All guides, by feature
Marketing Mix Modeling
Build marketing mix models and plan spend with the Budget Optimizer.
At a glance
- Where in Orbit
- Sidebar › Workspace › Marketing Mix Modeling
- Who can use it
- Anyone at your agency, once Orbit has switched Marketing Mix Modeling on for it. Training and the Budget Optimizer work for the clients in your Client Access.
- Availability
- If enabled for your agency
What it’s for
Marketing Mix Modeling (MMM) trains a statistical model on one client’s historical performance to estimate how much each channel really drives. Orbit uses Google’s Meridian, a Bayesian model.
Each model covers one client, one date range and one KPI: conversions or revenue. Its dashboard shows how well it fits, each channel’s ROI and marginal ROI, response curves, and a recommendation to scale up, maintain or reduce each channel.
The Budget Optimizer then uses a model to split a total budget across channels for the most return, and shows what the new split is expected to add.
Orbit switches Marketing Mix Modeling on per agency, and it’s off until then. Each agency can start 10 training runs in any 24 hours, with one model training at a time per client.
Guides
Questions
What is marketing mix modeling?
A statistical model trained on one client’s historical performance to estimate how much each channel drives, and how returns change with spend. Orbit uses Google’s Meridian.
Who can use Marketing Mix Modeling?
Everyone at an agency where it’s switched on, whatever their role. Training and the Budget Optimizer work only for clients in your Client Access.
How much data do I need to train a model?
Orbit recommends at least 365 days. With less, Model Builder warns “⚠️ Meridian recommends at least 1 year (365 days) of data for stable estimates.”
Can I choose the dates a model trains on?
Yes. Start Date * and End Date * fill in with the client’s full available range, and you can narrow them.
Should I choose Revenue or Non-Revenue (Conversions)?
Choose Revenue to “Optimize for total revenue generated”, or Non-Revenue (Conversions) to “Optimize for conversion count (purchases, leads, signups)”.
Can I build a geographic or regional model?
No. Models are National: one market for the whole client.
How long does a model take to train?
Orbit says it “typically takes 5-10 minutes”. It trains in the background, so you can leave the page.
Where do I find a model after it’s trained?
Under Trained Models on the Marketing Mix Modeling page, once it has finished. Use the client filter if you have several.
Can I train two models for the same client at once?
No. One model can train per client at a time. Models for different clients can train together.
How many models can we train a day?
Your agency can start 10 training runs in any 24 hours, counting every run, including failed ones. Email support if you need more.
Does a model update as new data comes in?
No. A trained model stays as it is. Train a new model to include newer data.
Can I change a model’s advanced settings or priors?
No. Model Builder offers the client, dates, KPI and model type only.
What does Model Accuracy mean?
How well the model fits the client’s data, as an R² percentage. Above 85% is “✅ Excellent fit”, above 70% “⚠️ Good fit”, and anything lower “❌ Needs improvement”.
What does the Health number on a model mean?
A 0 to 100 score for how well training converged. A model with Health 0 didn’t converge, so don’t rely on its results.
What’s the difference between ROI and Marginal ROI?
ROI is total return divided by total spend so far. Marginal ROI is the return from the next dollar spent, so it shows whether more spend would still pay.
Where does Total Revenue on a model come from?
It’s the return the model credits to marketing: each channel’s spend times its ROI, added up across channels.
How does Orbit decide Scale Up, Maintain or Reduce?
Scale Up when ROI is above 2 and Marginal ROI above 1.5. Reduce when ROI is below 1 or Marginal ROI below 0.5. Every other channel is Maintain.
What does the Uncertainty column mean?
How wide the channel’s ROI spread is compared with its ROI: 🟢 Low under 30%, 🟡 Medium under 60%, otherwise 🔴 High. The higher it is, the less sure the model is.
What does “If you double spend” mean on the response curves?
How much the channel’s return would grow if its spend doubled. Over 150% shows “Opportunity! Room to scale”; under 80%, “Diminishing returns - consider reallocating”.
What budget should I enter in the Budget Optimizer?
The total you want to plan. It starts at the “Current total”: the model’s spend over its dates. Enter another figure to test a scenario.
Plan a budget with the Budget OptimizerWhat the Budget Optimizer does
What do Increase, Decrease and Maintain mean in the optimizer?
They say which way to move a channel’s spend. Maintain means the recommended change is under 5%.
What is Expected Revenue Lift?
The extra return the model expects from the recommended split, compared with how the budget is split now.
Can I set a minimum or maximum spend for a channel?
No. The Budget Optimizer takes a total budget only. Allow for minimum spends and contracts when you act on its results.
Can I save or export optimizer results or a model dashboard?
No. Results aren’t saved and there’s no export. Run the optimizer again to see them.
Can I delete or rename a model?
No. Models can’t be deleted, renamed or compared.
Can I cancel a model that’s training?
No. Orbit has no way to cancel a training run.
Why does it say “Client is outside your assigned scope”?
The client isn’t in your Client Access, so you can’t train or optimize for them. An Owner or Admin can add them in Team.
Train a marketing mix modelClient Access: which clients someone sees
Can clients see marketing mix models in their portal?
No. Marketing Mix Modeling is for your agency only.
How it works
Explainer: What a marketing mix model is
A model is one training run for one client, over one date range, against one KPI. It estimates how much each channel contributed to that client’s results, and how returns change as spend changes.
Orbit trains it with Google’s Meridian. In Orbit’s words: “MMM uses Bayesian statistical modeling to measure the impact of your marketing activities on business outcomes.”
A trained model doesn’t update. To include newer data, train a new model for the client.
| Setting | What it means |
|---|---|
| Client * | The one client whose data the model learns from. |
| Start Date * and End Date * | The period it learns from. They start as the client’s full available range. |
| Non-Revenue (Conversions) | “Optimize for conversion count (purchases, leads, signups)”. |
| Revenue | “Optimize for total revenue generated”. |
| National | The only Model Type *: “Single national market (recommended)”. |
Limits: How much data a model needs
Orbit recommends at least a year of the client’s data. When you choose a client, Model Builder shows how many days are available and warns if it’s under 365.
- Model Builder shows “Available data:” with the number of days and the first and last date.
- Under 365 days, it warns: “⚠️ Meridian recommends at least 1 year (365 days) of data for stable estimates.”
- Start Date * and End Date * fill in with the whole available range. You can narrow them, at the cost of less data to learn from.
Explainer: Model statuses
A model is training, completed or failed. Only completed models are listed under Trained Models and in the Budget Optimizer.
| Status | What you see | What to do |
|---|---|---|
| Training | “Model Training in Progress” and “Training Model...” on the model’s page. | Wait. Orbit says it “typically takes 5-10 minutes”. |
| Completed | Listed under Trained Models with its KPI, channel count and Health score. | Open it to see its dashboard. |
| Failed | “Training Failed” on the model’s page. Orbit doesn’t show the reason. | Train a new model, or contact support if it keeps failing. |
- Only one model can train per client at a time. Another client’s model can train alongside it.
Explainer: The summary cards and Health score
The four cards at the top of a model’s dashboard say how well the model fits, and what marketing returned overall.
| Card | What it shows |
|---|---|
| Model Accuracy | How well the model fits the data, as an R² percentage. Above 85%: “✅ Excellent fit”. Above 70%: “⚠️ Good fit”. Otherwise “❌ Needs improvement”. |
| Overall ROI | Return on ad spend across all channels, as in “Every $1 spent returns $2.40”. |
| Total Spend | Spend across all channels over the model’s dates. |
| Total Revenue | The return the model credits to marketing: each channel’s spend times its ROI, added up. |
| Health | Shown on Trained Models. A 0 to 100 score for how well training converged. 0 means it didn’t converge, so don’t rely on its results. |
- The dashboard’s footer lists the model’s Date Range, Observations, Channels and a short Model ID.
Explainer: Reading the channel table
Channel Performance Breakdown lists each channel’s spend, ROI and share of results, with a recommendation to scale up, maintain or reduce it.
| Column | What it shows |
|---|---|
| Spend and % of Budget | What the channel spent over the model’s dates, and its share of all spend. |
| ROI (90% CI) | The channel’s average ROI, with a low and high figure underneath. |
| Marginal ROI | What the next dollar spent on the channel would return. |
| Revenue | The return the model credits to the channel. |
| Contribution | The channel’s share of the total return credited to marketing. |
| Uncertainty | How wide the ROI’s low-to-high spread is next to the ROI: 🟢 Low under 30%, 🟡 Medium under 60%, otherwise 🔴 High. |
| Scale Up | ROI above 2 and Marginal ROI above 1.5. |
| Reduce | ROI below 1, or Marginal ROI below 0.5. |
| Maintain | Every other channel. |
- Under the table, Channels to Scale, Channels to Maintain and Channels to Reduce count each recommendation.
Explainer: ROI, marginal ROI and response curves
Average ROI is what a channel has returned per dollar. Marginal ROI is what the next dollar would return. Response curves show how returns flatten as spend grows.
Orbit defines them on the dashboard: “Average ROI = Total return divided by total spend (historical performance)” and “Marginal ROI = Return from the next dollar spent (current efficiency)”.
A channel can have a high average ROI and a low marginal ROI. It has paid off so far, but extra spend there returns less.
| Label | When it shows |
|---|---|
| 🚀 High Performers | ROI above 2.0x. |
| 📈 Growth Potential | Marginal ROI above ROI, with ROI above 1. |
| ⚠️ Needs Attention | ROI below 1, under the break-even line at 1.0x. |
| Opportunity! Room to scale | Doubling the channel’s spend would add more than 150% to its return. |
| Moderate efficiency - maintain | Doubling spend would add between 80% and 150%. |
| Diminishing returns - consider reallocating | Doubling spend would add less than 80%. |
- Each response curve plots Marketing Spend ($) against Incremental Revenue ($), with a dashed line at the channel’s current spend.
- Under the chart, each channel shows “If you double spend:” and the change in return.
Explainer: What the Budget Optimizer does
The Budget Optimizer takes a total budget and uses a model’s response curves to split it across channels for the most return. It shows how much to move and what that’s expected to add.
| Item | What it is |
|---|---|
| Total Marketing Budget | The total to split. It starts at the “Current total”: the model’s spend across channels over its dates. |
| Expected Revenue Lift | The extra return the model expects from the new split, compared with the current one. |
| Optimized ROI | The return on investment the model expects from the new split. |
| Total Budget | The budget the plan spreads across channels. |
| Current Spend and Optimized Spend | Each channel’s spend now, and in the recommended split. |
| Change ($) and Change (%) | How far each channel moves. The biggest moves are listed first. |
| Action | Maintain if the change is under 5%, otherwise Increase or Decrease. |
- Each model’s dashboard already shows the best split of its current spend under Optimal allocation at your current budget. Use Scenario planning to try a different total.
- You can’t set a minimum or maximum for a channel, and results aren’t saved. Run it again to see them.
- Orbit’s own notes advise: “Consider practical constraints (minimum spends, contracts) when implementing” and “Re-run optimization quarterly or when market conditions change significantly”.
Limits: Marketing Mix Modeling: availability and limits
Orbit switches Marketing Mix Modeling on per agency. Where it’s on, anyone at the agency can train and use models, up to 10 training runs in any 24 hours.
- It’s off until Orbit switches it on for your agency. Until then, it isn’t in the sidebar.
- Every role can train models, open them and use the Budget Optimizer. Training and optimizing work only for clients in your Client Access.
- Your agency can start 10 training runs in any 24 hours. Every run counts, including ones that fail.
- One model can train at a time for each client.
- Models are national only.
- Models can’t be cancelled, deleted, renamed, compared or exported, and their advanced settings can’t be changed.
- The Budget Optimizer can’t take per-channel minimums or maximums, and doesn’t save scenarios.
- Clients don’t see Marketing Mix Modeling in their portal.
Explainer: Client Access: which clients someone sees
Client Access sets which clients a teammate can see, along with those clients’ plans, reports, pacing and data. You set it per person once they’ve accepted their invite.
| Option | What they see |
|---|---|
| All clients | Every client at your agency. |
| All Paid Search clients, or their own specialty | Every client with the service that matches their Specialty Role, including clients added later. |
| Assigned clients only | Only the clients you tick. Star one to mark them as that client’s lead. |
- Until you choose, a new teammate sees all clients. Set their access as soon as they show as ACTIVE.
- Changing someone to Specialist switches them to Assigned clients only and opens the panel so you can tick their clients. Changing them to another role gives them all clients again.
- Someone with no clients, such as Assigned clients only with none ticked, sees no client data, pacing or insights.