One Number for "Is Our Marketing Grounded?" — The Signal Quality Score, Explained

Last updated 2026-08-20

The Signal Quality Score is a deterministic 0–100 grade, computed per project, that tells you how well-grounded a client's marketing evidence base actually is. It composes four weighted ingredients — coverage, depth, first-party voice, and freshness — into one number, plus a named list of the gaps behind it. No LLM is involved: same inputs, same score, every time.

Here is what sits under that number:

• Coverage — 0.35. Do you have signal across the required source types at all?
• Depth — 0.25. Is there enough of it per type to say anything with confidence?
• First-party voice — 0.25. How much of it is your own customers and prospects talking, versus secondary material?
• Freshness — 0.15. How recent is it?

It composes machinery that already existed in the platform, and it unit-tests without a database — which is a boring engineering detail that matters to you for one reason: nothing about your grade is a model's opinion of your work.

Why you need a number here at all

You are running three to five engagements. Every re-entry means reloading personas, pipeline state, campaigns, and politics. Our own interviews put the hidden cost of that context-switching around [~$75,000/yr for a four-client solo practice] in unbillable re-entry time.

One prospect said it plainly: "with experience, it's hard for us to stay on top of everything. There's no way." And what they wanted was not a dashboard. It was this: "if you go that far, then it's the only thing I need. Like I, I can just focus on lead gen."

That is the job the score does. It answers "is the strategy under this client's content actually standing on evidence, or on my memory of a call in March?" in one glance — so you can stop auditing and go sell.

Read the score in 30 seconds

The grade is not the point. The named gaps are.

A score never comes back as "62." It comes back as 62 plus a list of actions in your language: customer interviews — missing. win/loss — 1 source, need 3. competitor scan — 14 months old. You do not interpret a number. You work a punch list.

Rough bands to calibrate against:

• 80–100 — defensible. You can put positioning in front of a CEO and cite where it came from.
• 60–79 — usable, thin somewhere. Usually first-party voice or freshness. Ship, but name the gap out loud in the engagement.
• 40–59 — inherited assumptions. The strategy is probably the founder's original story, unchallenged.
• Below 40 — you are writing from vibes. Everything downstream, every AI output included, inherits that.

Raise it in the right order — highest weight first

Do not optimize what is cheap. Optimize what is weighted. The four ingredients carry fixed shares of the 100 points — 35, 25, 25 and 15 — so work them in that order.

1. Fix coverage first (0.35). Missing source types cost the most per unit of effort — a third of the total grade sits here. If customer interviews — missing, one 30-minute call moves the score more than five new competitor notes ever will.
2. Then depth (0.25). Two interviews is an anecdote; the score treats it that way. Add sources within a type you already have until each stops being a single data point.
3. Then first-party voice (0.25). Swap secondary material for primary. A Gartner-style summary and a recorded prospect call are not equivalent inputs, and the score refuses to pretend they are. This is the ingredient most practices are quietly failing.
4. Freshness last (0.15). Lowest weight for a reason — re-running a stale competitor scan is real work with the smallest return. Do it on a cadence, not in a panic.

A practical cadence that keeps a project above 80 without heroics: one customer or prospect conversation per client per month, a competitor and market re-scan quarterly, and a coverage check at every engagement kickoff and renewal.

Why deterministic matters more than it sounds

If an LLM graded your evidence base, the grade would drift. You would re-run it, get 71 instead of 66, and learn nothing except that the model was in a different mood. A fixed-weight composition of four inputs — 0.35, 0.25, 0.25, 0.15 — returns the identical number on the identical evidence, every run.

That does three things a model cannot:

• It moves only when your evidence moves. A rising score is proof of work, not sampling noise.
• It is arguable. You can point at the weights in front of a client and defend why the interview mattered more than the blog post.
• It is a control, not a compliment. The gaps are named actions, not a number to admire.

What this is actually for

Your clients are increasingly running your deliverables through AI themselves. The question behind their question is not "could a model have written this?" — it is "what do you know that the model doesn't?"

The answer is your evidence base. First-party customer language, current, deep enough to argue from, locked into one strategy every channel and every AI output reads from — the difference between a 39 and an 85 on the same client, and the difference between defending a retainer and losing it.

The Signal Quality Score is how you prove that base exists — to the client, and to yourself, before you build a quarter of GTM on top of it.

Check the score on your weakest engagement first. That is the one where you already suspect the answer.