The basics
Attribution
Attribution is a method that assigns credit for each sale to the marketing touchpoints a customer clicked or viewed before buying.
Last-click attribution gives all credit to the final touchpoint; multi-touch attribution (MTA) spreads it across several. Both rely on tracking individuals, which privacy rules and cookie loss have weakened.
Attribution cannot see channels without clicks, such as TV or out of home, and each ad platform’s own attribution tends to credit itself.
Example
Meta and Google each claim a £100 sale that one customer made after seeing both ads, so platform reports add up to £200 of sales from £100 of revenue. (Illustrative.)
How Delphio uses it
Delphio does not replace attribution for daily optimisation, but gives an independent view of what each channel really adds, for budget decisions.
Related terms
Oct 19, 2026
Econometrics
Econometrics is the use of statistical models to measure relationships in economic and business data, such as how price or advertising affects sales.
Marketing mix modelling is the best-known use of econometrics in marketing. In the UK, “econometrics” and “MMM” are often used for the same thing.
Traditionally, econometrics has been delivered as bespoke consultancy projects taking weeks or months, which put it out of reach of smaller advertisers.
Example
An econometric model estimates that a 5% price rise reduces weekly sales volume by 3%, holding marketing and seasonality constant. (Illustrative.)
How Delphio uses it
Delphio applies the same econometric discipline as a bespoke project, on a platform with expert review, so scaling businesses can use it.
Related terms
Oct 19, 2026
Incrementality
Incrementality is the additional sales caused by a piece of marketing: the sales that would not have happened without it.
Many sales credited to ads would have happened anyway, from loyal customers, word of mouth or people already searching for the brand. Incrementality strips those out.
It can be measured with controlled experiments, such as holdout groups or geo-experiments, or estimated with marketing mix modelling.
Example
An ad platform reports 400 sales from a campaign. A holdout test shows 250 of those people would have bought anyway, so the campaign’s incremental sales are 150. (Illustrative.)
How Delphio uses it
Every contribution and return figure in Delphio is incremental, so budgets are set on what marketing actually changes.
Related terms
Oct 19, 2026
Marketing mix modelling
Marketing mix modelling (MMM) is a statistical method that estimates how much each marketing channel contributes to sales, using historical spend and sales data alongside other factors that move sales.
A model is fitted to weekly data covering media spend, sales, price, promotions, seasonality and market conditions. It separates the sales marketing caused from the sales that would have happened anyway.
Because it uses totals rather than tracking individuals, MMM covers every channel, including TV, radio and out of home, and needs no cookies.
Example
A retailer’s model finds that, of £2m in quarterly sales, £1.5m is base sales and £500k is driven by marketing, of which paid social contributes £180k. (Illustrative.)
How Delphio uses it
Delphio fits a Bayesian marketing mix model to your data, has an expert review it, and turns it into a planning tool with a range on every figure.
Related terms
Oct 19, 2026
How advertising behaves
Adstock
Adstock is the carry-over effect of advertising: the way an ad keeps influencing sales for days or weeks after it runs.
People don’t all respond the moment they see an ad. Adstock models this by letting each week’s advertising effect decay gradually rather than vanish.
The decay rate differs by channel. TV and out of home often carry over for weeks, while paid search effects fade within days.
Example
A TV burst in week 1 drives £40k of extra sales that week. With a 50% weekly carry-over, it adds about £20k in week 2 and £10k in week 3. (Illustrative.)
How Delphio uses it
Delphio estimates adstock separately for each channel, so a channel with a long tail is not undervalued by looking only at the week it ran.
Related terms
Oct 19, 2026
Base sales and incremental sales
Base sales are the sales a business would make with no marketing at all; incremental sales are the extra sales marketing causes on top.
Base sales come from brand strength, distribution, loyal customers and seasonality. They change slowly, though long-term brand building can raise them.
Splitting sales this way shows how much marketing really drives, and stops marketing being credited for sales that would have happened anyway.
Example
Of £500k in quarterly sales, a model attributes £380k to base and £120k to marketing. (Illustrative.)
How Delphio uses it
Delphio shows the base and incremental split on every results page, with the factors that drive the base, such as price and seasonality.
Related terms
Oct 19, 2026
Response curve
A response curve shows the incremental sales a channel is expected to produce at each level of spend.
The curve combines adstock and saturation. It usually rises steeply at first and then flattens as the channel saturates.
Comparing curves across channels shows where extra spend would earn most, and where cuts would cost least.
Example
Paid search’s curve shows £20k a week producing £60k of sales and £40k producing £85k. Doubling spend adds only £25k. (Illustrative.)
How Delphio uses it
Every Delphio model includes a response curve per channel, with a shaded range showing how certain it is.
Related terms
Oct 19, 2026
Saturation (diminishing returns)
Saturation is the point at which extra spend on a channel brings smaller and smaller increases in sales, also called diminishing returns.
Early spend reaches the most responsive audience. As spend rises, ads reach less interested people, see more repeat exposure and pay higher auction prices.
Saturation is why the best budget is rarely “more of whatever has the highest average return”.
Example
The first £10k a week on paid social returns £3 of sales per £1. The next £10k returns £1.50 per £1. (Illustrative.)
How Delphio uses it
Delphio fits a saturation curve per channel, which is what lets it say where the next pound works hardest.
Related terms
Oct 19, 2026
Measuring return
Marginal ROAS
Marginal ROAS is the sales revenue generated by the next pound of spend on a channel, rather than the average across all spend so far.
Because of saturation, marginal ROAS is usually lower than average ROAS, and it falls as spend rises.
Budget decisions should follow marginal ROAS: money should move from channels where the next pound earns least to channels where it earns most.
Example
Paid social has an average ROAS of 2.5x, but the last £10k spent drove £12k of sales, a marginal ROAS of 1.2x. Another £10k there would barely pay back. (Illustrative.)
How Delphio uses it
Delphio’s optimiser and scenario tools work on marginal returns, read from each channel’s response curve.
Related terms
Oct 19, 2026
ROAS (return on ad spend)
ROAS, or return on ad spend, is the sales revenue a channel generates for every pound spent on it, usually written as a multiple such as 2.5x.
Platform-reported ROAS counts every sale the platform can link to an ad, including sales that would have happened anyway. Incremental ROAS counts only the extra sales the ad caused.
ROAS is an average across all spend, so it says nothing about what the next pound will earn. That is marginal ROAS.
Example
Paid social costs £100k in a quarter and drives £250k of incremental sales, so its incremental ROAS is 2.5x. (Illustrative.)
How Delphio uses it
Delphio reports incremental ROAS for every channel, with a range, never platform-reported ROAS.
Related terms
Oct 19, 2026
Method and uncertainty
Bayesian marketing mix modelling
Bayesian marketing mix modelling is an approach to MMM that combines existing knowledge with your data and expresses every result as a range of likely values, not a single number.
A Bayesian model starts from priors (sensible assumptions, such as “advertising doesn’t reduce sales”) and updates them with the data. The output is a distribution of plausible values for each estimate.
This suits marketing data, which is often short and noisy, and makes uncertainty visible instead of hiding it.
Example
Instead of “TV returns 2.2x”, a Bayesian model reports “TV most likely returns 2.2x, and there is a 90% probability the true figure is between 1.7x and 2.8x”. (Illustrative.)
How Delphio uses it
Delphio is built on PyMC-Marketing, an open-source Bayesian modelling library, so its method can be inspected rather than taken on trust.
Related terms
Oct 19, 2026
Credible interval
A credible interval is the range within which a Bayesian model says the true value most likely lies, such as a 90% probability that a channel’s return is between 1.8x and 2.6x.
A narrow interval means the data pins the estimate down; a wide one means it doesn’t. Two plans whose intervals overlap heavily may not be meaningfully different.
Intervals are what let you tell a confident recommendation from a guess.
Example
Email’s return is 3.4x with an interval of 3.1x to 3.7x: well established. Out of home’s is 1.6x with an interval of 0.6x to 2.9x: the data barely constrains it. (Illustrative.)
How Delphio uses it
Every return, contribution and scenario outcome in Delphio comes with its credible interval.
Related terms
Oct 19, 2026
Geo-experiment
A geo-experiment is a test that measures a channel’s incremental effect by changing spend in some regions and comparing their sales with similar regions where spend stays the same.
Because the regions are otherwise alike, the difference in sales shows what the change in spend caused. It is one of the most direct ways to measure incrementality without tracking individuals.
Geo-experiment results can be used to check, or calibrate, a marketing mix model.
Example
A brand pauses paid search in a set of test regions for four weeks. Sales there fall 4% against matched control regions, giving a direct read on paid search’s incremental effect. (Illustrative.)
How Delphio uses it
Where a model’s range is wide, Delphio’s experts can recommend a geo-experiment to sharpen it.
Related terms
Oct 19, 2026
Priors
Priors are the assumptions a Bayesian model starts with before it sees your data, such as the likely range of a channel’s return or how long its effect lasts.
Good priors stop a model reaching impossible conclusions, such as a channel destroying sales, when the data is thin. Where data is strong, it overrides the prior.
Priors should be stated openly, because they shape results most where data is weakest.
Example
Before fitting, a model assumes TV’s carry-over most likely lasts between one and six weeks. The data then narrows this to two to four weeks. (Illustrative.)
How Delphio uses it
Delphio shows which estimates lean on priors and which are driven by your data.
Related terms
Oct 19, 2026
Planning
Budget optimisation
Budget optimisation is finding the split of a marketing budget across channels that is expected to produce the most incremental sales or profit.
An optimiser reads each channel’s response curve and moves money until the next pound earns about the same everywhere. It can also show what happens if the total budget rises or falls.
Results are only as good as the model behind them, so ranges matter: a shift smaller than the uncertainty is not worth making.
Example
Moving £20k a month from generic paid search to paid social is expected to add £30k of sales, with a likely range of £10k to £50k. (Illustrative.)
How Delphio uses it
Delphio’s optimiser recommends a channel split and checks your total budget against history, peers, affordability and growth targets.
Related terms
Oct 19, 2026
Scenario planning
Scenario planning is testing several possible plans, such as different budgets, prices or promotions, against a model to compare their likely outcomes before committing.
Markets do not sit still while you plan. Scenario planning lets you model changes in budget alongside price, distribution and demand, and compare the outcome ranges side by side.
It turns a model of the past into a tool for deciding the next quarter.
Example
A business compares three Q1 plans: baseline, £150k more media with no price change, and a 5% price cut with flat media. The ranges show the extra-media plan raises the ceiling while the price cut mostly raises risk. (Illustrative.)
How Delphio uses it
Scenario planning is Delphio’s core planning tool. Every input is labelled as fitted by the model, forecast, or your own assumption.
Related terms
Oct 19, 2026
