How many signals should typically clear the filter each week?

For a typical mid-market team with a few thousand ICP accounts, expect 20 to 60 high-conviction alerts per week after stacking and scoring. Significantly more than that usually means the threshold is too low. Significantly fewer means the threshold is too high or the signal pipeline is undersourced.

Should weak signals be ignored entirely?

Not entirely, but they should not drive outreach on their own. Use them as priors. A weak signal at an ICP account does not justify a sequence, but it can raise the priority of any future stronger signal at the same account.

What signals are most over-rated?

Single page visits, single intent topic flares, generic social engagement, and account growth alerts that fire on routine hiring. These have low base correlation with buying decisions and high false positive rates, especially when used as standalone triggers.

From Company Noise to Signal: Extracting What Actually Matters

DemandScience analyzed 5,903 companies and found only 62 were genuinely in-market. The job is no longer collecting signals, it is filtering 99 percent of what comes in so the rep only sees what is worth a sequence.

Why Noise Drowns Most Teams

Volume of observable activity overwhelms reps. Intent feeds prioritize recall over precision, training reps that alerts are not worth checking. Alert fatigue follows.

Filtering Criteria

Three layers: ICP fit, recency, and intent depth. Single touches on low-intent surfaces are noise. Multiple touches across high-intent surfaces in tight windows are signal.

Signal Stacking

Require two corroborating signals from different categories before firing an alert. This drops volume 80 to 90 percent and raises conversion.

Weighted Scoring

Depth times frequency times seniority times ICP fit produces a single score that decides which alerts the rep works first.

Common Failure Modes

Acting on every signal, no recency cutoff, and scoring without quarterly outcome review undermine the value of the signal stack.