Test Before You Bet: 6 Rules for Running a Pilot Before You Launch a New Product or Market

On April 23, 1985, Coca-Cola replaced the most famous formula in the world. The company had not guessed. It had run roughly 200,000 blind taste tests, and the new recipe beat both the old one and Pepsi. Seventy-nine days later, after a flood of angry letters and calls, the original was back on shelves as Coca-Cola Classic.

The test was not too small. It measured the wrong thing. People preferred a sip of the new drink. Nobody had asked what they would feel when the old one was taken away.

A pilot is only as good as the question it is built to answer, and most companies build pilots that answer the question they hope to hear. That is why so many new products, new markets and new locations are "validated" and then fail at scale.

Now look at the opposite case. Target entered Canada in 2013 with no real pilot at all. It opened stores at a pace few retailers had attempted, ran into empty shelves and broken supply chains, and in January 2015 announced it would close all 133 stores. About 17,600 people lost their jobs and the company took a write-down of roughly $5.4 billion. A handful of stores run for a year would have exposed most of those problems for a fraction of the cost.

In a group that has been building businesses since 1890, across wine, real estate, hospitality and mobility, I have learned that the expensive mistakes are rarely the bold ideas. They are the bold ideas launched at full size before anyone knew what would break. Here are the six rules I apply before we bet real money on something new.

Rule 1: Write the kill criteria before you start

Most pilots end in an argument, because nobody agreed in advance what failure looks like. The team that built the idea reads every number as encouraging.

Decide, in writing and before day one, the number that will make you stop. Not a vague "we will see how it goes," but something like: if repeat purchase is below 20% after eight weeks, we do not expand. If the new location does not cover its direct costs by month six, we close it.

Write three outcomes, not two:

• Scale: the result that justifies a full launch.
• Adjust: the range where you change one variable and test again.
• Kill: the result that ends the project, with no second chance.

The kill line is the most valuable sentence in the whole plan. It protects the company from the sunk-cost logic that keeps weak projects alive for years. If you want the full framework for ending things, I wrote about it in which products you should kill.

Rule 2: Test the behavior, not the opinion

Coca-Cola measured preference in a blind sip. What mattered was loyalty, identity and habit. Surveys, focus groups and tastings tell you what people say. A pilot has to tell you what they do.

The only metric that predicts a full launch is a paid, repeated action by a real customer. Did they buy at the full price? Did they come back? Did a distributor reorder without being pushed?

When McDonald's tested its McPlant burger in the United States, it did not ask diners whether they liked the idea of a plant-based burger. From February 2022 it put the product on sale in about 600 restaurants in the San Francisco Bay Area and Dallas-Fort Worth. By that summer the company said the test had "concluded as planned," and analysts reported weak demand. The burger was not rolled out nationally. That is a pilot doing its job: a clear answer, at a cost that did not touch the rest of the business.

Rule 3: Make it small enough to lose, big enough to learn

Two mistakes kill pilots. The first is testing so small that the result means nothing: ten friendly customers, one enthusiastic distributor, one good weekend. The second is testing so big that failure hurts the core business.

Size a pilot by the answer you need, not by your enthusiasm. Ask how many customers, stores or orders you need to see before a result is more than luck, and stop there.

A few practical guidelines:

• Pick representative ground. Test in a typical market, not your best one. A pilot in your strongest city tells you how the idea performs where everything already works.
• Cap the downside in money. Set a maximum spend, including people's time, that you can lose without changing any other plan.
• Run it long enough to see the second purchase. Launch-week sales are curiosity. Week eight is the business.

Rule 4: Protect the pilot from the organization

Here is the uncomfortable part. A pilot inside a successful company is usually run by people who are also running the existing business. They give it leftover attention, or they give it heroic attention nobody could sustain at scale. Both results are false.

Staff the pilot the way you would staff the real launch, or the numbers mean nothing. If the full rollout will depend on the regular sales team, the pilot must depend on the regular sales team too, not on the founder personally closing the first accounts.

The same goes for special treatment. If the pilot store gets the best manager, extra marketing and weekly visits from headquarters, you have tested the manager, not the concept.

Rule 5: Instrument everything from day one

A pilot you cannot measure is a pilot you will argue about. Decide what data you need before the first sale, and set up the systems to capture it.

Track the full unit economics, not just revenue. A new product can sell well and still lose money once you count the discounts, returns, extra logistics and time from the team. For every pilot I want to see:

• Customer acquisition cost and how it trends week by week.
• Repeat rate or reorder rate.
• Gross margin after every real cost, including the hidden ones.
• Operational friction: what broke, how often, and what it cost to fix.

That last line is where Target Canada would have found its answer. Distribution problems do not show up in a forecast. They show up the first time a real truck is late.

Rule 6: Treat the scale decision as a new investment

A successful pilot does not approve the launch. It only earns the right to ask for the money.

Scaling is a separate capital decision, and it should compete for funding with every other project on the table. The pilot proved the idea works at small size. The scale plan must now prove it can work at large size, with the organization you actually have, at the returns you require. I use the same discipline I described in how we decide which capital projects get funded.

Taco Bell is a good example of patience paying off. The Doritos Locos Taco went through a three-year development process before its national launch on March 8, 2012. It then sold 100 million units in its first ten weeks, the most successful product launch in the chain's history at the time. The years of testing were not a delay. They were the reason the launch could be that big.

Key Takeaways

• A pilot is only as good as the question it is designed to answer; define that question first.
• Write scale, adjust and kill criteria in advance, with numbers, before the first sale.
• Measure paid, repeated behavior, not opinions, surveys or tasting scores.
• Size the pilot so you can afford to lose it but large enough that the result is not luck.
• Staff and support the pilot as you would the real rollout, or the data is false.
• Track full unit economics and operational friction, not only revenue.
• A successful pilot earns a scale proposal, not an automatic launch.

Frequently Asked Questions

What is a business pilot program?

A business pilot program is a limited, time-bound launch of a new product, service, location or market, run with real customers to test whether it works before committing full resources. It differs from research because customers actually buy and use the offer. The goal is a clear decision: scale, adjust or stop.

How long does a pilot program typically last?

Most commercial pilots run between two and six months, long enough to see repeat purchases and operational problems rather than only launch-week curiosity. Software pilots can be as short as a few weeks. Set the length by how long it takes a normal customer to buy a second time.

How do you test a product before launch?

Start with concept tests and prototypes to remove obvious flaws, then sell the product at full price to real customers in a limited, representative market. Measure purchase, repeat rate and margin, not stated interest. Compare the results against kill criteria written before the test began.

What are some examples of test marketing?

McDonald's tested its McPlant burger in about 600 US restaurants in 2022 and chose not to roll it out nationally. Taco Bell developed the Doritos Locos Taco for three years before a national launch that sold 100 million units in ten weeks. Consumer brands also commonly test in single cities or regions before going national.

What is a better term for a pilot program?

Depending on the context, companies also call it a test market, soft launch, proof of concept, beta or limited rollout. The name matters less than the discipline: a defined scope, a fixed budget, measurable success criteria and a decision date.

One Thing to Do This Week

Pick the new initiative your company is most excited about right now. Write one sentence that starts with "We will stop if…" and put a number in it. If nobody can agree on that sentence, you are not ready to launch. You are ready to design a pilot.

Discover how Manzanos Enterprises builds and tests new businesses across a group founded in 1890 and now selling in more than 75 countries.