how much?

Where the same rule is useful

Entrepreneurship and strategy. In a new market, there may not be enough repeated evidence to estimate the probability with confidence. A precise expected-return calculation can look more certain than the evidence supports. Affordable Loss replaces prediction with control: limit the initial downside, expose the idea to reality, and let customers, partners, and technical evidence shape what becomes possible next. The effectuation research community emphasises that two people can set different loss limits for the same venture because they have different money, obligations and room to recover.

Finance and portfolio design. An investor can limit how much money goes into one appealing idea so a failure does not end their ability to invest in later opportunities. The same rule applies to repeated bets: start with limited exposure, then increase it when the evidence improves. Product work can leave capabilities and relationships behind even when the original bet fails, but exhausted people and damaged trust don't recover as neatly.

Software and product work. A team can release a change to a small group first, keep a switch that turns it off, move data in stages, or test away from live customer data. Each choice limits how much a failed experiment can affect and keeps a clear way back. The limit must still match the harm. Testing with one percent of customers does not contain a privacy leak or a database change that can destroy data, because one bad event may already be irreversible.

In each case, limit how much one unproven assumption can damage so the next test is still possible.

When this model is useful

Use Affordable Loss when:

• The outcome range is wide and probabilities are poorly grounded.
• Acting can reveal something that research alone can't.
• The exposure can be staged, isolated or reversed.
• You expect to make several bets rather than one final bet.
• Failure would otherwise create pressure to continue through sunk costs.
• Other people need a clear promise about the risk they won't be asked to carry.

A practical sequence is:

1. State the decisive uncertainty.
2. Name the smallest action capable of testing it.
3. Set maximum exposure across money, time, attention, trust and safety.
4. Identify who bears each loss.
5. Define what evidence earns another block of investment.
6. Define the boundary that stops the experiment immediately.
7. Preserve enough resources for the next useful move.

Set the loss limit before the work starts, then review it only when new evidence changes the decision.

Where a cheap experiment tells you too little

An experiment can be too small to learn.

A team may give a difficult idea two afternoons, produce a predictable failure and claim the idea was disproved. The test was cheap because it was too weak to answer the important question.

Affordable Loss also becomes dangerous when the decision-maker declares someone else's loss affordable. A platform may tolerate a small fraud rate while the affected customer loses rent money. A founder may cap cash exposure while employees absorb nights and anxiety. Count the loss to every person affected, especially people who didn't choose the bet.

Some acts remain unacceptable even when the organisation could survive them. Safety, consent, legal rights and irreversible ecological damage are constraints, not prices on a risk budget.

Finally, downside discipline still leaves room for the upside. Use Affordable Loss while the forecast is mostly guesswork. Once the probabilities and returns have credible evidence, bring expected return and opportunity cost back into the decision.

Nearby models

Expected Value compares probability-weighted outcomes and chooses the option with the best estimated average result. Affordable Loss is useful when those probabilities and payoffs are too speculative to rely on. Affordable Loss starts with the amount you can lose and still continue.

Margin of Safety adds room for error around a credible plan whose basic economics or load estimates are believed to be sound; use it when the plan could be wrong at the edges. Use Affordable Loss when you are still testing whether the plan is worth pursuing.

Before taking the bet, check that a failure will teach you something and still leave enough capacity to use what you learn.