Phase Gates: How to Design Project Checkpoints

What is a phase gate, and what does it actually decide?

phase gates is a key topic in modern project management and teamwork. A phase gate is a scheduled review at the boundary between two phases, where a group with resource authority decides whether the next phase is funded. The decision is the product. The deck, the evidence pack, and the agenda exist to make it defensible.

The stage-gate model formalised this in product development. Robert Cooper’s version breaks innovation work into stages with a gate before each one, and defines every gate with three parts: required deliverables, criteria for judging the project, and defined outputs. Gates are staffed by gatekeepers, senior managers from the functions that own the next stage’s resources.

Two design choices matter more than the diagram.

First, criteria are set at the previous gate, not invented on the day. The team knows what “ready” means while doing the work, which removes the usual argument about whether the evidence is sufficient.

Second, gates get heavier as money gets bigger. In Cooper’s five-stage system, Gate 3 is the last point at which a project can be killed before heavy development spend. After that, the cost of stopping rises faster than the value of the information you gain.

Government and aerospace reviews follow the same logic under different names. NASA publishes entrance and success criteria before each key decision point, and its criteria are deliberately blunt: requirements must be “clear, complete, consistent, feasible, traceable, testable”. The US Department of Health and Human Services frames its stage gate outcomes as approve, conditionally approve, or not approved. Different vocabulary, same instrument: evidence in, decision out.

What decisions can a gate make?

Five outputs cover almost every real situation, and each needs a different written record. If your template offers only “approved” and “not approved”, you will get “approved” every time, because it is the only option that avoids a difficult conversation.

Conditional go is the most useful and the most abused. A real conditional go names three things: the gap, the owner, and the date. A conditional go that says “proceed, but tighten up the risk log” is a go with softer language. Treat it as a go in your reporting, and it will behave like one.

> Important note: every gate should record what happens to the money if the answer is kill. A gate where nobody knows which budget line is released is a gate that cannot afford to say no.

How do you design a phase gate in six steps?

Design the gate before the phase starts, in this order. Doing it in the reverse order, writing criteria after seeing the results, is how gates become rubber stamps.

Step 1: Find the irreversible commitments

List every point where the project commits money, headcount, or a contractual obligation that cannot be unwound cheaply. Those are your gates. A design approval that locks a supplier, a construction start that commits a site, a launch date that commits marketing spend: each deserves a gate in front of it.

Step 2: Split criteria into knockouts and scores

Knockouts are yes/no questions where a no stops the project. Scores are weighted judgments used to compare and prioritise. Keep knockouts to five to eight items, each answerable with evidence rather than opinion.

Step 3: Name one decision owner and the quorum

A gate needs a chair with authority to stop the work, and a defined quorum. If the person who can say no is absent, the gate is adjourned, not passed.

Step 4: Define the evidence pack, not the deck

List the artefacts due 48 hours before the review: updated business case, phase deliverables, test results, risk register, cost-to-complete, next-phase plan and budget. A pre-read makes a 60-minute decision possible; a 40-slide deck read aloud makes it impossible.

Step 5: Timebox the meeting and separate presenting from deciding

The project manager presents for fifteen minutes, reviewers ask questions for twenty, and the decision is taken with the project manager out of the room. People defend their work in front of their sponsor, and reviewers soften in front of the person whose bonus depends on the answer.

Step 6: Write the output the same day

Decision, conditions, owners, dates, next gate. Circulate it within 24 hours. An unwritten gate decision becomes a debate again three weeks later, usually with a different conclusion.

Where should the gates sit in a project?

Most projects need four to six gates, positioned at commitment points rather than at convenient calendar intervals. A generic project can use this map.

Gate 3 is the one most organisations get wrong. It is the last cheap exit, and it is usually scheduled in the same week as a contract signature, which makes a kill commercially awkward. Move it two weeks earlier and it regains its purpose.

Gate 5 is the one most organisations skip. A review six to eighteen months after launch is the only gate that compares promises to outcomes, and it is how an organisation learns whether its Gate 2 forecasts are honest.

What criteria belong in a gate?

Criteria should cover five dimensions: strategic fit, business case, delivery readiness, risk, and capacity. Capacity is the one teams forget, and it protects the rest of the portfolio, because a project can pass every other test and still be the wrong thing to fund this quarter if the people it needs are committed elsewhere.

Scoring works only if the weights are agreed before the review and applied across projects. Otherwise each project is judged on the dimension its sponsor cares about.

Worked example: a scorecard that says go, and a knockout that says stop

A customer portal rebuild reaches Gate 3. Reviewers score it out of five:

• Strategic fit: 4 × 25% = 1.00
• Business case: 3 × 25% = 0.75
• Delivery readiness: 4 × 20% = 0.80
• Risk: 3 × 20% = 0.60
• Capacity: 2 × 10% = 0.20

Total: 3.35 out of 5, or 67%. On score alone, that is a conditional go with capacity as the named gap. Then the knockout check finds the data migration plan has no owner and no signed data-protection approval. That is a must-meet failure, so the decision becomes recycle regardless of the 67%.

That is why the lists stay separate: a weighted score invites trading a weakness against a strength, and a knockout does not.

Worked example: what a late kill costs

An insurer screens 40 ideas at Gate 1. Twelve proceed to business cases at an average cost of $18,000 each. Five enter development, where average spend to Gate 4 is $640,000. Three launch.

The number that matters is the difference between kill points. Killing at Gate 1 costs about $2,000 in analysis; killing at Gate 4 costs roughly $640,000 plus a year of team time. Moving two kills a year from Gate 4 to Gate 2 saves more than a million dollars.

What does running a gate cost?

A 90-minute gate with seven senior participants at a loaded cost of about $95 an hour is roughly $1,000. Add 12 hours of team preparation at $70 an hour, and the gate costs around $1,840.

Compare that with the alternative. A programme burning $38,000 a week that continues six weeks past a gate it should have failed spends $228,000. The gate is under 1% of that, so even one genuine stop in twenty pays for the process many times over.

There is a quieter cost too: schedule. If four gates each slip three business days because evidence arrives late, the project loses twelve working days, about 2.4 weeks, or 6% of a 40-week plan.

How do you keep gates honest?

Three mechanisms do most of the work: a decision log, a kill rate you are willing to state publicly, and conditions with expiry dates.

The decision log. One row per gate: date, attendees, evidence version, decision, conditions, next gate. Review it quarterly. If every row says go, you have your answer about the process.

A visible kill rate. Portfolio-level, not project-level. In a healthy pipeline, a meaningful share of initiatives stop at early gates. If nothing is killed before launch, the gates are recording, not screening.

Conditions with expiry. If a condition is not closed by its date, the project reverts to “not approved for the next phase” until the chair confirms closure. Without expiry, conditions become permanent footnotes.

> Practical tip: ask the project manager to write the strongest case for killing their own project, one page, as part of the pre-read. It surfaces the risks the deck omits, and it makes the kill conversation possible without making it personal.

What are the common failure modes?

• Criteria written after the results. The gate becomes a justification exercise. Fix: criteria are appended to the previous gate’s decision record.
• No decision owner in the room. The meeting ends with “we’ll take it offline”, which means nobody decided.
• A gate with no kill option. If the project cannot stop, the organisation loses the ability to reallocate.
• Financial criteria only. Business cases for genuinely new work are often wrong by wide margins, so purely financial screens reject the most innovative projects and approve the safest-looking ones.
• Gates as presentations. A 40-slide deck read aloud consumes the hour and prevents the decision.
• Skipping the gate under pressure. The project most in need of a gate is the one most likely to argue it has no time for one.
• Never reviewing gate quality. Nobody checks whether Gate 2 forecasts matched Gate 5 actuals, so the process never learns.

When are phase gates the wrong tool?

Gates suit work with expensive, irreversible commitments and long feedback cycles: construction, hardware, regulated products, large IT programmes, anything with a contract award. They are weak in fast, reversible, discovery-heavy work, where the cost of a gate exceeds the cost of the experiment it reviews.

Three signals that gates are misfiring: the team optimises for the evidence pack rather than the outcome; every function adds its own review until preparation eats a fifth of delivery time; and decisions arrive after the commitment has already been made informally.

The alternative is not “no checkpoints”. It is lighter checkpoints at shorter intervals with delegated authority, plus a few real gates at commitments that cannot be reversed. Many organisations land on a hybrid: quarterly investment gates for funding, and a fortnightly delivery review with no funding authority but real dependency and risk decisions.

The trade-off is honest. Fewer gates mean faster decisions and more reliance on local judgment. More gates mean stronger central control and slower delivery. Choose based on how expensive a wrong commitment is, not on how much control you would like.

How does Doitify support phase gates?

Doitify holds the evidence a gate needs beside the work itself: milestones with dates and owners, tasks and sub-tasks with checklists for phase deliverables, definition-of-done and QC checks, a risk and constraint register with owners, WBS dependency links, workload views for the next phase, and reports that compare planned with actual.

That matters because the weakest part of most gate processes is not the meeting. It is assembling the evidence pack from five tools the week before. Doitify is our product, so read that with the usual caveat: if your gates already run well on documents and spreadsheets, a platform improves preparation time, not decision quality.

Conclusion

A phase gate is a small, cheap instrument with one job: make the continue-or-stop decision explicit while the money is still reversible. It works when criteria are written in advance, evidence arrives early, one person can say no, and the decision is recorded.

Design four to six gates at your irreversible commitments. Split criteria into knockouts and scores. Timebox the meeting and separate presenting from deciding. Then check, once a quarter, whether any gate has actually said no. If none has, you do not have a gate process. You have a ritual, and the project that should have stopped is still running.

If this post on phase gates was helpful, you might also enjoy How to Reduce Tool Sprawl in Your Team (2026 Guide) and Free Weekly Project Report Template (2026).