Diagnostics
How to Find the Bottleneck in a B2B Pipeline
A repeatable diagnostic for locating the real constraint in a B2B pipeline using volume, conversion and latency measured separately at every stage.
- Author
- Felipe SozaFounder, GTM Systems Architect
- Published
- Updated
- Last reviewed
Executive summary
- Measure volume, conversion and latency separately. Conflating them hides the constraint.
- Work upstream to downstream and stop at the first stage that fails its threshold.
- Fix one constraint at a time, then re measure, or you will not know what worked.
Key thesis
There is exactly one binding constraint at a time. Find it with arithmetic, fix it, then find the next one. Everything else is decoration.
Step one: lay out the stages honestly
Write the actual stages a prospect passes through, including the ones nobody owns. A typical outbound path: sourced, valid, eligible after suppression, sent, delivered, replied, positive reply, meeting booked, meeting held, qualified opportunity, closed.
Step two: measure three numbers per stage
- Volume: how many entered the stage in the period.
- Conversion: what fraction moved to the next stage.
- Latency: median time spent in the stage.
Latency is the number teams skip, and it is frequently where the constraint hides. A 30 percent reply to meeting rate with a two day response delay is not a messaging problem.
Step three: apply thresholds, upstream first
- 01Data valid rate below 90 percent means stop here. Everything downstream is polluted.
- 02Delivery or placement degradation means stop here. You are measuring an infrastructure fault, not market response.
- 03Positive reply rate collapsing while total reply rate holds means a targeting or offer problem.
- 04Meeting held rate below 70 percent of booked means a qualification or confirmation process problem.
- 05Opportunity to close falling with healthy earlier stages means a sales execution or fit problem, not a pipeline problem.
Stop at the first failing threshold. Downstream numbers computed on polluted upstream data will send you in the wrong direction with confidence.
Step four: cohort before you conclude
Split by segment, by sending unit and by list build date. A single decaying cohort can drag a blended average enough to look like a general decline. If one cohort is fine, the system is fine and the input is not.
Common failure modes
- Changing three things at once, then attributing the result to the favorite one.
- Using blended averages across cohorts with different ages.
- Ignoring latency because no tool reports it by default.
- Diagnosing from a dashboard instead of a raw export.
- Fixing a non binding constraint and being surprised that output did not move.
What to do with the answer
Write the constraint down as a sentence with a number in it. Assign one owner and one date. Re measure the same three numbers after the change. If output did not move, your diagnosis was wrong and that is useful information, provided you only changed one thing.