A business is a flow system. Demand enters, decisions occur, value is delivered, and cash or durable customer value exits. The current constraint is the stage that prevents the system from producing more of the output that matters.
Acquisition, conversion, retention, and capacity are candidate locations. The data must show which one is limiting the next unit of valuable output.
Click a node to inspect the decision logic. The path is deliberately sequential: no intervention earns priority until the mechanism and economic consequence are explicit.
Map acquisition, conversion, retention, and capacity with the same period, scope, and definition of valuable output. Pick the lane with the strongest combination of economic impact, evidence, controllability, and learning speed.
flow map → comparable evidence → candidate constraintDo not use one metric to diagnose every lane. Each constraint produces a different pattern in the flow data.
Acquisition is suspect when the input stream is too small or too poorly matched to load the system with viable opportunities. Measure qualified opportunity rate, source quality, response latency, and fit rejection. Raw lead volume is an input signal, not the output.
Conversion is suspect when qualified work enters but stalls before a clear decision. Locate the exact stage where time accumulates or progress stops. Measure stage conversion, age distribution, next-step completion, and contribution per win.
Retention is suspect when acquired customers do not reach repeated value. Read cohort behavior, activation, usage, support, contraction, and cancellation timing together. A retention percentage without cohort scope can hide the leak.
Capacity is suspect when work accumulates at a constrained role, handoff, decision, or delivery step. Track work in process, throughput, cycle time, queue age, rework, and defects. Do not purchase capacity before using the existing constraint well.
A ranking table does not replace judgment. It makes the assumptions visible enough to challenge.
Rank with one time window. Use the same week, month, or cohort definition for every candidate. Score economic impact, evidence strength, controllability, and time to learn. The score is a forcing function; it is not permission to pretend a complex system can be reduced to one exact number.
Ask the counterfactual. If the lane improved materially, what output would rise next? If the answer is “nothing until another lane changes,” you have found a symptom or a non-constraint. Stop the project before it consumes scarce attention.
The protocol prevents premature hiring, spending, and optimization theater.
Draw the flow, locate the queue or drop-off, and write the exact step that limits valuable output. “Marketing is weak” is not a constraint statement. “Qualified opportunities from the target segment fall below the sales team’s usable capacity” is testable.
Remove interruptions, incomplete inputs, unnecessary approvals, unpriced customization, and low-value work. Use the constraint’s scarce time on the work with the highest contribution or learning value.
Limit work in process, sequence requests, protect handoffs, and pause work that competes with the constraint. A local team may appear less busy while the total system becomes faster and more predictable.
Add people, spend, tooling, or distribution when the old constraint remains after exploitation and subordination. Define the added throughput, cost, quality guardrail, and payback logic before the investment.
Measure the baseline and the downstream output. If the constraint moved, return to the flow map. If it did not, reject the story and reopen the diagnosis.
There can be several serious problems, and a system can contain linked constraints. The operating question is which one currently limits the next unit of valuable output. If two constraints are genuinely coupled, state the coupling and define the first intervention rather than creating four parallel priorities.
Start with the narrowest flow boundary you can measure. Separate common-cause deterioration from a single bottleneck. Check definitions, cohort mix, seasonality, data delays, and denominator changes before assigning the problem to a commercial lane.
Hire when the evidence shows a repeatable constraint remains after avoidable loss has been removed, adjacent work has been subordinated, and the expected added capacity has a defined economic and quality case. Hiring can be the elevation step; it should not be the identification step.
Long enough to observe the relevant leading signal and enough downstream output to avoid reacting to noise. The correct window depends on the cycle of the system. Define the review date and stop condition before launch rather than selecting a duration after seeing the result.
That is often the expected result. A moved constraint means the system changed. Record the new queue, drop-off, or economic leak and return to the first step. Do not keep optimizing the old constraint because its dashboard is familiar.
Bring one scoreboard, one flow map, one suspected constraint, one owner, and one bounded attack. The goal is not more analysis. It is a measurable system move.
Open Diagnostic Sequence