What does a high score mean?

The score is directional only. The panel narrows risk before spend.

AI Marketing Panel Workflow Kit | rTech Digital Consulting

Field sheet

Find the objection before the market finds it for you.

Most feedback arrives after the spend: the campaign underperforms, the page does not convert, the offer lands wrong. AI Marketing Panel moves the objection earlier. Each seat reads the same artifact independently, so reactions are not contaminated by consensus, and a separate reviewer reports where they agree, where they split, and who would actually act.

The core of this kit is open source under the MIT license: AI Marketing Panel on GitHub. What rTech adds on top is the human review checkpoints that decide what ships.

Worked exampleSample data only

Scenario: A fictional regional services firm is about to launch a new offer page. Names, seats, and reactions are invented for demonstration.

Panel builtSeven seats: three plausible buyers, two wrong-fit, one hard-no, one prior customer who churned. Each with documented needs, budget authority, and history.

Artifact washedEvery seat reads the same page cold, without seeing the others' reactions. If the tool being used cannot keep the seats apart, the run is labeled degraded rather than presented as independent.

Split surfacedThe buyers accept the promise. The churned customer names an unaddressed failure from last time. Neither view is averaged away.

Rewrite and retestThree rewrites address the specific objection, then face the same panel again.

Input

What it needs

The decision being made, your own customer evidence, and the artifact under evaluation: a page, offer, email, script, or positioning statement. Setup is not finished on sample data: the first panel is built from your customers and run against your artifact, and the kit will not build a panel until you supply both.

Output

What it produces

One evidence-labeled dossier per seat, independent reactions, a synthesis that preserves dissent, and a weighted score against owner-approved criteria, with the arithmetic and the cited quotes checked mechanically before the run is accepted. You read the split and decide what ships; the kit never releases anything.

Stop condition

When it refuses to run

It stops and names the gap: evidence too thin for honest composites, a roster with nobody who can say no, a panel outside five to fifteen seats, scoring weights that do not total 100, or a request to invent customer evidence. It also stops after three rewrite cycles that have not moved the objection, and recommends real customer interviews instead.

Processing boundary

Local by default. External by decision.

Panel definitions, dossiers, and run records are ordinary files. The operator chooses what is appropriate to send to an external model.

Runs locally

Panel definition and seat roster
Customer evidence and dossiers
Run records and prior reactions
Scoring weights and criteria
Prediction and outcome ledger

May run externally

Seat reactions from approved, non-sensitive context
Synthesis across recorded reactions
Alternative rewrites for review

Never automatic: sending confidential customer records, publishing an artifact, buying media, or treating a panel score as a measured result.

How to read the output

Disagreement is the signal.

A hard-no saying no is healthyIt confirms the panel has real range instead of a room built to agree.
A wrong-fit seat saying yes is a warningThe artifact may be attracting the customer you do not want.
An all-positive panel is an integrity failureUnanimous approval means the composites were built too agreeable to be useful.
A high score is directional onlyThe panel narrows risk before spend. Every run records its prediction so the real outcome can be added later and correct the panel that made it.

Human review checklist

Before the artifact ships.

Is each seat grounded in real customer evidence, not invention?
Does the panel include seats that can say no?
Did every seat react independently, without seeing the others?
Does the synthesis preserve dissent instead of averaging it?
Are the scoring weights ones the owner actually approved?
Is the prediction recorded so the real outcome can correct it later?

Discuss this workflow Back to all kits