Scale, Hold, Cut, Pause: A Directive Framework for Marketing Capital Allocation

Marketing dashboards show data. They do not tell you what to do. Allocera's CDAI engine produces exactly one of five enforceable directives — Scale, Hold, Cut, Pause, or Flag — for every active campaign, each with a confidence score, reason codes, and explicit budget guidance. Here is the framework, the thresholds, and the 80 percent measured accuracy result.

Every marketing analytics platform produces some version of the same artifact: a dashboard. The dashboard surfaces data — conversion counts, cost per lead, ROAS, attribution credit by channel. Interpretation is left to the marketing team. The team looks at the numbers, has a discussion, and makes capital allocation decisions on what to scale, what to cut, what to keep running.

That model works fine when the marketing team has the analytical sophistication and time to make sense of the data, and when the data is reliable enough to act on. For large enterprise operations with dedicated analytics teams, that's the correct division of labor. For mid-market operations spending six figures monthly without an in-house analytics function, it's a gap. Showing the dashboard is not the same as making the call.

Allocera's CDAI engine was built around a different model. Instead of reporting data and leaving interpretation to the team, the engine reconciles all seven cost layers to true contribution margin per campaign and then issues exactly one directive per campaign per evaluation cycle. The directive is not a suggestion. It is a specific capital action — Scale to 1.5x budget, Cut by 50 percent, Hold, Pause immediately — with the underlying margin math attached and a confidence score in the 70 to 97 percent range.

This piece walks through the five directives, the contribution margin thresholds that trigger each one, what the confidence scoring means, and why the 30-day retest validates the framework against actual outcomes. We covered the calculation methodology behind contribution margin in our complete guide, and the retest mechanics in our 30-day retest analysis.

The Five Directives

How Each Directive Works

SCALE

The engine issues SCALE when contribution margin is at or above 30 percent, no risk signals are firing (refund rate within range, no fraud elevation, quality scores stable), and the campaign has been operating long enough to produce statistically meaningful data. Confidence score: 88 percent. Recommended action: increase budget to 1.5x current spend.

Why 30 percent as the threshold: at margins below 30 percent, the variability in week-to-week performance is high enough that scaling can compress margin into single digits unexpectedly. At margins above 30 percent, the campaign has enough cushion to absorb the natural quality drift that comes with increased volume.

HOLD

The engine issues HOLD when contribution margin runs in the 18 to 30 percent range with stable signals. This is the engine's "more data needed before directional move" call. Confidence score: 70 percent — the lowest of the directive set, deliberately so. HOLD is not a high-conviction call. It's a flag that the campaign is in a middle band where small shifts in cost or revenue could move it in either direction, and the engine is waiting for more data before issuing a SCALE or CUT.

CUT

The engine issues CUT under three conditions: contribution margin below 10 percent, campaign spending with no revenue (typical of attribution-broken campaigns or campaigns burning budget into the void), or refund rate exceeding 15 percent. Confidence score: 85 to 90 percent. Recommended action: reduce spend by 50 percent or eliminate the campaign entirely depending on severity.

Every directive carries a confidence score, reason codes, and a specific capital action. Then 30 days later, the engine grades itself against the actual outcome.

PAUSE

The engine issues PAUSE on emergency conditions: fraud rate above 20 percent of leads, contribution margin worse than negative 50 percent, or chargeback rate above 15 percent. Confidence score: 93 to 97 percent — the highest of any directive type, because PAUSE only triggers on unambiguous severe distress signals. Recommended action: emergency stop the campaign immediately.

PAUSE is the lowest-frequency directive in normal operations. When it issues, the underlying signal is severe enough that continued spend produces catastrophic margin damage every additional day.

FLAG

The engine issues FLAG when CPL distortion exceeds 50 percent between true reconciled cost and dashboard-reported cost, or when the engine detects a data anomaly the underlying directive logic cannot resolve automatically. Confidence score: 85 to 90 percent. Recommended action: human review required, data integrity issue likely. FLAG is the engine's way of saying "the math says this campaign is severely off, but the data quality is questionable enough that I want a human to look at it before I issue Cut or Pause."

Confidence Scoring

Every directive carries a numeric confidence score in the 70 to 97 percent range. The score reflects the engine's certainty about the directive based on data volume, signal stability, and the strength of the underlying contribution margin signal. A SCALE issued on a campaign with 90 days of stable data at 35 percent contribution margin lands near the top of the SCALE confidence range. A SCALE issued on a campaign with 30 days of data at 31 percent margin lands near the bottom.

Each directive also includes reason codes — explicit labels for the metric values that triggered the directive. A SCALE might carry reason codes like high_margin_stable, quality_signal_strong, no_fraud_flags. A CUT might carry margin_negative, spend_no_revenue, refund_rate_high. The reason codes make every directive auditable: any technically literate reviewer can trace the directive back to the specific contribution margin and quality signal values that triggered it.

What Validates the Framework: The 30-Day Retest

The directive framework above produces specific predictions. SCALE predicts that margin will hold or increase after the budget increase. CUT predicts that margin will improve or losses will stop after the cut. HOLD predicts margin will stay within plus or minus 5 percent of the baseline. Each of these predictions is testable against what actually happens 30 days later.

That's exactly what Allocera's engine does. Every directive issued is recorded with its pre-directive contribution margin state. Thirty days later, the engine automatically retests against the post-directive contribution margin and classifies the outcome as CORRECT, INCORRECT, or INCONCLUSIVE based on deterministic rules per directive type.

As of May 2026, the engine has issued 56 directives, scored 55 of them (one within the 30-day window), and produced 80 percent measured accuracy — 44 correct, 11 incorrect. This is substantially above the 60 to 70 percent directional accuracy typical of multi-touch attribution and marketing mix models, and it is a measured number, not a marketing claim. We covered the retest methodology in detail in our 30-day retest deep dive and the broader methodology breakdown in our analysis of the 55 directives scored at 80 percent.

What Other Marketing Tools Do Not Do

Three things separate Allocera's directive framework from what the major attribution and CRM platforms produce.

• Directives, not reports. HubSpot, Salesforce, Triple Whale, Rockerbox, and the major attribution platforms produce reports. They do not issue specific capital actions per campaign. The recommendation layer is left to the marketing team to construct from the data. We covered the architectural difference with Salesforce specifically in our Allocera vs Salesforce Marketing Cloud comparison, and the attribution-tool comparison in our Triple Whale vs Rockerbox vs Allocera analysis.
• Confidence scores tied to retest outcomes. The 70 to 97 percent confidence scores in Allocera's framework are calibrated against measured 30-day outcomes, not modeled. Most attribution platforms either don't issue confidence scores or issue them as model fit metrics that don't translate to whether their recommendations produced the predicted result.
• The engine refuses to issue directives on bad data. If incoming data is stale or campaign attribution is missing, the engine's health monitor flags the data set as not directive-safe and no directives issue. This behavior is independently verified across two real businesses in our published validation case study. A wrong directive is worse than no directive.

The Question This Framework Answers

The framework exists to answer one operational question: for every active campaign right now, what should I do this week — and what should I do based on?

A dashboard answers neither. It surfaces data, and the interpretation work is the marketing team's problem. The directive framework answers both: the action (Scale to 1.5x, Cut by 50 percent, Hold, Pause, Flag), the basis (the reason codes), and the confidence (the 70 to 97 percent score). Then 30 days later, the engine grades itself and updates the confidence calibration based on what actually happened. We covered the underlying contribution margin calculation in our methodology guide and the foundational dashboard-vs-reality framing in our True CAC analysis.