SOP: MMM Analysis Service Delivery

Standard Operating Procedure for Agencies Using MMM Pilot

This SOP covers the end-to-end delivery workflow for an MMM Analysis engagement — from lead qualification to post-delivery follow-up. It is designed for internal agency use and assumes that MMM Pilot is the primary analysis platform.

The guiding principle of this workflow: MMM Pilot automates the statistical heavy lifting. Your effort should be invested where it creates the most value — in data quality, strategic interpretation, and client communication.

Roles and Responsibilities

Role	Responsibilities
Project Lead	Client-facing point of contact; scope management; final sign-off on deliverables
Analyst	Data collection, preparation, model execution, QA, and documentation
Account Manager	Scheduling, billing, project tracking, NPS collection

Phase 0 — Pre-Sales and Qualification

Before any project begins, confirm the client is a good fit for MMM.

Steps:

Run a discovery call. Understand business goals, active marketing channels, and how they currently measure performance.
Check minimum eligibility:
- ≥ 12 months of historical data?
- Spend across two or more active channels?
- A defined revenue or conversion metric?
Deliver a short MMM explainer — set expectations clearly. MMM is a strategic planning tool, not a real-time attribution system.
Scope the project: number of channels, analysis horizon, desired outputs.
Send proposal and commercial offer.
Collect signed contract and NDA.

Disqualification criteria: Less than 12 months of sales data. Only one marketing channel. Long B2B sales cycles (> 3 months). No revenue or conversion data available.

Typical duration: 1–2 weeks from initial contact to signed contract.

Phase 1 — Onboarding and Data Collection

The largest single risk to timeline is slow or incomplete data from the client. The goal of this phase is to reduce friction to zero.

Step 1.1 — Kickoff Meeting (45–60 minutes)

Introduce the agency team and confirm roles on both sides.
Confirm agreed scope, timeline, and key deliverables.
Walk through the Data Intake Pack — templates, instructions, and deadlines.
Assign a single data owner on the client side.
Set up the communication channel (email, Slack, or project management tool).

Step 1.2 — Data Collection

GA4 Data: Use MMM Pilot’s native Google Analytics integration to import conversion and revenue events directly — this eliminates manual exports and reduces formatting errors.
Ad Spend: Pull channel spend data (Meta, Google, and others) from agency accounts or ask the client for platform exports.
Internal Data: Request the promotional calendar, pricing history, and any seasonal flags in CSV format.
All data should be aggregated to weekly granularity before upload. Weekly is preferred over daily for SME-level transaction volumes.

Step 1.3 — Data Quality Audit

Upload the data to MMM Pilot and run a Quick Data Check (1–2 credits). This confirms that date formats are read correctly, column names have no conflicts, and the model structure is valid before committing to a full run.
If gaps exceed 20% of required data points, notify the client immediately and request clarification.
Issue a brief Data Audit Report (1–2 pages) summarizing completeness, any gaps, and how they will be handled.
Project Lead sign-off required before the analysis phase begins.

Typical duration: 1–2 weeks (depends on client responsiveness).

Phase 2 — Analysis and Modeling

This is the phase where MMM Pilot does the work. Your job is to direct it and evaluate its output.

Step 2.1 — Enrich the Model Context

Use the “Discover Economic Indicators” feature in MMM Pilot to identify 1–2 macroeconomic variables (e.g., Consumer Confidence Index, Inflation Rate) that correlate strongly with the client’s revenue trend.
Import selected indicators as an additional data source. This improves model accuracy and adds a layer of real-world context that strengthens the final presentation.

Step 2.2 — Standard Analysis Run

Use the AI Suggest feature to generate an initial hyperparameter configuration. This eliminates the need for manual statistical expertise and produces a well-calibrated starting point.
Launch the Standard Analysis run (~8–12 credits, approximately 2 hours of processing time).
The run executes in the cloud. Use this time for other project work — no manual oversight is required during processing.

Step 2.3 — Review AI Evaluation and Iterate

Once the run completes, open the AI Evaluation — an automatically generated report that summarizes model quality, flags any statistical concerns, and recommends improvements.
If the AI Evaluation suggests hyperparameter adjustments, use the “Duplicate with Suggested Hyperparameters” button to generate an optimized follow-up run with pre-filled settings. This eliminates the need to reconfigure from scratch.
Confirm model validity before proceeding:
- R² ≥ 0.75
- Sanity check: do the results align with business intuition?
- Peer review: a second team member reviews the output before it goes to reporting.

Typical duration: 1–2 weeks (varies with number of channels and data volume).

Phase 3 — Reporting and Deliverable Preparation

Turn model outputs into client-ready strategy. This is where your analytical work becomes business value.

Step 3.1 — Export and Interpret

Export the Contribution Breakdown charts, Spend Efficiency curves, and CSV data from MMM Pilot.
For each channel, write one clear, plain-language insight. Example: “Meta Ads is hitting diminishing returns above €2,000/month. Every euro beyond that generates a fraction of what it did at lower spend levels.”

Step 3.2 — Executive Summary

Two to four pages. Non-technical language throughout.
Structure: Top 3 findings → What they mean for the business → Recommended next steps.
Review and sign-off by the Project Lead before delivery.

Step 3.3 — Budget Optimization Scenarios

Present at least two to three alternatives (e.g., Conservative / Current / Optimized Growth).
Each scenario should show: channel-by-channel allocations, projected incremental revenue impact, and clear assumptions.

Typical duration: 3–5 business days.

Phase 4 — Client Presentation and Handover

The goal is understanding and buy-in — not a data dump.

Step 4.1 — Pre-Send

Send the Executive Summary and presentation deck 24–48 hours before the session.
Brief the client: “Read through the summary, bring your questions.”

Step 4.2 — Strategy Session (60–90 minutes)

Walk through: context → findings → channel-by-channel insights → Budget Optimization scenarios → agreed next steps.
Invite pushback. If results contradict the client’s intuition, explore why — this conversation often surfaces important context that improves the next model.
Document all questions and answers during the session.

Step 4.3 — Post-Session Follow-Up

Send a meeting summary and agreed next steps within 24 hours.
Deliver the final pack: PDF deck, Excel scenario tables, methodology note (brief, plain language).
Send NPS survey 1–2 days after the session.

Typical duration: 2–4 days (scheduling, session, and follow-up).

Phase 5 — Retention and Account Development

This phase is often skipped. It is where your most valuable revenue comes from.

Review NPS results within one week.
Run an internal retrospective: what went well, what to improve.
Schedule a 30-day check-in call: has the client acted on any recommendations? What has changed?
If the client is satisfied, present the Co-Pilot retainer — explain that a quarterly model refresh will validate whether the budget changes worked, and set up the next planning cycle.
If the client consents, request a case study or testimonial.
Archive all project files (data, model outputs, deliverables) in the central agency knowledge base.

Total Timeline at a Glance

Phase	Description	Duration
Phase 0	Pre-Sales and Qualification	1–2 weeks
Phase 1	Onboarding and Data Collection	1–2 weeks
Phase 2	Analysis and Modeling	1–2 weeks
Phase 3	Reporting and Deliverables	3–5 business days
Phase 4	Presentation and Follow-Up	2–4 days
Phase 5	Retention and Development	Ongoing (30 days)
Total	Kick-off to Final Delivery	4–6 weeks

The most common cause of delay is slow data delivery from the client. Assigning a named data owner at the kickoff is the single most effective mitigation.

Quality Gates (Non-Negotiable Checkpoints)

No project advances past these points without explicit sign-off:

Gate	Criterion	Owner
QG-1: Data Readiness	≥ 80% data completeness + Data Audit Report approved	Project Lead
QG-2: Model Validity	R² ≥ 0.75 + peer review completed	Analyst + Project Lead
QG-3: Deliverable Review	Executive Summary + Deck approved internally before client delivery	Project Lead
QG-4: Client Handover	Final pack received and confirmed + NPS collected	Account Manager

When QG-2 Is Not Met

If the model does not reach R² ≥ 0.75 after two runs (including the AI Evaluation iteration), do not proceed to reporting. The required response depends on how far below the threshold the model fell:

R² Result	Action
0.60 – 0.74	Deliver with explicit confidence caveats. Disclose the R² score. Label all budget scenarios as directional only. No refund required.
0.45 – 0.59	Do not deliver budget recommendations. Deliver a Diagnostic Report: root cause, findings at low confidence, and a data improvement roadmap. Discuss a reduced fee or credit.
< 0.45	Stop. Deliver a transparent explanation only. No analytical conclusions. Agree a commercial remedy with the client before the Strategy Session.

Before taking any commercial action, diagnose the root cause: is this a data problem (fixable — the client can provide better data) or a structural factor (not fixable — the business model or market makes MMM inherently difficult)? The root cause determines who bears responsibility and what remedy is appropriate.

See: Handling Low Model Accuracy — the full guide on diagnosis, delivery decisions, the refund question, alternatives to cash refunds, and client communication. This is the reference document for any engagement where QG-2 is not met.

Pre-Presentation Checklist

Before every client delivery, verify:

Every channel insight is written in plain business language — no statistical terminology.
There is one clear, actionable recommendation per channel analyzed.
The results pass the business sanity check — are they logically coherent given what the client knows about their market?
Budget scenarios include explicit assumptions and confidence levels.
Limitations have been noted clearly: MMM is backward-looking and works with aggregated data.
If R² is below 0.75, the model fit score is disclosed in the report and all findings are framed as directional.

High-Level SOP: MMM Analysis Service Delivery