Project Funnel Metrics: Why Drop-Off Belongs Beside Cycle Time and Throughput15 min read

A project funnel can look busy while nothing inside it actually gets decided. Work moves between stages, dashboards fill up, and approved projects quietly age.

Three numbers tell you what’s really going on:

• Cycle time, or how long a stage takes to clear.
• Throughput, meaning how many projects formally leave a stage each period.
• Drop-off: how often a gate ends in a stop somebody actually wrote down.

One boundary first, because it decides everything downstream. A project funnel starts after an idea is approved, and ends when the delivered outcome is closed.

Early screening and funded delivery carry different risks, different evidence, and different decision cadences. Idea intake belongs in a funnel of its own.

Cycle time and throughput get dashboards built for them. Drop-off gets treated as a leak, which is where the trouble starts.

Why These Three and Not Twelve

Innovation programs collect plenty of metrics. The ones that describe activity can be reported in full without anyone changing a decision.

Deloitte’s 2021 study found that organizations with a self-described leading innovation capability are likelier to set targets and actively measure idea conversion, 50% versus 39% of average respondents.

The gap runs upward too. BCG’s 2024 survey found just 12% of companies report a strong link between business strategy and innovation strategy.

Definitions drift between departments faster than data does. Our innovation dashboard glossary is the reference to settle terms before anyone argues about a number.

What Each Metric Actually Measures

The Kanban Guide defines throughput as the number of work items finished per unit of time, and cycle time as elapsed time from started to finished.

Those definitions carry into a governed funnel with one adjustment. Here, the unit of completion is a gate decision, and one of the available decisions is to stop.

That adjustment is what makes drop-off a peer of the other two. A recorded stop is an output of the funnel, with a reason and a cost attached.

Start With Event Data, Not Status Labels

You can’t diagnose flow from a current status column. Diagnosis needs an event history recording when every transition and decision actually happened.

Store these as events, never as editable summary fields. If a recycled project re-enters concept development, open a new stage instance rather than overwriting the first.

Keep one calendar rule. Wall-clock days capture the delay stakeholders feel, working days suit internal capacity work, and mixing the two ruins comparability.

The seven-stage project funnel gives you a workable vocabulary here. Your stages can differ, as long as their boundaries stay explicit and stable.

An exit also needs criteria someone can apply under pressure. Our stage-gate execution checklist sets out entrance criteria, success criteria, and decision outcomes per gate.

Define the Three Metrics Precisely

Each metric below gets a formula, a counting rule, and the field it reads from. Precision here is what stops two teams reading the same funnel differently.

Cycle Time

Stage cycle time is the elapsed time between a project entering a stage and formally exiting it.

Stage cycle time = stage_exit_at - stage_entry_at

ProKanban warns that averages are highly susceptible to outliers, so one slow project moves the number. Report the median and 85th percentile, not the mean alone.

Track unfinished work separately, as current stage age: today minus stage_entry_at. Completed work tells you what the funnel did, and aging work tells you what it’s about to do.

Cycle time has its own blind spot. A stage can post a healthy median while its slowest projects define what the business actually experiences.

Throughput

Stage throughput is the count of formal stage exits recorded in a fixed reporting period.

Stage throughput = count of stage_exit_at events in the period

A formal exit covers both advancement to the next stage and termination from the funnel. Both are governance outputs, though only one moves delivery forward.

So split the count in two: advancement throughput and closure throughput. A decisive quarter and a productive one shouldn’t read the same.

That split is also where drop-off enters the throughput number, which is the first place these three metrics stop being independent.

The arrival side is the easiest to miss. A stage clearing ten projects a month looks stable until you notice only four are arriving.

Drop-Off

Gate drop-off is the share of gate decisions resolving continuation, Go or Kill, that formally terminate a project.

Gate drop-off = Kill / (Go + Kill) x 100

Hold and Recycle stay out of the denominator, because neither has resolved continuation. Publish their counts next to the rate so deferred decisions can’t hide inside it.

No decision is different again. It means a project reached the gate with no gate_decision recorded, so it’s an absence in the log.

Drop-Off Is a Peer, Not a Byproduct

Drop-off is the number teams instinctively want to see fall, and the one worth watching closely. A gate that never stops anything isn’t a gate.

The base rate is sobering. Cooper’s analysis of pipeline gridlock cites a PDMA estimate: one in four projects entering heavy-spending development becomes a commercial success.

The Kill Decision Most Gates Never Make

Robert Cooper’s diagnosis is direct. Gates are often non-existent or lack teeth, and the Kill option is rarely exercised because management doesn’t know how to say no.

Cooper adds that after the initial Go, gates amount to little more than project reviews or milestone checkpoints, not serious Go/Kill investment decisions.

McKinsey reports the same pattern, noting that funding is rarely delayed or halted at a traditional stage gate.

Cooper names the other half of the problem. A hollow Go, approval without committed resources, leaves no limit on people availability, so too many projects get approved.

What Late Attrition Costs

Delay doesn’t make a stop cheaper. Research on IS projects by Keil, Mann, and Rai suggests that between 30% and 40% exhibit some degree of escalation.

In Staw’s 1976 experiment, business students responsible for an earlier funding choice allocated $11.08 million in a simulated second round, against $8.89 million from others.

Keil and Montealegre put it plainly: managers frequently continue pouring in more resources. Their California case grew from $75.5 million to an estimated $260 million.

Attrition Is Normal, and Smaller Than Folklore Says

None of this means a healthy funnel kills most of what enters it.

The claim that 80% of new products fail was challenged in the Journal of Product Innovation Management. Empirical studies since 1977 put the rate at 40% or less.

Three different things get called failure:

• Product failure after launch, which is a market outcome.
• Project cancellation at a gate, which is a governance decision with a reason attached.
• Silent stalling between gates, which the log records only as an absence.

The large-sample failure datasets track products already in market, not projects moving inside a portfolio.

One study of 83,719 new packaged-goods SKUs found a quarter were no longer bought a year after launch, rising to roughly 40% after two.

So treat drop-off as a rate you interpret, not a rate you inherit. What matters is whether each stop was evidenced, timely, and recorded.

Read the Three as One Signal

Alone, each metric supports a comfortable story. Together they show whether the funnel is selecting, flowing, or just recording motion.

That last one has a name.

A 2024 Technovation paper on innovation theater describes symbolic activities that are highly visible and have no impact on the innovation process.

Little’s Law offers a consistency check: WIP = throughput x cycle time.

That relationship holds only under strict conditions. Little’s 1961 paper requires finite means, strictly stationary processes, and an arrival process that is metrically transitive with nonzero mean.

A funnel that’s still filling, shrinking, or losing projects to unrecorded limbo isn’t stationary. Use the relationship as a smell test, then go back to the stage data.

Utilization is the other force acting on all three numbers.

Queues grow nonlinearly well before a system looks full. Estimates of the tipping point vary, from around 50% to near 80% utilization, so watch your own queue times.

A Worked Example You Can Copy

The worksheet below follows the 48 projects that entered in one quarter, tracked to a single reporting cutoff rather than to full completion.

Days are wall-clock, and percentiles cover completed stage instances only. The No Decision column counts stage instances with no gate decision at cutoff.

Drop-Off, No Decision, and the spread between median and 85th percentile days carry the argument here.

What the Numbers Say

Feasibility drop-off is 6 / (29 + 6) x 100 = 17.1%. Its two Holds and two missing decisions stay visible beside it without diluting the rate.

The largest drop-off sits at validation and pilot, at 19.0%. That’s a good place for it, assuming pilot evidence disproved something and each stop carries a written reason.

What to Investigate

Implementation has the lowest drop-off and the widest absolute spread, 96 median days against 154 at the 85th percentile. Read the projects above 154 first.

One unresolved implementation project is probably a reporting cutoff. The same gap three quarters running is silent stalling, and it needs a name and a date.

Only the first four stages here clear 20 formal exits, so read the later percentiles as provisional. That’s the normal state of a first baseline.

Build Your Own Baseline

A borrowed benchmark carries the source’s project mix, gate depth, and review calendar with it. Start with your own history, then compare like with like.

Published ranges, including the ones in our glossary, are orientation and not targets. They tell you whether you’re unusual, not whether you’re wrong.

Old funnel ratios are especially tempting here.

Stevens and Burley’s 1997 paper popularized a 3,000-ideas-to-one-success chain, but it’s paywalled, its methodology isn’t publicly checkable, and it counts raw ideas rather than funded projects.

How Much History You Need

Sample size decides whether a percentile means anything:

• Higher-volume stages: eight to twelve weeks, covering at least one full decision cadence.
• Lower-volume stages: two quarters, or a rolling twelve months.
• Any stage: roughly 20 to 30 formal exits before you quote a percentile.

Treat those as operating rules, not thresholds. A stage with six exits doesn’t have a trustworthy 85th percentile, however precisely the dashboard prints one.

Label early numbers provisional in the report itself. It costs one word and prevents a working figure from hardening into a target.

Segment Before You Compare

Separate cohorts whose evidence requirements genuinely differ. Project type, investment band, regulatory exposure, and strategic horizon are the usual dividing lines.

A lightweight process improvement shouldn’t set the expected cycle time for a new platform. Blend them and you get a baseline describing neither.

Freeze the Definitions

Fix your definitions for the length of the baseline period. When a gate boundary moves, mark the date and start a new series instead of blending two designs.

Recalculate monthly, and resist rebaselining after one unusual quarter. A rolling view with annotated changes in staffing, policy, or mix will age better.

Diagnose the Movement, Then Act

A metric should trigger a question, an evidence check, and an owner. Automatic target-setting can turn a diagnostic into theater.

Notice that most actions in that last column change the system, rather than setting a target on a number.

Work down the row that matches your signal, then stop. Fixing two things at once makes the next reading harder to interpret.

Protect the Incentives

The moment drop-off becomes a target, people manage the number instead of the decision behind it.

Marilyn Strathern’s 1997 formulation still says it best: when a measure becomes a target, it ceases to be a good measure.

Reward managers for low drop-off, and weak projects stay alive. Reward raw throughput, and Kills get counted as wins, distorting in the opposite direction.

Decision time is scarce enough without that. In McKinsey’s global survey, 61% of respondents said most of their decision-making time was used ineffectively.

Keep the Decision Record

When a project stops, the record is what survives it. Our zombie-project diagnosis covers what projects cost when they’re never formally stopped.

Keep four things on every closure:

• The reason, as a standard code plus one sentence.
• The evidence that made the decision defensible at the time.
• The cost to date.
• The learning somebody else can reuse.

This is where software earns its place. Automated timestamps, controlled statuses, and gate records reduce the manual reconstruction that turns portfolio reviews into memory exercises.

Clustering closure reasons and flagging projects past their expected stage age are worth automating. Neither makes the Kill decision itself, which still belongs to the accountable gate owner.

The harder constraint is cultural.

In NewVantage Partners’ annual survey, reported by MIT Sloan Management Review, 80% of respondents said the main barriers to becoming data-driven were human rather than technological.

Count the Stops as Carefully as the Starts

Cycle time, throughput, and drop-off are peers, not runners-up. Read them together and you can see whether the funnel selects, flows, or only performs.

Put them to work in four moves:

• Find the stage where the signal actually moved.
• Separate what was formally recorded as attrition from work that simply stalled.
• Test your explanation against the event records, not the status column.
• Assign one action, one owner, and one review date.

A healthy funnel won’t optimize any single one of these numbers. It records stops as readily as starts, and moves resources toward projects that still deserve them.

Accept Mission customers who standardize their funnels report implementation running 25 to 30 percent faster. Standardized gates are what make the stops countable.

Download our free ebook Project Portfolio: From Opportunities to Value to learn how to structure stage-gate governance, prioritize work, and track portfolio health over time.

Request a demo to see how Accept Mission gives you stage-level timestamps, recorded gate decisions, and live portfolio views across every project in your funnel.