One-Way vs Two-Way Door Decisions: A Better Framework for Making Business and Technology Decisions

Popularized by Amazon founder Jeff Bezos, the “one-way door” and “two-way door” framework has become one of Amazon’s most useful principles for making decisions quickly without taking unnecessary risks. Bezos formally explained the idea in Amazon’s 2015 Shareholder Letter, distinguishing between consequential decisions that are difficult to reverse and decisions that can be changed, tested, or rolled back. The philosophy later became deeply embedded in Amazon’s management culture and is also widely used in AWS guidance around architecture, experimentation, and innovation.

At its core, the framework asks a surprisingly simple question:

If this decision turns out to be wrong, can we easily go back?

That question can fundamentally change how businesses, executives, software engineers, data engineers, and architects approach decision-making.

Every organization makes hundreds of decisions.

Some involve technology. Some involve people. Others concern products, pricing, infrastructure, data, or long-term strategy.

But not all decisions deserve the same amount of time, analysis, or approval.

This is exactly where the concept of one-way doors and two-way doors becomes valuable.

What Is a One-Way Door Decision?

A one-way door decision is difficult, expensive, or sometimes practically impossible to reverse.

Once you walk through the door, going back may not be realistic.

Think about cracking an egg.

Once you crack it, you cannot change your mind five minutes later and return it to its original state.

That is a simple example of a one-way door.

In business, one-way doors are usually decisions with significant consequences.

Examples might include:

• Selling a major business division
• Acquiring another company
• Permanently shutting down an important product
• Signing a restrictive long-term contract
• Migrating away from a platform when the original data or infrastructure cannot easily be restored
• Making a large capital investment with limited resale value
• Changing a core business model
• Making an architectural choice that creates extreme vendor lock-in

Because these decisions are hard to reverse, they deserve deeper analysis.

The organization should carefully consider cost, risk, alternatives, dependencies, opportunity cost, long-term consequences, and possible failure scenarios.

What Is a Two-Way Door Decision?

A two-way door decision is reversible.

You can walk through the door, observe the outcome, and walk back if it does not work.

Suppose a team stores data using the Amazon S3 Standard storage class.

Later, the team realizes that some of the data is rarely accessed.

It can transition that data to another S3 storage class that better matches its access pattern and cost requirements.

The original decision did not permanently trap the organization.

It could be changed.

That is a two-way door decision.

Other examples include:

• Testing a new landing page
• Running an A/B test
• Introducing a feature behind a feature flag
• Trying a different internal workflow
• Changing cloud instance sizes
• Testing a new dashboard
• Running a small pilot program
• Experimenting with a new marketing campaign

For these decisions, the cost of being wrong is relatively small.

That means the organization can move faster, learn from the result, and adjust.

Decide → Test → Measure → Learn → Adjust

That is very different from treating every choice as a major irreversible commitment.

The Most Important Lesson

The framework is not telling organizations to make every decision quickly.

It is telling them to distinguish between decisions that require extensive deliberation and those that do not.

Jeff Bezos described irreversible decisions as Type 1 decisions. These should be made methodically and carefully.

Reversible Type 2 decisions, however, can usually be made much faster by individuals or smaller teams. Amazon has repeatedly emphasized that treating too many Type 2 decisions like Type 1 decisions creates unnecessary bureaucracy and slows innovation.

And that leads to a powerful rule:

Do not use a one-way-door decision process for a two-way-door problem.

Why This Matters for Architecture

This concept becomes especially powerful when applied to software, cloud, enterprise, and data architecture.

Good architecture should not merely work today.

It should make tomorrow’s changes easier.

An architecture that allows:

• rollback,
• component replacement,
• incremental migration,
• experimentation,
• versioning,
• feature toggles,
• loose coupling,
• portable data,
• automated deployments,

creates more two-way doors.

By contrast, tightly coupled systems often turn ordinary technical choices into expensive one-way doors.

Changing one component requires changing five others.

Replacing a database requires rewriting an application.

Switching providers requires rebuilding the platform.

The system may technically function, but the organization has lost flexibility.

This gives us a deeper definition of good architecture:

Good architecture preserves options.

Turn One Big One-Way Door Into Several Two-Way Doors

Perhaps the most practical lesson from the framework is this:

When you encounter a one-way door, ask whether you can break it into several smaller two-way doors.

Suppose a company wants to replace its entire CRM.

Instead of migrating the whole organization at once, it could:

1. Choose one department.
2. Migrate a small dataset.
3. Run both systems temporarily.
4. Test integrations.
5. Measure productivity.
6. Collect feedback.
7. Expand gradually if the results are positive.

The original decision looked large and risky.

But the organization transformed it into a sequence of smaller reversible decisions.

Software engineers use exactly the same principle.

Instead of replacing an entire monolith:

1. Extract one service.
2. Route a small percentage of traffic to it.
3. Monitor the results.
4. Compare performance.
5. Increase traffic gradually.
6. Roll back if necessary.

That is architecture being used as a business risk-management mechanism.

Technology Gives Us Tools for Creating Two-Way Doors

Many modern engineering practices exist partly because they make change reversible.

Examples include:

• Feature flags
• Blue-green deployments
• Canary releases
• Infrastructure as Code
• Database backups
• Versioned APIs
• Containers
• Automated rollback
• Data replication
• Event-driven architectures
• Loose coupling

Consider a feature flag.

Instead of releasing a new feature permanently to everyone, an organization can enable it for:

• employees,
• beta customers,
• 1% of users,
• one geographic region,
• or one customer segment.

If something goes wrong, the feature can be disabled.

A potentially risky deployment has been transformed into a two-way door.

Decision Speed Should Depend on Reversibility

A practical rule is:

For reversible decisions: move faster, experiment, measure results, and correct mistakes.

For difficult-to-reverse decisions: investigate more deeply, model consequences, understand dependencies, and involve the appropriate stakeholders.

The critical skill is not simply making decisions quickly.

It is recognizing which decisions deserve speed and which deserve caution.

A Practical Checklist

Before making an important decision, ask:

1. Can we reverse it?
2. How much would reversal cost?
3. How long would reversal take?
4. What dependencies will this create?
5. What happens if we are wrong?
6. What happens if we delay the decision?
7. Can we test the idea on a smaller scale first?
8. Can we isolate the change?
9. Can one large decision be divided into several smaller reversible decisions?

These questions are useful far beyond software engineering.

They apply to business strategy, hiring, marketing, product development, investment, operations, and entrepreneurship.

Reversibility Can Become a Competitive Advantage

Organizations often say that they want innovation.

But innovation requires experimentation.

Experimentation inevitably produces some failures.

Therefore, an organization that wants to innovate must make failure inexpensive.

One of the best ways to make failure inexpensive is to make decisions reversible.

Instead of asking:

“How can we ensure that we never make a wrong decision?”

a better question is:

“How can we design our systems and processes so that a wrong decision is inexpensive to correct?”

That is a much stronger foundation for innovation.

Successful organizations will still make incorrect decisions.

The difference is that resilient organizations can recognize them, reverse them, learn from them, and move forward quickly.

Final Thought

Jeff Bezos’s one-way and two-way door framework is powerful precisely because it is so simple.

Before making your next major business or technology decision, ask:

Is this really a one-way door?

If it is, give it the analysis it deserves.

But if it is a two-way door, do not let unnecessary bureaucracy turn a reversible experiment into a month-long decision.

And whenever possible, design your business, your software, and your architecture so that you have more doors you can walk back through.