What the framework is
Every marketing team makes far more decisions than it can study properly. Some are small and easily undone: an email subject line, a homepage banner, a bid adjustment. Others are large and sticky: a new brand identity, leaving a sales channel, a change in pricing model, a long agency contract. The framework names the two kinds as doors. Through a two-way door, a team can step back if the room is wrong. Through a one-way door, it cannot, or only at a high price.
The point is to match the evidence to the stakes. The guide to research design and the research brief makes the same argument from the research side: the decision a study serves sets the standard of evidence it needs. The guide to choosing an attribution approach applies it to measurement, where moving a small creative budget can rest on rough attributed figures while moving a large share of the media plan cannot. This page is the classification itself: how to tell the two kinds of decision apart, how to route each one, and how to turn a one-way door into a sequence of two-way ones.
The failure the framework guards against runs in both directions. Teams that treat every decision as a one-way door research a banner for a month and fall behind. Teams that treat every decision as a two-way door rebrand on instinct and spend the following year undoing it.
The two kinds of decision
| Kind | Signs | Evidence needed | How to decide |
|---|---|---|---|
| Two-way door | Cheap to undo; affects a small audience or a short period; nothing is signed; customers barely notice a reversal | A clear expectation of what should happen and a measure to check it against | Act, watch on a set date, reverse if the result misses the expectation; test where traffic allows |
| One-way door | Costly or slow to undo; changes what customers expect; involves contracts, assets or public commitments; loses accumulated learning or reputation if reversed | Research into the reasons behind the choice, a test or pilot that isolates the effect, and a view of the downside | Make a smaller reversible version first; decide with a named owner on evidence agreed in advance |
Most decisions sit somewhere between, and the useful habit is to ask what makes a decision harder to reverse than it looks. Three questions do most of the work. What is spent that cannot be recovered, such as production costs, contract terms or sunk media? What changes in other people's heads, such as customer expectations of price, a partner's view of the relationship, or staff trust in a new process? What accumulated advantage is lost by reversing, such as search rankings, a platform's learning about who converts, or a list of subscribers who have been trained to expect a particular kind of email?
Classifying marketing decisions
Once a decision is classified, it follows one of three routes.
Act and watch. Most copy, creative, scheduling and bid decisions are two-way doors. The team writes a one-line hypothesisA testable statement of the change to make, the effect expected on a named metric and the evidence-based reason for expecting it. of what should happen, ships the change, and checks on a set date. A change that misses goes back. This route needs a measure and a calendar entry, not a meeting.
Test before committing. Decisions that are reversible in principle but expensive to get wrong at full scale, such as a new checkout offer, a landing page for the main paid campaign or a different discount structure, are worth an A/B testControlled comparison of two variants. where traffic is sufficient. The guide to average order value levers orders its levers partly this way: a free-delivery threshold or a set of cross-sells can be tried and withdrawn, while financing or volume pricing changes the store's economics and needs modelling first.
Research and pilot first. One-way doors, such as a rebrand, a new market, leaving a channel, a move onto a marketplace or a large annual commitment, go through research into why customers would respond, then a pilot that can be stopped. The research brief template records the decision the research serves, which keeps the study focused on what would change the choice rather than on everything that might be interesting.
Classification is a judgement, so write it down. A short line in the brief, saying which kind of door this is and why, lets others challenge the call before the work starts rather than after it goes wrong. The same classification helps a manager decide how much authority to hand over with a piece of work, as the guide to delegating marketing work explains. A marketer taking over a live channel they have never run can use the same sort to choose safe first changes, as the tutorial on learning a new marketing channel on the job shows.
Making one-way doors smaller
The most valuable use of the framework is not sorting doors but redesigning them. Many decisions that look irreversible can be broken into a reversible first step that answers most of the uncertainty.
- Pilot in one place. Launch in one region, one store group or one segment before committing everywhere, with a gate that decides whether to continue.
- Hold some customers back. A holdout groupA randomly chosen part of an audience kept from exposure to measure what exposure caused. that does not receive the change shows what would have happened without it, so the team can judge its incrementalityConversions caused by marketing. rather than its correlation with results.
- Run old and new side by side. Keep the existing channel, page or programme running at reduced weight while the replacement proves itself, instead of switching off the old one on launch day.
- Time-limit the commitment. A promotion with an end date, a contract with a break clause or a trial agency project is a two-way door; an open-ended version of the same thing is not.
- Change the inside before the outside. Test a new positioning in paid search copy and on landing pages before repainting signs, packaging and the brand guidelines.
Each of these costs something: pilots are slower, holdouts give up some revenue, parallel running doubles some effort. That cost is the price of keeping the door open, and it is worth paying only when the decision is genuinely hard to undo. A parallel runRunning the old and new versions of a process, platform or tag set side by side for an agreed period, compared on one measure, before the old one is retired. also needs an end date, or the doubled effort becomes permanent.
Worked example (illustrative)
A toy retailer sorts its quarter's decisions
Tumblewhit Toys, an invented online toy retailer with 38,000 email subscribers, lists 11 marketing decisions for the quarter and sorts them. Nine are two-way doors, among them subject lines, homepage banners, the order of categories in navigation, bid changes on 3 shopping campaigns and a new post-purchase email. Each gets a one-line expectation and a review date; two are run as A/B tests because they touch the checkout. One banner test lifts category clicks by 9% and stays; a subject line style cuts opens from 31% to 27% and is reversed after one send.
Two decisions are one-way doors. A rebrand would replace packaging, signage at 4 pop-up stands and a logo customers know; selling on a large marketplace would mean price matching and handing customer data to a platform. The rebrand goes to research first: 14 interviews with parents and a test of the new name in paid search ads. Interviews show the old name is trusted, so the team narrows the rebrand to a refreshed logo. The marketplace move is piloted with 60 products for 10 weeks, with 20% of comparable products held back. The pilot brings £8,400 of extra sales but also pulls repeat buyers off the retailer's own site, so the range is not extended.
Limits and criticisms
Reversibility is a spectrum, and it is often misjudged. Decisions sit along a continuum from trivially undone to permanent, and the two-door language can hide that. Teams also tend to misjudge in a consistent direction: the person proposing a change usually sees it as more reversible than the people who would have to clean up after it. Writing down what reversal would actually involve, and asking someone outside the proposing team to check it, catches the worst errors.
Reversible changes can carry hidden costs. A paid campaign switched off and on again may lose the delivery learning the platform had built up. A price cut withdrawn after a month can leave customers feeling misled. A new email style abandoned after a few sends may still have trained part of the list to ignore the brand. The door looked two-way, but walking back through it cost something real. Counting trust, data and learning as costs, not only money, makes the classification more honest.
Speed on two-way doors can become an excuse to skip measurement. "We can always reverse it" is true only if someone checks whether it should be reversed. A team that ships many reversible changes without expectations or review dates ends up with a site and a programme shaped by accumulated guesses, none of which were ever undone, because nobody looked.
The framework also says nothing about which option is right. It routes a decision to an appropriate level of evidence; the evidence still has to be gathered and read well, and a one-way door researched badly is no safer than one walked through on instinct.
Related frameworks
The experimentation loop is the machinery for the middle route, turning two-way doors with a measurable effect into ordered tests and recorded learning. For one-way doors that are split into steps, phase-gate reviews supply the decision points between the steps, with criteria agreed before each one. And for the act-and-watch route, control charts show whether a metric moved after a change by more than its normal week-to-week variation, which is how a team knows a two-way door needs walking back through.
Common mistakes
- One standard for every decision. Applying a rebrand's level of scrutiny to a subject line, or a subject line's to a rebrand.
- Counting only money. Lost platform learning, search rankings and customer trust make reversals more expensive than the budget line suggests.
- No review date on reversible changes. A two-way door nobody revisits works like a one-way door.
- Classifying after deciding. Calling a favoured idea reversible to avoid the research it needs.
- Skipping the smaller version. Committing to the full one-way door when a pilot, a holdout or a time-limited trial would have answered most of the question.
- Letting the proposer classify alone. The person who wants the change is the least likely to see what undoing it would cost.
