Marketing mix modeling agency vs software: what each offers, and how to choose
A marketing mix modeling agency or consultancy builds and runs the model for you, usually as a project or retainer, and presents the findings. MMM software gives your own team a platform to build, refresh and use models directly. An agency suits businesses that want the work done for them. Software suits teams that want to own the measurement.
The two are often presented as rivals, and in practice many businesses use both. Many businesses also meet a third option: the media agency that buys their ads offering to measure them as well. That arrangement raises a separate question about independence. This page explains what each option includes, compares them on six dimensions, covers who should audit the measurement, lists what to ask each kind of provider, and sets out when each option is the better choice, including the hybrid many businesses end up with.
What a marketing mix modeling agency does
A marketing mix modeling agency, or measurement consultancy, takes the business's data, builds a model, and delivers findings: each channel's contribution, response curves, and a recommended budget split. Most also run scenario planning and present the results to marketing and finance.
Engagements are usually a project, often once or twice a year, or a retainer with regular refreshes. The agency's analysts do the modeling, and the client supplies data and reviews the output.
The strength of the model is the people behind it: experienced analysts who bring judgement about data problems, industry patterns and how to present results. The limitation is that the model lives with the agency, and results arrive on the agency's schedule.
Providers come in several forms: specialist econometrics consultancies, the analytics teams inside large media agency groups, and data companies that add their own retail or panel data to the model. They differ in price, depth and independence from the media being measured, so the category name alone says little about what a given engagement includes.
What marketing mix modeling software does
Marketing mix modeling software gives the business a platform to run models itself. It typically connects to ad platforms and sales systems, fits and refreshes models, and provides dashboards, scenario planning and budget allocation tools.
Some platforms are fully self-serve. Others add support from the vendor's analysts, which places them between software and consultancy. Some include experiment design and calibration, and some do not.
The strength is ownership and speed: the team can refresh the model, test scenarios and answer questions without waiting for a project cycle. The requirement is someone in-house with the time and skill to use it well.
Software also changes how often measurement informs decisions. A model that refreshes with new data can be consulted before each budget change, instead of once a year when a study is delivered.
When the media agency measures its own work
Media agencies increasingly offer measurement alongside media buying. An agency that recommends media and reports on its performance produces the evaluation of the party being evaluated. That is a feature of the arrangement and says nothing about anyone's intentions.
It becomes a practical question when a result has to convince finance or a board. The usual safeguard is to make sure someone other than the author can inspect the method, change the assumptions and reproduce the result. That can be the client's own team, a separate consultancy, or a platform the client controls.
An example: a media agency's model shows that the channels it manages performed well last quarter. The finding may well be right. It carries more weight with a finance team when the client can open the same model, see the assumptions behind each channel and rerun it.
The differences, dimension by dimension
Who does the work. Agency: the agency's analysts. Software: your team, with or without vendor support.
Speed and cadence. Agency: results on a project schedule, often quarterly or annually. Software: refreshes as often as the data allows.
Control over assumptions. Agency: set by the agency, and visible if it chooses to share them. Software: visible and adjustable where the platform allows it.
Skills needed in-house. Agency: enough to supply data and question the findings. Software: enough to run the model and interpret it.
Cost structure. Agency: project fees or a retainer, mostly for analyst time. Software: a subscription, plus the internal time to use it.
Continuity. Agency: the model and its history sit with the agency. Software: they sit with the business, on the platform.
Who should audit the measurement
Whichever option is chosen, the result is more useful when someone outside its production can check it. The practical test is whether a reviewer can see the specification, the assumptions and who set them, the fit on past data, and the raw outputs behind each recommendation.
Agencies vary widely here. Some share the full model, and others share only the findings. Software makes the model visible to the client by default, but only to the extent the platform exposes it. Asking to see the assumptions and to change one, in a demo or a pitch, shows quickly where a provider sits.
Questions to ask an MMM agency
Can we see the model specification and the assumptions? And who sets the assumptions: the agency, the client, or both.
How often will the model be refreshed, and at what cost? A once-a-year model is out of date for most of the year.
How are experiments used? A model calibrated with geo tests or lift studies holds up better than one fitted to history alone.
What do we keep if we leave? The data, the outputs and the model settings. Agreeing this at the start avoids rebuilding the history later, when the relationship ends or the team decides to bring measurement in-house.
Is the agency also buying our media? If so, agree how the measurement will be checked independently.
Questions to ask an MMM software vendor
How much of the work does our team do? Setup, data mapping, modeling choices, refreshes and interpretation.
Can we see and change the assumptions? And can we export the outputs.
How are experiments handled? Designed and run in the platform, accepted as inputs, or not supported.
What support is included? Onboarding, analyst time and review of results, and whether that support continues after the first model is live.
Can an agency partner use it on our behalf? Relevant for businesses that want an agency to run the work on a platform the business controls.
When an agency is the better choice
An agency or consultancy is the better choice when the business has no analytics team with time for modeling, when the need is a one-off or annual strategic review, when the data is complex and needs experienced hands, and when the findings have to be presented by an outside expert.
It also suits businesses in categories where an experienced consultancy brings useful external context, for example knowledge of how price and promotions behave in that sector.
When software is the better choice
Software is the better choice when the business wants to own its measurement, when results are needed more often than a project cycle allows, when the team has an analyst who can run the platform, and when the model's assumptions need to stay visible and adjustable to the people who use it.
A hybrid is common: software owned by the business, with an agency or consultancy running the work on it. The business keeps the model and its history, and the agency supplies the time and expertise.
What neither option fixes
Neither option removes the need for data. A model needs enough history behind each channel, and enough movement in each channel's spend, whoever builds it.
Neither makes a model causal on its own. A model fitted to history can mistake a channel that moved with demand for one that drove it, and experiments supply that separation.
And neither guarantees the results get used. Findings have to arrive when budgets are set, in terms the planners use. A report delivered after the budget is agreed describes a decision that has already been made.
How to choose
No analytics capacity, and an annual strategic question: an agency or consultancy.
An analyst who can run it, and a need for frequent answers: software.
An agency relationship you want to keep, and a model you want to own: software run by the agency on your behalf.
Your media agency offers to measure its own work: agree first how the method will be checked by someone else.
Cassandra is a marketing mix modeling platform, on the software side of this comparison. Its production engine is Bayesian, priors are co-set with the client, and geo experiments run in the same system, using the same method family as GeoLift. Reads sit at campaign level, in the planning layer.
White-label is live, running today on partner domains, so an agency can deliver measurement on the platform under its own brand while the method underneath stays inspectable. The dashboards run on the partner's domain, and there is no embed inside the partner's own product.
A price is published.
Moving from an agency to software
Start by asking the current agency for the data it has prepared, the latest outputs and the model's main settings. That material shortens setup and gives a reference point for the new model.
Run the new model on the same period as the last agency study and compare channel by channel. Differences will appear, and each is a chance to see which assumption drove it.
Decide who will run the software day to day before switching. The most common reason a platform goes unused is that nobody has the time to own it.
Consider keeping the agency for the first cycle in an advisory role. It knows the data, the history and the questions the business asks, and that knowledge helps set up the first model and check its results. Many agencies are willing to work on a client's platform, which keeps the relationship while the business takes ownership of the model.
Questions
What does a marketing mix modeling agency do?
It builds and runs a model on the business's data and delivers findings: each channel's contribution, response curves and a recommended budget split, usually as a project or a retainer.
Who are the best vendors for marketing mix modeling?
It depends on the model of service. Consultancies suit businesses that want the work done for them, and software suits teams that want to own the model. Our best-of page lists five platforms and the criterion used to choose them.
Should our media agency measure our marketing?
It can, provided someone else can inspect the method and reproduce the result, because an agency that buys the media and measures it is evaluating its own recommendations.
Is MMM software cheaper than an agency?
Often over time, because the cost is a subscription rather than analyst time for each project. It also needs internal time to run, which should be counted.
Can an agency use MMM software for us?
Yes. Many businesses keep the model on a platform they control and have an agency or consultancy run the work on it.