Google published a full user guide and a hands-on tutorial for the Meridian Scenario Planner on August 5 and 6, 2026, moving the marketing mix modeling dashboard into open beta and documenting, for the first time, how shared reports handle credentials.

The Google for Developers documentation page for Meridian Scenario Planner carries an "Open Beta" label and a last-updated stamp of August 6, 2026. One day earlier, on August 5, 2026, the Google Analytics YouTube channel published a seven-minute, thirty-two-second walkthrough titled "Google Meridian | Scenario Planner Overview," presented by Jeff, a developer relations engineer working on Meridian. Together the two artefacts constitute the most detailed public account of how the tool behaves since the feature itself arrived in February 2026.

According to the documentation, Meridian Scenario Planner "is a tool that automates the creation of comprehensive reports that analyze the effectiveness of various marketing channels and their impact on sales or KPIs." The output is a dashboard that supports interactive budget optimization rather than a static deck.

What is new is not the concept but the operating detail: the formulas governing how far spend can be shifted, the error states that break the report, the difference between two classes of dashboard link, and an explicit security recommendation naming which of those links to share.

How a report gets generated

The workflow starts outside the dashboard. A Colab notebook is the designated entry point for report generation, and it assumes at least one trained Meridian model already exists in Google Drive or Google Cloud Storage, with access granted to the notebook. Where none exists, the documentation points to the Meridian Demo Colab.

In the video, the sequence compresses into a few steps. A data science team trains the model in Python and saves the model object to Drive, though other storage connectors are described as viable depending on enterprise setup. Running the notebook processes the model output and generates a unique dashboard link, which opens the report once authorised with a Google account.

Jeff described the intent of the layer in blunt terms: "this is essentially where we transform our marketing budget from guesswork to decisive action." He added that the tool "bridges the gap between your data science team and leadership."

The dashboard is organised into three tabs: Analyze, Plan, and Optimize.

Validating the model before spending against it

The Analyze tab is a diagnostic layer. Its model fit section reports three statistics, each defined in the documentation.

R-squared measures the amount of variation in the data explained by the model, with values closer to 1 indicating a more accurate fit. Mean Absolute Percentage Error (MAPE) measures the mean absolute percentage difference between expected and actual values, with values closer to 0 indicating greater accuracy. Weighted Mean Absolute Percentage Error (wMAPE) applies weighting by actual revenue or KPI. The documentation states that the weighted version is generally preferred because it is less influenced by geographies and weeks carrying relatively small revenue.

The documentation then undercuts any temptation to treat those numbers as proof. Goodness-of-fit metrics, it states, "don't give a complete picture of how good a model is for causal inference," and "The best model for causal inference may differ from the best-fitting model, because causal inference models must also estimate the unobserved baseline."

Jeff made the same point, framing R-squared as "an important confidence check before we start spending money, rather than the ultimate proof of success." The primary goal, he said, is causal inference rather than predictive accuracy.

An expected versus actual outcome chart sits alongside the metrics table, plotting expected, actual, and baseline outcome. Hovering over a point surfaces the time period, the numerical values for each of the three series, and the confidence interval.

Channel contribution and the baseline problem

The channel contribution section separates what marketing did from what would have happened anyway. Baseline, in the documentation's definition, is what would have occurred for a brand or product line in the absence of any marketing channel contribution. Any company expects some sales without marketing, and the documentation notes that brands differ on this point because of established brand recognition and seasonal demand patterns.

The worked example is a 12 percent contribution from television advertising, presented as the kind of figure that helps justify a spend level. Charts break contribution down by baseline, price promotions, and marketing channels, and a separate view compares each channel's share of media spend against its share of incremental revenue or KPI.

In the video, that comparison was described as a fast efficiency test: "If a channel makes up, say 22% of your spend share but drives 34% of your outcome. That's a heavy hitter."

Non-media variables such as pricing and promotions entered the Meridian framework in an update published on September 30, 2025, alongside channel-level contribution priors and binomial adstock decay. The Scenario Planner surfaces those inputs as separate contribution bands rather than folding them into media effects.

Three return metrics that do not agree

The return on investment section carries the distinction most likely to change a budget decision. The documentation defines ROI as incremental outcome per dollar spent, effectiveness as incremental outcome per media unit, and marginal ROI (mROI) as an indicator of the efficiency of additional spend.

The three can point in different directions. According to the documentation, "channels with high ROI but low mROI are likely in the saturation phase," meaning additional investment is unlikely to return what the initial investment did. Channels showing both high ROI and high mROI are described as likely to continue yielding returns as spend increases.

Jeff put the practical reading of mROI more directly: "This tells you the expected return of your very next dollar you spend." Bubble charts allow the three metrics to be compared across channels simultaneously.

Fixed budgets, flexible budgets, and an error state

The Plan tab is where constraints are set. Jeff characterised it as the point where "we act like the CMO and set guardrails for our future budget." In the demonstration, the initial budget was set to 150 million, drawn either from a historical period or entered manually.

Two constraint types are available. A fixed budget finds the optimal allocation across channels for a given total, maximising overall ROI. A flexible budget maximises total incremental outcome, or determines the maximum spend per channel, depending on the target selected.

Flexible budgets require a target type. Target Total ROI maximises total incremental outcome while maintaining a minimum overall ROI. Target channel mROI determines the maximum spend per channel while still achieving a minimum marginal ROI on each.

The documentation flags a hard failure mode: for a fixed budget, setting the parameters intended for the flexible case will cause an error. It also notes that where a target total ROI is set for a flexible optimization, the optimized total ROI may come in below the minimum target value if the spend limit is reached.

The demonstration dashboard showed television at roughly $31 million with an increase of $1.6 million, and social at roughly $23 million with an increase of $4.0 million, as examples of the recommended shifts.

The spend constraint formulas

Channel spend constraints determine the search space the optimizer explores. The documentation publishes the arithmetic.

The minimum channel spend in the exploration space is calculated as one minus the min_spend_shift_ratio, multiplied by historical channel spend. The maximum is one plus the max_spend_shift_ratio, multiplied by the same historical figure. By default the entire exploration space is used.

Additional channel-level bounds can be entered on top of that default. The documentation's example lists Search and Video as spend-bounded channels, with lower bound ratios of -0.1 and -0.2 and upper bound ratios of 15 percent and 25 percent. That configuration sets the available range for Search at between -10 percent and 15 percent of its original spend.

Two formatting rules apply. Lower bound ratios must be non-positive and upper bound ratios non-negative. Decimals and percentages cannot be mixed inside the same input box.

The valid ranges are not fixed. A low initial budget provides less room to adjust the lower bound, and a high initial budget less room to adjust the upper bound, because both are governed by the shift ratios set at the budget step.

In the video, the constraint step was framed as the mechanism that keeps output usable. Jeff described bounding one channel so it neither drops nor increases by more than 10 percent, and another so it does not drop at all but cannot increase by more than 20 percent, adding: "This ensures the planner gives you a realistic executable media plan."

Known failure modes in the report itself

Two warnings in the documentation concern the report software rather than the model.

The first advises pausing report computation before configuring the optimization and resuming afterwards, because the configuration page takes inputs in steps and partial states can cause conflicts and unnecessary computation.

The second is more consequential. According to the documentation, a known issue exists within Data Studio where updating charts on another page while a system error is present may cause those charts to become unresponsive. The stated remedy is to verify that the review table renders correctly before navigating away, and to refresh and restart the report if the issue occurs.

Response curves and what the optimizer actually did

The Optimize tab opens with a summary table comparing optimized total budget, total ROI, and total outcome against non-optimized figures for the selected period, repeated at channel level. Below it sit charts showing the change in optimized spend per channel, the percentage allocation, budget spend per channel, and incremental outcome per channel.

The response curves explain the reasoning. Each plots incremental outcome as a function of spend, with a solid line representing the constraint bounds set for the optimization and a dotted line representing spending outside them. A circle marks current spend; a square marks optimized spend.

A reach and frequency chart is displayed alongside by default, but the two are not synchronised. Selecting a channel from the media channel dropdown updates the response curve, and the reach and frequency chart follows only where such data exists for that channel. Where it does not, the chart is disabled.

Jeff summarised the mechanism as the optimizer "moving dollars away from channels that are flatlining at the top of the curve and pouring them into channels that are still in that steep, high growth phase."

A footnote clarifies what is being maximised. Outcome is typically revenue, but where the KPI is not revenue and revenue-per-KPI data is unavailable, Meridian defines the outcome as the KPI itself.

The section with the clearest operational consequence concerns sharing. The documentation states that Data Studio generates two distinct types of dashboard link.

A temporary creation link, generated from Colab, points to an unsaved instance. A persisted saved link exists only after a user clicks "Edit and Save" on a temporary dashboard.

The credential behaviour differs sharply. When a persisted dashboard is shared, the sharer remains the owner, and viewers effectively impersonate that owner, reading the data on the owner's credentials without any direct access to the underlying source. When a temporary dashboard is shared, the recipient cannot view anything until they save it themselves, at which point they become the owner and must authenticate with their own credentials and access rights.

The documentation's recommendation follows from that asymmetry: "To maximize data security, share the Temporary Dashboard link." Doing so, it states, eliminates the risk of unintended data exposure through owner credential impersonation.

Granting editor access on a persisted dashboard is described as discouraged, because an editor can potentially modify data connections and view unauthorized data using the owner's credentials. The documentation directs users who need to edit toward making their own copy of the report. The broader guidance is to give minimum access and prefer viewer over editor rights.

For organisations whose MMM sits on aggregated media spend, pricing, and revenue data, the distinction is not academic. A saved dashboard forwarded to an agency or an external partner carries the originator's data reach with it.

Looker Studio, Data Studio, and where the page now sits

The documentation and the video use different product names for the same surface. The video describes an interactive Looker Studio dashboard; the documentation's security section refers to Data Studio and to link patterns on the datastudio.google.com domain. That reflects Google's decision, announced on April 11, 2026, to retire the Looker Studio name and return the product to Data Studio with free and Pro tiers.

Within the Meridian documentation tree, the guide sits under Scenario Planning, between the Unified MMM Schema page and an FAQs and troubleshooting page, with content licensed under Creative Commons Attribution 4.0 and code samples under Apache 2.0.

Why this matters for marketing teams

Meridian has expanded steadily since it first appeared in March 2024 on a limited basis and opened globally in January 2025. The framework gained a geo-experimentation layer and an enterprise platform in the pre-Google Marketing Live announcements of May 5, 2026, then moved inside Google Analytics 360 at GML 2026 on May 20, where a Scenario Planner is also offered natively. A May 28 developer episode set out the Hill curve and adstock mathematics behind those products.

The open beta guide addresses a different constituency from those announcements. Analytics 360 integration serves advertisers on an enterprise contract. The Colab-plus-dashboard path documented here serves teams that already run the open-source model themselves and now need to hand its output to people who will not read Python.

That handover is where the discipline has historically leaked. Research found nearly 40 percent of marketers said their organisations struggled to connect MMM outputs to business decisions. IAB's State of Data 2026 report, published February 7, 2026, found up to 75 percent of buy-side decision-makers considered attribution, incrementality tests and marketing mix models to underperform on rigour, timeliness, trust and efficiency. Incubeta research released on May 6, 2026, recorded 70.4 percent of leaders confident their budgets were deployed effectively while 41.6 percent conceded waste.

Published constraint formulas and named error states change the terms of that debate slightly. A media plan produced through the Scenario Planner can now be audited against documented bounds rather than accepted as an opaque recommendation, which matters when the plan is used, as Jeff put it, "when you need to justify budget shifts to your finance team or leadership like your CMO or CFO."

The governance question remains open. A PPC Land analysis published on April 2, 2026, examined the structural dynamic when a platform builds the MMM that guides budget allocation, using Meta's Robyn as the case. CIMM returned to the same territory in late July 2026, warning that default priors and embedded assumptions in widely adopted open-source frameworks can reconfigure data asymmetry rather than remove it. Publishing the exploration-space arithmetic is a partial answer to that critique. It does not settle it.

Timeline

Summary

Who: Google published the material through its Google for Developers documentation site and the Google Analytics YouTube channel. The video walkthrough was presented by Jeff, a developer relations engineer working on Google Meridian. The audience is marketing measurement teams, data scientists running Meridian models, and the marketing and finance leadership who receive the resulting media plans.

What: The Meridian Scenario Planner user guide is published as an open beta, covering report generation from a Colab notebook, model fit diagnostics using R-squared, MAPE and wMAPE, channel contribution and baseline separation, three distinct return metrics including marginal ROI, fixed and flexible budget optimization with target total ROI and target channel mROI options, channel spend constraint formulas built on min and max spend shift ratios, response curve interpretation, and a data security section distinguishing temporary from persisted dashboard links.

When: The tutorial video was published on August 5, 2026. The documentation page carries a last-updated stamp of August 6, 2026. The Scenario Planner itself launched on February 19, 2026.

Where: The user guide is hosted on Google's developer documentation site under the Meridian Scenario Planning section, positioned between the Unified MMM Schema page and an FAQs and troubleshooting page. The dashboard runs in Data Studio, formerly Looker Studio, with models loaded from Google Drive or Google Cloud Storage.

Why: Marketing mix modeling outputs have repeatedly failed to reach budget decisions, with research finding that nearly 40 percent of marketers struggle to connect MMM results to business choices and up to 75 percent of buy-side decision-makers rating current measurement methods as underperforming. Documenting the optimizer's exploration space, its error states, and its credential handling makes the resulting media plan auditable rather than opaque, and identifies a data exposure path that exists whenever a saved dashboard is forwarded outside the team that built it.