MMMAgentic Marketing Mix Modeling · Part of ScanmarQED

Six MMM engines. One dataset. Results you can repeat .

Run Robyn, Meridian, PyMC and our own engines side by side, in the app or from Claude, ChatGPT and Copilot, with MMM experts on hand when you need them. Choose your model on evidence, not habit.

Insights PROJECTBig CheeseDATASETBig CheeseCONFIGURATIONNational - Logged SAMPLE DATA · ILLUSTRATIVE RESULTS
Executive SummaryROI ResultsDecomposition ChartsResponse Curves
ROI ResultsReturn on spend by channel for the chosen model.
SELECT MODEL TYPE
Ad Spend, Contribution, and ROI by Variable OLS · R² 0.91 · Current plan
Ad Spend
TV Search OOH Social Video Radio
$ 0$ 3.0M$ 6.0M
Contribution
TV Search OOH Social Video Radio
$ 0$ 5.0M$ 10.0M
ROI (x)
TV Search OOH Social Video Radio
0.02.55+
MOLS
CATEGORY⋮AD⋮CONTRIBUTION⋮ROI (X)
›TV———
›Search———
›OOH———
›Social———
›Video———
›Radio———
© 2026 MMMLabs · sample data, illustrative resultsContact Us
  • Part of ScanmarQED: 25+ years in marketing measurement
  • 300+ brands
  • ISO 27001:2022 certified · GDPR compliant
  • EU and US data residency available
  • Listed in Claude and ChatGPT directories

Agentic MMM. From raw data to budget plan in one conversation

Work with the MMM Labs agent in the app or in Claude, ChatGPT and Copilot. It prepares your data, runs the engines and drafts the budget plan. Our engines calculate every number, our experts can review or take over any step, and nothing changes until you approve it.

Ask

Ask about your own model results and get a clear answer, with every figure traced back to the engine that calculated it.

Act

The agent prepares your data, runs and compares the engines, adds your priors and builds the budget plan inside MMM Labs. It shows you what it will create or change and waits for your go-ahead.

Automate

Your own AI agents and workflows can use the same MCP tools to trigger model runs, retrieve results and bring insights into your reports. Combine MMM Labs with your files and other tools in one workflow.

WORKS WITH
Claude
ChatGPT
Copilot
Any MCP tool

Listed in Claude's Connectors Directory and the ChatGPT app directory. Also works with Microsoft Copilot and any MCP-compatible assistant.

Built on MCP, the open standard for connecting AI assistants to tools, so you are not tied to one AI vendor.

Claude
MMM Labs connected
Here's this year's weekly data for Big Cheese. Run the engines on it. XLSbig_cheese_weekly.xlsx
✓run_data_actionupload_file

It's uploaded. I found 104 weeks, 6 media channels, 3 controls and sales as the KPI. The readiness check passed, so I started all six engines. OLS was back in seconds and all six were done in under 10 minutes.

Which one should I trust?
✓run_results_actioncompare_models✓run_results_actionget_model_status

Meridian predicts the held back weeks best and its ROI ranges are the tightest. Robyn comes close behind. All six agree on which channels lead, so the ranking holds up.

Holdout fit, R², top four of sixExample values
Meridian0.90
Robyn0.88
PyMC0.87
OLS0.84
Fold in what we already know about OOH.
✓config_actionapply_business_prior✓model_actionrerun_model

Your lift test is in as a prior on OOH, range 0.2 to 4.0, and the model is rerun. The proposed annual budget is $4.2M, budget neutral: 8% more into TV, 5% out of OOH. Shall I apply it?

Which channel is saturated?
✓run_results_actionget_response_curves✓run_results_actionget_roi

TV has flattened out. Past about $80k a week, each extra dollar returns less than Search, which still has room to grow.

Reply to Claude↑
Example conversation with example values. Real MCP tool names.
WORKFLOW
MMM LABS RESPONDS

Variables are categorized and a readiness check runs.

Same platform underneath. One source of truth for chat and dashboard.

MCP uses the same tools as the in app assistant and writes the same saved files as the UI. You sign in with your MMM Labs login, every call runs with your own permissions in your US or EU region, and deletes wait for your approval.

Same tools, same saved files as the app
Your login, your own permissions
Data stays in its US or EU region
Deletes wait for your approval
Claude
MMM Labs connected
Delete last quarter's test scenario.
✓confirm_destructive_actiondelete_scenario
Confirm delete or cancelTest scenario, Q3
Advanced Diagnostics PROJECTBig Cheese CONFIGURATIONNational - Logged
Model Fit SummaryContributionsPaid Media AnalysisDriver AnalysisParameters
Test Dataset Performance
R²
90%
Coefficient of Determination
SMAPE
5%
Symmetric MAPE
Full Dataset Performance
R²
93%
Coefficient of Determination
SMAPE
4%
Symmetric MAPE
Same result, same saved fileThe champion, open in the web app

The problem. Does this sound familiar?

Data everywhere, but nowhere to start

Spend, sales and promo data sit in different systems and formats. Weeks go into cleanup before the first model runs.

Platform core.
Every major MMM engine, in one place

Set up your data once. Run any engine, or all of them, on the same configuration and compare the outputs side by side.

06 ENGINES

Bring what you already know

Lift tests, geo experiments and earlier studies go into the model as priors, so results start from what your team has already proven. Set a range per channel, rerun, and the source stays recorded next to every prior.

ModelingPROJECTBig CheeseCONFIGURATIONNational - Logged
Configurations LibraryModel SettingsData ReadinessRun ModelsLeaderboard
Business priors
CHANNELSOURCEROI RANGESTATUS
OOHLift test, Q20.2 – 4.0
TVGeo experiment0.5 – 3.0
SearchPrevious MMM1.0 – 5.0
3 of 3 priors applied · stored with National - Logged · example values
0.1–5.8 → 0.6–2.4
OOH ROI range before and after adding the lift-test prior. Example values.

Test more, trust more

See fit, ROI, contribution and elasticity for every engine in one table. When independent methods agree, you have evidence. When one model disagrees, you see it before it moves budget. Every contribution figure is labeled with the reporting setting behind it, so when two people quote different numbers for the same model, you can see why.

Advanced DiagnosticsPROJECTBig CheeseCONFIGURATIONNational - Logged
Model Fit SummaryContributionsPaid Media AnalysisDriver AnalysisParametersData
Actual vs. Fitted Sales
HOLDOUT
Actual salesFitted · OLSMAPE 3.8%

Plan the next budget, not just report the last one

Turn model results into forecasts, scenarios and an optimized allocation within your own constraints, with the projected lift shown before you commit.

Scenario PlannerPROJECTBig CheeseCONFIGURATIONNational - Logged
ConstraintsResults
SPEND SHARE · DEMO DATA
Spend Share by Channel
—
—
—
—
—
—
TVSearchOOHSocialVideoRadio

Know what works, and when to stop spending

Response curves show where each channel saturates and what the next dollar returns (marginal ROI), so you cut before returns flatten, not after.

InsightsPROJECTBig CheeseCONFIGURATIONNational - Logged
ROI ResultsDecomposition ChartsResponse Curves
Response Curve · TV
SPEND $1.24M · mROI 1.00Optimal

Always-on MMM. New data in. Your model setup stays.

Drop next month's files into a dataset you have already configured. MMM Labs adds the new weeks and keeps your categorization, hierarchies, priors and model configurations.

Bring data from Excel, CSV, Google Sheets, Snowflake, BigQuery, Databricks, PostgreSQL, MySQL or SQL Server. Auto Categorize classifies every column, while readiness checks flag issues before any model runs. Then set variables, pooling and the ROI you already believe as a business prior.

From
Excel Excel CSV CSV Google Sheets Google Sheets Snowflake Snowflake BigQuery BigQuery Databricks Databricks PostgreSQL PostgreSQL MySQL MySQL SQL Server SQL Server
See a refresh in the demo →
Data HubPROJECTBig CheeseCONFIGURATIONNational - Logged
DatasetsSourcesData Setup
Need a starting point? Download sample template (Excel or CSV)
↻ Process Data
Contact Us
01

Drop in CSV or Excel. Files land in the dataset you already configured, no new project needed. Click the drop zone to upload.

See every change before it is saved

Each refresh is checked against your existing data first. New rows, restated values and new columns are listed sheet by sheet, and nothing is saved until you confirm.

Your models stay protected

A refresh that would break a model is blocked, with the affected configuration named. Anything that no longer fits the data is flagged, never silently deleted.

Refresh on your decision cycle

Update monthly, or whenever the budget conversation needs it. Each refresh is a routine update, not a new modeling project.

Multi-engine runs. Stop choosing an engine. Run all six in minutes.

OLS, Constrained OLS, Robyn, PyMC, Adaptive MMM and Meridian all read the same configuration, so results compare directly. OLS returns in seconds and full runs finish in minutes. PyMC runs went from 98 minutes to 5.9, from 51 to 16 and from 26 to 4.5.

Example run, all six engines at once.
PyMC runs cut from 98 to 5.9 minutes.
ModelingPROJECTBig CheeseCONFIGURATIONNational - Logged
Data ReadinessRun ModelsLeaderboardContributions ManagerResults Compare
Run ModelsBig Cheese › Big Cheese Weekly › Modeling 6 COMPLETED · 0 RUNNING
ENGINESTATUSRUNTIME
OLSRegression Completed 14s
Constrained OLSOurs Completed 2m 41s
Adaptive MMMOurs Completed 4m 13s
RobynMeta Completed 6m 6s
PyMCPyMC Marketing Completed 7m 49s
MeridianGoogle Best fit 9m 31s
© 2026 MMMLabs · Big Cheese demo dataContact Us

Honest validation. The winner is chosen on weeks it never saw.

Every run is scored on held back weeks, and the run with the best holdout accuracy becomes champion. You can override it, and the champion feeds Insights, the Scenario Planner and the deck.

MeridianCHAMPION

Best holdout accuracy of six engines in the Big Cheese example project. You can override it.

93%
R² on the full data
90%
R² on held out weeks
4%
Error, symmetric MAPE
2.2x
Return on media spend
ModelingPROJECTBig CheeseCONFIGURATIONNational - Logged
Contributions ManagerResults Compare
Model Statistics
METRICFULL DATASETHOLDOUT SET
R293%90%
ADJ_R292%N/A
DW1.961.66
MAPE4%5%
NRMSE0.050.07
SMAPE4%5%
Every quality check, full data, train and holdout

Big Cheese demo data

Security. Ready for enterprise security and IT review

ISO 27001:2022GDPRAI OPTIONAL

Certified security

ISO 27001:2022 certified and GDPR compliant, from a company that has handled client marketing data for 25+ years.

Support. Work with us the way your team works today

Same platform, three levels of support. The agent does the legwork in every mode; our experts take on as much as you want. Move between them as your team grows.

Change how you work at any time. No re-platforming, no lost work, no new contract to learn a new tool.

Find the right fit →

Process. From upload to budget decision in three steps

See the engines and methodology →

Use cases. Built for your industry

Media agencies

Run MMM across every client on one platform. Refresh each client's data without rebuilding the setup, and model every market without hiring more modelers.

Food and beverage

Separate promotions and seasonality from media to invest where marketing really moves demand.

Finance

Know which channels bring in high-value customers, not just applications.

Retail and e-commerce

Find the real drivers of growth behind the platform ROAS numbers.

Gaming

Optimize ad spend for active users, not just installs.

Tech and durables

See how media shapes demand across long purchase cycles, from brand to sales.

koen-pauwels-headshot520x520px
Prof. dr. Koen Pauwels Distinguished Professor of Marketing Northeastern University Strategic Advisor, ScanmarQED, Co-founder, MMM Labs

Co-founder. Built with Prof. dr. Koen Pauwels

Prof. dr. Koen Pauwels co-founded MMM Labs and serves as Strategic Advisor to ScanmarQED. MMM Labs joined the ScanmarQED Group in 2026.

FAQ. Questions buyers ask

Agentic MMM is Marketing Mix Modeling run by an AI agent. You give it a task, and it prepares the data, runs and compares models, and builds the budget plan. In MMM Labs, the agent proposes each step and waits for your approval. Every number comes from the modeling engines, never from the AI. Our experts can check the agent's work, guide your team or run it for you.

See all six engines run on data like yours