Marketing Mix Modeling
On this page
A marketing mix model estimates how much each channel contributed to an outcome across your spend history, and what a channel returned per unit of spend. Ask for it in a conversation when a straight comparison of channel metrics cannot answer the question, typically budget allocation: where the next increment of spend is worth putting, and which channels have stopped paying back.
The model captures two effects a query cannot: carryover, where spend keeps working after the period it was spent in, and diminishing returns, where extra spend in a channel buys progressively less. Once a fit is complete, the same model can split a future budget across your channels.
There is no separate screen for it. You ask in Chat, the model is fitted in the background, and the result comes back into the conversation.
Prerequisites
-
The Editor role or above to fit a model or optimize a budget. A Viewer can read results, see Permissions
-
The spend and outcome data in a connected data source or an attached file
-
Workspace storage configured, if you want to optimize a budget from the fit afterwards. The fitted model is saved there
The model is fitted on a table you assemble in the conversation, so ask for the figures first, then for the model. The table needs:
Scroll horizontally to see the full table.
Column | Requirement |
|---|---|
Date | One row per period at a consistent cadence: daily, weekly, or monthly. No missing and no duplicate dates |
Channel spend | One numeric column per channel, in its reporting currency. Do not rescale it |
Outcome | The numeric measure you are modelling, for example revenue or conversions |
Controls (optional) | Numeric columns for other drivers you want held constant, for example price or promotion |
A fit needs at least 30 rows of history, and nothing is fitted below that. Thirty rows is a floor rather than a target: carryover and seasonality are estimated far better from two years of weekly data.
Decide these before you ask
The model can only answer the question you specify, and the specification is a business decision rather than a technical one. Adverity Atlas asks you for these in a short form before it fits, with what it can infer already filled in. The outcome is never filled in for you.
-
Which outcome. Revenue, conversions, bookings, or pipeline are different models with different answers. Pick the one the budget decision turns on, and be specific about whether it is gross or net.
-
Which channels. Leaving a real channel out pushes its effect onto the channels you kept, which inflates them. Include everything that drives the outcome, not only paid media.
-
What counts as spend. Working media only, or media plus fees. Keep it consistent across channels.
-
Which period. Longer history identifies the effects better. A window that spans a major change in the business, a repricing or a rebrand, needs a control column for it.
The data review before the fit
Adverity Atlas reviews the table before it fits anything and tells you what it found, check by check. Each finding comes with the fix that suits it.
Scroll horizontally to see the full table.
Check | What it looks for | Blocks the fit |
|---|---|---|
Incomplete result | The query hit the row limit, so the table is cut off | Yes, until the data is aggregated in the query |
Missing values | Empty or non-numeric values, missing or duplicate dates | Yes, until fixed |
Columns that never vary | A channel or control with the same value in every row | Yes, until the column is dropped |
Cadence | Gaps in the date series | No, needs your decision |
History length | Less than a year of periods (or one to two years, as a note) | No, needs your decision |
Zero spend | A channel with no spend in more than 30% of periods | No, needs your decision |
Outliers | Periods far outside the usual range | No, needs your decision |
Channels that move together | Two channels whose spend is strongly correlated | No, needs your decision |
The fixes offered are concrete: fill a channel's spend with zero for the periods it was off, insert the missing periods, merge two channels that always moved together, or keep a real one-off event and add a control column for it. Missing values are never filled in silently, because a gap would be read as zero rather than as unknown. More than a tenth of a series missing is a problem to fix at the source, not to fill.
A chart of the outcome and total spend over time renders under the review with the findings drawn on it: outlier periods as red points, gaps as yellow bands, periods with a missing value as dashed lines.
Blocking findings have to be fixed before a fit starts. The others need a decision from you, and the fit starts once you have made it.
Fitting a model
-
Ask for the data, for example weekly spend per channel and weekly revenue for the last two years.
-
Ask for the model. Adverity Atlas sends the short form described above, then runs the data review.
-
Settle the review's findings. The fit starts in the background and the conversation stays usable while it runs. It also shows under Tasks in the account menu as Fit marketing mix model, see Task Executions.
-
When it finishes, a Marketing Mix Model card appears in the conversation on its own, with a Show result button, and you get a notification. You do not need to wait on it or ask again.
Fitting takes minutes rather than seconds and stops at a 30 minute limit. A fit that hits it can be retried with a shorter sampling run, fewer channels, or a shorter history window. A fit that fails for another reason says why in the conversation, and re-running it is safe.
You can ask for the model settings to be adjusted, most usefully a longer sampling run when a first pass comes back with wide ranges, a seasonality term for a business with a strong annual cycle, or a longer carryover window for brand channels.
Reading the result
Show result renders the channel breakdown with a chart comparing each channel's share of spend with its share of the effect. The result covers each channel you included:
-
Return per unit of spend. What the channel returned over the period, relative to what was spent on it. Channels with little spend produce unstable and often implausibly large values, so treat an outlier on a small channel as an artefact rather than a finding.
-
Contribution share. How much of the modelled outcome the channel accounts for. Shares are relative to the total the model explains, so they are useful for ranking channels rather than as absolute amounts.
-
Carryover. How long spend in the channel keeps working, reported as the number of periods it takes for the effect to halve. Brand and video channels typically carry for longer than search.
-
Diminishing returns. Where the channel sits on its response curve, which tells you whether more spend in it is still productive.
The model's estimates are reported as ranges rather than single numbers, and the width of the range is part of the answer. A channel whose range is wide enough to include both a good and a poor return has not been measured well enough to reallocate budget on, and is a candidate for a proper test instead.
Adverity Atlas also checks whether the fit settled before it reports, and flags values it does not trust. Treat a flagged number as unavailable rather than as a weak signal.
Ask how well the model fits, or about the response and carryover shapes, and three more charts render: response curves per channel with current spend marked, actual against fitted over the fit window, and carryover per channel.
Optimizing channel budgets
Once a fit is complete, ask Adverity Atlas to allocate a budget across its channels for a future window, for example "split 400k across these channels for next quarter". It uses the fitted response and carryover curves to find the per-channel spend that maximizes the expected outcome.
You set the total, the start and end of the planning window, and any limits that matter. By default every channel is kept between half and twice its spend in the fit window, because that is the range the model has actually seen. Widen it for a more aggressive reallocation, hold a channel where it is, or cap one that cannot absorb more. A channel that spent nothing in the fit window stays at zero. You can also ask it to lean on the channels the model is confident about instead of maximizing the raw expected return, which is the honest choice when several channels have wide ranges.
The optimization runs in the background with a 10 minute limit. When it finishes, a Budget plan card appears in the conversation, and Show plan brings back the allocation next to current spend, the change per channel, and the expected outcome with its range, against the outcome of the current split. Four charts come with it: current against optimized spend over the window, the change per channel, the response curves with both the current and the proposed spend marked, and the spread of the expected outcome for each split. The response curves are the explanation: money moves off channels sitting on the flat part of their curve and onto channels still on the steep part.
Each plan carries a verdict:
Scroll horizontally to see the full table.
Verdict | Meaning |
|---|---|
Usable | Every channel's curve was pinned down by the fit, and no proposal goes past the spend the model has seen |
Directional | At least one channel was held at its current spend because its curve was not pinned down, or a proposal goes past the highest spend in the fit. Read it as a direction, not a number |
Not allocatable | No channel's curve was pinned down, so there is nothing to reallocate on |
The plan is a starting point, not an instruction. The optimizer maximizes the model's expected response and knows nothing about reach, creative capacity, or contractual commitments you did not give it as limits.
Looking up an earlier fit
Ask about a past model or budget plan in any conversation and Adverity Atlas retrieves it, including its status if it is still running. The notification for each finished fit links back to the conversation it ran in. Retrieval is scoped to the workspace the fit was run in.
Permissions
Scroll horizontally to see the full table.
Action | Required role |
|---|---|
Read a marketing mix model result or budget plan | Viewer |
Fit a marketing mix model, optimize a budget | Editor |
The Analyst role can chat about data but cannot fit a model. See Members and roles.
What's next
-
Custom Models, specify the model form yourself, or model price instead of media
-
Glossary, define the outcome measure before you model it
-
Knowledge, record a result so the next model can build on it
-
Artifacts, turn the breakdown into something you can share
Adverity Atlas 2026.11.4. Powered by AI.