The single biggest fix for slow operations is not a smarter framework. It is sorting every decision into two piles first: the ones you can undo cheaply, and the ones you cannot. Reversible decisions should be pushed down and made in minutes. Irreversible ones deserve the meeting, the data, and your signature.

Most teams do the opposite. They agonise over the reversible calls (which vendor for a trial, how to lay out a shift roster) and rush the irreversible ones (a lease, a headcount plan, a data-migration cutover). The result is a calendar full of low-stakes debate and a handful of expensive mistakes that nobody had time to think through.

This guide gives you three practical tools that work together: a reversibility test to decide how much rigour a decision needs, a decision-rights map so people stop asking "who owns this?", and a simple rule for how much data to gather before you commit. Each one is something you can put in place this week.

Sort every decision by reversibility first

Amazon popularised the "one-way door versus two-way door" idea, and it holds up on the operations floor. A two-way door is reversible: if it turns out wrong, you walk back through and try again at low cost. A one-way door is hard or impossible to undo — you are committed once you step through.

Strong looks like this: a COO who names the door out loud. "This is a two-way door — pick something reasonable by Thursday and we will adjust." Or: "This is a one-way door, so I want the analysis and a dissenting view before we sign." The team learns to match effort to stakes without being told each time. Weak looks like this: every decision gets the same treatment. Either everything goes to committee (and the business stalls), or everything is fired from the hip (and a one-way door slams on you). Both are a failure to sort.

How to actually do it: before any decision, ask three questions. What does it cost to reverse this in three months? Who is harmed if we are wrong, and can we make them whole? Is there a point of no return baked in (a signed contract, a deleted dataset, a public announcement)? If reversal is cheap and nobody is badly harmed, delegate it and move on. Consider a warehouse trialling a new pick-path layout: if it slows things down, you revert in a day, so it is a two-way door and the shift lead should just decide. Compare that to migrating the order database to a new provider — reversing a botched cutover can mean days of downtime and lost records, so it earns a full plan and a rollback path.

The payoff is speed with safety. When you protect your rigour for the one-way doors, you can afford to move fast everywhere else — which is exactly where most of your decisions live.

Fix decision rights so nobody asks "who owns this?"

The second cause of operational drag is unclear ownership. A decision floats between three managers, each assuming another will call it, and a week passes with nothing decided. Reversibility tells you how much rigour a decision needs; decision rights tell you who actually makes the call.

The most reliable tool here is RACI — for any decision or task, name who is Responsible (does the work), Accountable (the single person who owns the outcome and makes the final call), Consulted (gives input before), and Informed (told after). The non-negotiable rule: exactly one Accountable per decision. Two accountable owners means no accountable owner.

Strong looks like this: a supply-chain reorder threshold has one accountable owner — the operations manager — who consults finance on cash impact and informs the sales team of the new lead times. Everyone knows their lane. The decision happens in a day. Weak looks like this: "the leadership team" owns it. That is code for nobody. Or the CEO is quietly accountable for forty operational calls a week and becomes the bottleneck for the entire company. A clear operating partnership between CEO and COO usually starts by drawing exactly this line: which decisions are the COO's to make alone, which need the CEO, and which the COO delegates further down.

How to actually do it: take your ten most frequent recurring decisions — hiring backfills, capital spend under a threshold, pricing exceptions, vendor selection, incident escalation — and write a one-page decision-rights grid naming the accountable owner and the dollar or scope limit for each. Delegate as far down as competence allows. A useful default: if someone will still be accountable for the result, give them authority over the decision. Publish the grid where the team can see it, and update it when roles change. This is also where you set escalation triggers, so a front-line manager knows the exact point at which a call stops being theirs and becomes yours.

Let data set the confidence bar, not the answer

Data does not make decisions. It tells you how confident you are allowed to be, and it turns arguments about opinions into arguments about numbers. The job is not "collect all the data" — it is "collect enough to clear the bar this decision requires."

Tie the data bar back to reversibility. A two-way door needs only enough evidence to start; a quick read of one metric is often plenty. A one-way door needs a real analysis: the base case, the downside, and what would have to be true for this to fail. Trying to hit the same evidence bar for both is how teams get analysis paralysis on trivial calls and thin evidence on the expensive ones.

Strong looks like this: the metric is agreed before the decision. "We will keep the new routing if average handle time drops by 10% over four weeks without a rise in errors." That is falsifiable — you set the threshold in advance, so you cannot rationalise a bad result later. This kind of discipline is the backbone of any data-driven operation, and it depends on having operational metrics you actually trust rather than numbers assembled after the fact to justify a preference. Weak looks like this: pulling a dashboard after the decision to find a chart that supports what the loudest person already wanted. That is confirmation bias wearing a data costume. It feels rigorous and is worse than no data, because it launders a gut call as evidence.

How to actually do it, for a decision that matters: (1) write the decision and the one or two metrics that would prove it right or wrong; (2) set the threshold before you look at results; (3) gather data from a source everyone accepts, so the debate is not about whether the number is real; (4) name the downside case explicitly — "if we are wrong, here is what breaks and how we recover"; (5) decide, record the reasoning in two lines, and set a review date. The written record matters more than it looks: six months later, when someone asks why you chose this vendor, the two-line rationale tells you whether the decision was sound and the outcome unlucky, or the decision was genuinely flawed. That distinction is how a team actually learns.

A practical limit on data-gathering: past roughly 70% of the information you wish you had, more analysis usually costs more than the certainty it buys — especially on a two-way door you can correct later. The skill is knowing which decisions deserve the extra 20% and which do not.

Where these three tools meet

Run them in order and they reinforce each other. Reversibility sets the rigour. Decision rights set the owner. Data sets the confidence. A two-way door with a clear owner and a quick metric check gets decided today. A one-way door escalates to a named senior owner, pulls a real analysis, and books a review. Under pressure this ordering is what stops a team from freezing — and it is why disciplined crisis decision-making leans on pre-agreed decision rights: you cannot invent who-decides in the middle of an incident. The same clarity makes any operational change land faster, because people accept a decision far more readily when they understand who made it and on what basis.

Key takeaways

• Sort by reversibility first. Two-way doors get pushed down and decided fast; one-way doors get the analysis, a dissenting view, and a named senior owner.
• One accountable owner per decision. Use RACI, delegate as far down as competence allows, and publish a decision-rights grid for your most frequent calls.
• Set the metric before you decide. A pre-agreed, falsifiable threshold prevents the after-the-fact chart-hunting that dresses up bias as data.
• Match the evidence bar to the stakes. Enough to start on a reversible call; a full downside case on an irreversible one.
• Write two lines of rationale and a review date. It lets you tell a bad decision from bad luck later — the only way a team improves.
• Stop gathering data near 70%. On reversible decisions, extra certainty usually costs more than the correction would.

Frequently asked questions

How do I tell a reversible decision from an irreversible one when it is not obvious? Ask what it costs to walk it back in three months, whether anyone is permanently harmed, and whether there is a hard point of no return like a signed contract or deleted data. If reversal is cheap and nobody is badly hurt, treat it as a two-way door and decide fast. When you are genuinely unsure, default to treating it as reversible but add a small, cheap safeguard such as a pilot or a rollback plan. Won't pushing decisions down cause chaos and inconsistency? Not if you pair delegation with clear rights and limits. Delegation without a decision-rights grid does cause chaos, but a one-page grid naming the accountable owner and the dollar or scope limit for each recurring decision gives people a fenced sandbox. Within their limits they move fast; past them, a defined escalation trigger sends it up. The COO's job shifts from making the calls to setting and maintaining the fences. What is the difference between the person Responsible and the person Accountable in RACI? Responsible people do the work; the single Accountable person owns the outcome and makes the final call. A team of three can be responsible for launching a new process, but exactly one person is accountable for whether it succeeds and answers for the result. If you cannot point to one name as accountable, the decision has no owner and will stall or drift. How much data is enough before I commit? Match the data to the stakes, not to your anxiety. A reversible decision needs only enough evidence to start, because you can correct course cheaply; an irreversible one needs a base case, a downside case, and a clear view of what would make it fail. As a rule of thumb, once you are near 70% of the information you wish you had, more analysis usually costs more than the certainty it buys — so spend the extra effort only on the one-way doors. How do I stop my team from using data to justify decisions they have already made? Set the success metric and its threshold before you look at any results, and write both down. When the bar is fixed in advance and falsifiable — "keep it only if handle time drops 10% in four weeks" — there is no room to hunt for a supporting chart afterward. Pulling a dashboard after the fact to confirm a gut call is confirmation bias in disguise, and a pre-committed threshold is the cleanest defence against it. Should the CEO be involved in operational decisions? Only the genuinely irreversible or company-shaping ones. If the CEO is accountable for dozens of operational calls a week, they become the bottleneck and the COO role is not doing its job. Draw an explicit line between decisions the COO makes alone, those that need the CEO, and those delegated further down, then keep the CEO's list short and reserved for the true one-way doors.