The One-Three-Five Year Experiment
Forecast your environment at 1, 3 and 5 years, then run today's informing experiment
- Difficulty
- Moderate
- Time to result
- ~months to results
- Steps
- 7
- Confidence
- 74%
This is Hoffman's answer to what a founder can do today to be more profitable tomorrow, and he starts by rejecting the framing — not literally tomorrow. The mechanism is two-stage. First, forecast the environment your business will occupy at one, three and five years, across three pattern classes: technology, competition, and delivery including supply chain. Note he forecasts the environment, not the outcome, and is explicit that overly specific predictions are how you look foolish. What you produce is a probable guess. Second — the load-bearing move — use that guess to design an experiment you can run today that informs which big choices you need to make across those horizons. Hoffman is careful about the payoff: the experiment won't necessarily make you profitable. Its output is information. It either points at the thing that gets you to more profit, or it sorts which things in your future market are living creatures and which are dinosaurs.
Origin
Hoffman offered this as his answer to Young and Profiting's closing question on what an entrepreneur can do today to become more profitable tomorrow, describing it as part of how he thinks about the world.
Core principles
- 01Forecast the environment, not the outcome — environments are more predictable than results
- 02Three forces drive environmental change: technology, competition, and delivery
- 03A probable guess is enough to design an experiment against
- 04The experiment's job is to inform a choice, not to produce profit
- 05Big choices made blind are how you end up a dinosaur
- 06Overly specific predictions are the fastest way to look foolish
How to run it
- 1
Set the three horizons
Write down the environment your business will be operating in at one year, three years, and five years. Environment, not results.
Pro tip Hoffman's own guardrail: the future is sooner and stranger than you think. Bias the one-year column toward sooner.
Watch out Overly specific predictions are, in his words, a great way to look foolish in the future. Aim for shape, not detail.
- 2
Run the technology pattern
For each horizon, ask what the environment changes based on patterns of technology. Extrapolate the pattern rather than naming a product.
Pro tip People in the 80s predicted AI and got the internet and mobile phones instead. The direction was right; the instantiation was not.
Watch out Do not bet the forecast on one named technology arriving on schedule.
- 3
Run the competition pattern
For each horizon, ask what changes based on patterns of competition. Who can do what you do, and what does the cost of doing it look like at each horizon?
Watch out Your current competitor set is the least useful input here. The pattern usually introduces new entrants.
- 4
Run the delivery pattern
For each horizon, ask what changes in the patterns of how you deliver your business — supply chain and everything else in the path to the customer.
Pro tip Delivery is the most-skipped column and often where the disruption actually lands first.
- 5
Name the big choices
From the probable guess, identify which big choices the one, three and five year time frames will force you to make. These are the decisions the experiment must serve.
Pro tip If a forecast implies no choice, it is trivia. Cut it.
Watch out Skipping this step turns the whole exercise into a forecasting hobby with no action attached.
- 6
Design the informing experiment
Ask Hoffman's actual question: how do you run the experiment today that informs you about which big choices you need to make in that one, three and five year time frame?
Pro tip Grade the experiment on information yield, not revenue. He is explicit that it won't necessarily make you profitable.
Watch out An experiment that cannot change any of the big choices from step 5 is not an experiment. It is just more work.
- 7
Sort the creatures from the dinosaurs
Read the result. The experiment either gives you the thing that gets you to more profit, or it helps you navigate the changes by revealing which parts of the future market are new living creatures and which are dinosaurs.
Pro tip Both outcomes are wins. The dinosaur identification is often worth more than the upside finding.
Watch out Do not re-run the experiment hoping for the profitable answer. Its job was information, and it did it.
In the wild
Around 2014-2015 Hoffman ran exactly this pattern on himself. The technology pattern was clear: not a new algorithm — the fundamentals were somewhat there already — but the convergence of scale compute, learning machines, and data to learn from. His probable guess about the environment: every computational device loses its interface and manual, everybody has multiple agents, anything with computational units becomes much more intelligent. He did not predict a specific product. He then made the big choice: he went to his partners at Greylock and said he was going to focus entirely on AI, while telling them to keep doing the crypto work that was still making money.
→ An environmental forecast, held loosely on specifics, drove a full allocation shift years before the wave arrived.
Hoffman's argument for forecasting environments rather than outcomes. Go back thirty years and tell someone there will be jobs called web designer and data scientist, and they will call you a crazy person from the future. The specific instantiation was unforecastable. But the environmental pattern — computation spreading into everything, creating roles around it — was legible. He expects the same again: oh, didn't realize that was going to be the job. And it's cool.
→ Demonstrates why the method forecasts the environment and holds the specifics loosely.
Common mistakes
Grading the experiment on profit
Hoffman explicitly says the experiment won't necessarily make you profitable. Killing it for not producing revenue destroys the only thing it was built to produce — information about the big choices.
Forecasting outcomes instead of environments
Specific predictions are, in his words, a great way to look foolish. People in the 80s predicted AI and got the internet — the pattern was right, the instantiation wrong. Forecast the environment and hold products loosely.
Forecasting without designing the experiment
The one-three-five picture is worthless on its own. The whole mechanism is the bridge from probable guess to an experiment you can run today.
Is it for you?
Best for
Entrepreneurs and operators whose market is being reshaped by a technology or competitive shift and who need to act now on a future they cannot yet see.
Not ideal for
Stable, slow-changing markets, or situations demanding an immediate decision with no room to run an experiment first.
From the transcript
“What do you think the environment your business is going to be in in 1, three, and five years? And what do you think that…”
“By looking at that kind of like probable guess, how do you run the experiment today that informs you about which big choices you need…”
“The experiment won't necessarily make you profitable, but the experiment may give you the thing that either gets you to more profit or also helps…”
From the episode
Reid Hoffman: Unlocking Limitless Human Possibilities With AI
Reid Hoffman