Derive the threshold from your data, never from a template

The fastest way to ruin a sound model is to import someone else's cut-off. An MQL threshold copied from a vendor blog is a guess about a stranger's funnel, their ICP, their channel mix, their buyers, none of which is yours. The point values you chose are a hypothesis; the threshold is where that hypothesis meets reality, and reality is in your own data, not in a template.

Let the leads accumulate first

Resist the urge to draw a line on launch day. Run the model for around sixty days and let real leads flow through it scored but unfiltered, so you build a population with both a score and an outcome. You cannot find where the model separates winners from time-wasters until you have enough scored leads with known results to see the separation. Sixty days is the patience the model demands before it can tell you anything trustworthy about where to cut.

Read the curve, do not guess it

Now do the work the template skips. Bucket your scored leads into bands, then chart two things per band: the sales-acceptance rate and the opportunity-conversion rate. A pattern emerges. Some bands convert at a real multiple of your baseline, and some are the bands your follow-up effort quietly ignores because the leads in them never go anywhere. The threshold is not a round number you like, it is the point on that curve where acceptance breaks, the boundary between the band that earns attention and the band that does not. Set the cut-off there, derived from the break in your own data, and it will mean something.

Use the reference points as a compass, not a map

There are sane starting references, and they are starting references only. An MQL often lands near sixty-five points, roughly the top fifteen to twenty per cent of a database, and an SQL near eighty-five. Use these to sanity-check that your derived threshold is in a plausible neighbourhood, not to replace the derivation. If your curve says the break is at fifty-eight, trust your curve over the blog. The reference tells you roughly where to look; your data tells you exactly where to cut.

Know what good operating numbers look like

Once the threshold is live, you have benchmarks to hold it against. A healthy B2B model runs MQL-to-SQL acceptance above sixty per cent, SQL-to-opportunity conversion above thirty per cent, and time from MQL to first sales contact under four hours. If acceptance is sitting well below sixty per cent, your MQL threshold is too generous and junk is leaking through; if your best reps are starved, it may be too strict. The threshold is not a one-time decision, it is a dial you tune against these operating targets as the curve underneath it moves.