Multi-touch attribution and budget optimisation for every channel, online and offline.
Search, social, affiliates and price comparison sites all claim the same order. See which channels really drive profitable sales, after returns.
Most marketing budgets are set by last year's split, gut feeling, or whichever channel reports the most conversions. But channels don't work alone. TV drives search, display prepares the click, the catalogue brings people to the store. Our data-driven attribution model learns from every customer journey, converted and non-converted, online and offline, and measures how your channels work together. We then set your budget based on these interactions, not on each channel's past results in isolation.
Same budget, better performance. See which channels really earn their share, move money away from the ones that only collect credit, and get more from the spend you already have.
Last-click attribution credits whoever touched the ball last. This short film shows what that does to your budget, and how we credit every channel for its contribution.
Voice and music are AI-generated. Values shown are illustrative.Read the transcript
When a striker scores, the scoreboard shows one name.
But the goal started with the defender who won the ball, and the playmaker who found the pass.
Last-click attribution is the scoreboard. It rewards whoever touched the ball last, so budgets flow to brand search while social, newsletters and price comparison get cut.
Our data-driven attribution model learns from every customer journey, converted and not, online and offline, and measures how your channels work together.
Then our budget optimisation moves spend to where it works hardest, and re-optimises it every month.
Credit every channel based on its contribution to success. Book a free 30-minute call at prodata.ai.
Your priorities
What eCommerce marketing leaders are asked in 2026, and how we help
Ad costs rise every quarter, returns eat into margin, and finance wants proof before approving the next budget. These are the questions we hear most, and what changes with the solution.
“Ads cost more every quarter.”
Auction prices on social and shopping ads keep climbing, so every euro has to work harder.
What changes
You see where extra spend still pays back and where it no longer does, and move budget before the next price increase.
“ROAS looks fine. Profit doesn't.”
Channels are judged on revenue, before returns, discounts and costs.
What changes
Every channel is measured on contribution margin after returns, so budget follows profitable orders, not just orders.
“Every platform claims the same order.”
Add up the platform reports and you get more sales than the shop actually made.
What changes
One independent view across all vendors and networks, including the ones Google and Meta do not measure.
“New customers keep getting more expensive.”
Retargeting and brand search collect the credit for customers other channels won.
What changes
You see which channels really bring new customers, and what those customers are worth.
“We can see less and less.”
Consent banners, shorter cookies and AI search leave gaps in every journey report.
What changes
Sessions are stitched into complete journeys, and offline, TV and AI signals are measured instead of ignored.
“Finance wants proof.”
A budget shift based on platform numbers rarely survives the CFO.
What changes
Every recommendation shows its assumptions, comes as a range, and is tested in the market before rollout.
The problem
Five rules, one blind spot
Most reports split the credit for an order with a fixed rule, and only between clicks. Here is how five common rules, and our data-driven model, divide the same journey.
The journeyTV spotno click→Display→SEO→SEA→Direct→Order
First clickAll credit to the first click
TV: not seen
Display100%
Last clickAll credit to the last click
TV: not seen
Direct100%
LinearEqual shares for every click
TV: not seen
Display25%SEO25%SEA25%Direct25%
Position-basedFirst and last click weigh most
TV: not seen
Display40%SEO10%SEA10%Direct40%
Time decayThe later the click, the more
TV: not seen
Display10%SEO20%SEA30%Direct40%
ProDataAILearned from dataLearned from all journeys, TV included
TV: counted
TV30%Display20%SEO15%SEA25%Direct10%
The TV spot never produces a click, so click-based rules cannot see it at all. Last click even crowns Direct, the channel that merely closed the order. ProDataAI values are illustrative.
All of these rule-based methods look only at journeys that converted, and only at clicks. They're hypotheses, not analysis.
Why the numbers mislead
Last-touch gets gamed
It rewards whatever happens to be last, not what persuaded the customer. Brand search and email clicks look like heroes; upper-funnel display and video look useless.
Channels are judged alone
Each channel is scored on its own numbers, although channels amplify each other: TV lifts search, catalogues lift store visits.
Fixed rules ignore your business
Your customers' decision process depends on your product, offer and creative. The model has to learn from your data.
Offline and AI are invisible
TV, radio, OOH, print, catalogues and stores sit in a separate silo, often while being the largest budget items. AI assistants barely register at all.
Unstable numbers
A model whose channel credits jump every time it is re-run can't be used as a KPI, or to negotiate with agencies and networks.
Results
Running in production, improving the numbers that matter
Our attribution and budget optimisation is deployed in production at multiple organisations. By moving budget to the channels that really drive sales, it has improved three numbers our clients report on.
Revenue
More sales from the same budget, because spend follows the channels that create demand, not the ones that collect the last click.
Contribution margin
Channels are judged on profitable orders after returns, so money stops flowing to channels whose sales come back.
New customers
Channels that win new customers get the credit they earn, instead of brand search and retargeting taking it.
In the first call we walk you through comparable projects from retail and fashion, telecom, travel and online marketplaces.
Our approach
All journeys, not just the winners
Whether a channel persuades is a comparison: how often does it appear in journeys that end in a purchase, and how often in journeys that don't? That question can only be answered with both.
Our model8 of 8 journeys used
Rule-based models4 of 8 used: half the evidence is thrown away
Order
TVAI assistantSEADirect
seen by rule-based models
No order
DisplaySocial
ignored by rule-based models
Order
CatalogueSEOStore
seen by rule-based models
No order
AffiliateSEA
ignored by rule-based models
Order
RadioOOHNewsletterDirect
seen by rule-based models
No order
TVDisplay
ignored by rule-based models
Order
SocialAI assistantPrice comparisonDirect
seen by rule-based models
No order
OOHSocialSEA
ignored by rule-based models
Our approach uses all journeys, converted and non-converted.OnlineOfflineAI assistant
What the comparison reveals
DisplayAffiliate
Only in journeys without an order
However often they show up, they did not persuade anyone here. Rule-based models never notice, because they never look at these journeys.
AI assistant
Only in journeys that ended in an order
A strong signal, even though it never gets the last click. Last-click reporting would give it nothing.
SEASocial
In winning and losing journeys alike
Their value depends on the channels they are combined with, which is exactly what the model measures.
Illustrative journeys. In a real project the model compares millions of them.
ProDataAI
Journeys without conversion are used too
Credit is assigned with a mathematical method based on conditional probabilities within journeys
Journey combinations with a higher response are weighted more heavily
Every journey is needed, so this is a real big-data problem, and our pipeline is built for it
vs.
Rule-based attribution
Only converting journeys count
Simple models with many assumptions; the “correct” weight of each touchpoint is never estimated
A hypothetical, not an analytical, approach
Most of the data is discarded before the rule is applied
Modules
Three modules, built in order
The first module lays the foundation: your data, your customer journeys and a validated attribution model. It takes the most time and is built once. The other two turn it into answers, and you can add one or both.
Module 1The foundation Takes the longest: built once, during the pilot
Attribution Modeler Module
Built once, used by everything that follows
This is where most of the work happens, and it is what makes every later answer trustworthy. We bring your online and offline data together, rebuild complete customer journeys, calculate the attribution from them, train the TV and radio response models, and validate everything before anyone relies on it.
Data foundation
Ad exposures, clicks, costs, orders, CRM and offline response connected in one pipeline on your cloud.
Complete journeys
Sessions and devices stitched into full journeys, including AI referrals and offline touchpoints.
Your business rules
Returns, cancellations, service contacts and brand search handled the way your business counts them.
A validated model
Credit for every order split across its touchpoints, checked for accuracy and stability before anyone relies on it.
What you have at the end
One trusted record of which channels earned each order: the single source for reporting, analytics and budgeting.
Every budget meeting asks the same three questions: how much, where and when. The module answers them with scenarios your finance team can follow and sign off, based on how your channels actually influence each other.
How much?
The total budget that still pays back, and the point where extra spend stops working.
Where?
The split across channels, online and offline, including the ones that prepare a sale rather than close it.
When?
Timing per month or week, so spend lands when it moves the most.
Where the next euro works hardest
Every channel reaches a point where more money buys less. The module shows where each of your channels sits on that curve, so budget moves out of saturated channels and into the ones with room to grow.
Walk into the budget meeting with options, not one number
Budgets are tight, and finance wants to see what each euro buys. Instead of defending one plan line by line, you show three: what you would cut, keep or add, each with the expected effect on the number your board tracks, from revenue and contribution margin to new customers. Your commitments stay fixed: booked TV, minimum brand presence, channel caps.
Conserve
Budget down 15%
Which cuts hurt least?
Today
Recommended
Maintain
Same budget
What is the better split of the same money?
Today
Recommended
Accelerate
Budget up 15%
Where does extra money pay back first?
Today
Recommended
TVRadio & OOHSearchSocialDisplay
Respects your rules: TV contracts, minimum brand presence, channel caps
Illustrative splits
A monthly loop
1
Measure
The attribution model credits every order to the channels that actually triggered it, so the plan starts from the right numbers.
2
Predict
A forecast of how each channel responds to more or less budget, including its effect on the other channels, for example TV lifting search. Forecasts come as a range, not a single number, and are updated every month.
Budgets are based on how channels influence each other, not on each channel's past results in isolation.
3
Plan
The recommended split per channel and period for each scenario, within your constraints and aimed at the goal you set.
4
Learn
Plan versus actual, every month. What we learn feeds the next cycle, and holdout tests keep the model honest.
What your finance team gets
Traceable
Every recommendation shows the assumptions and constraints behind it. No black box.
Tested
Budget shifts are validated with a holdout test before they are rolled out.
Tracked
Plan versus actual every month, in the dashboard you already use.
Is this for you?
Built for companies with serious online marketing
Custom attribution pays off when there is enough budget and data to move. Here is an honest check.
A good fit if
You have a dedicated web analytics or data team
You invest serious money in marketing, typically from around €5 million a year, spread across several paid channels
You have enough orders and customer journeys for patterns to show, ideally with two years of history
You want budget decisions you can defend to finance, not platform reports
Probably not (yet) if
You are a smaller company without web analytics people: a ready-made attribution tool from a software provider is the better, cheaper choice
You run only one or two channels
You have little order history
Not sure? That is what the free first call is for. We tell you honestly if a standard tool fits you better.
A faster start with the data you already have
We connect all your data sources and bring them together. If you already export your GA4 data and can provide the data of your other channels, setup is much simpler and the whole process much faster.
How we work together
Three steps. You decide after each one.
You never commit to more than the next step. Each one ends with a clear result and a decision that is yours to make.
1 2-3 weeks
Assessment
What happens
We connect your data and put your channels side by side: what last click says they deliver, and what they really contribute. Then we set your current budget against each channel's real performance.
What you get
Your channels side by side: last click vs. data-driven
Your budget set against each channel's real contribution
A data check: what is there, what is missing, what it takes
Your decision
Is it worth going further? If not, we tell you, and you stop here.
2 8-12 weeks
Pilot
What happens
We build the attribution foundation on your data and make a first budget recommendation, then test it in the market on part of your budget.
What you get
The attribution model running on your data, with proof it is reliable
What each channel, network and campaign really contributes
A budget recommendation, tested in the market before you roll it out
Your decision
Did the recommendation hold up in the market? Then roll it out.
3 every month
Run
What happens
Every month the numbers are refreshed and the budget recommendation updated, so your plan keeps up with the market.
What you get
A monthly budget recommendation per channel
A what-if simulator: “what if we move 10% from TV to social?”
A dashboard in the tool you already use: Looker, Power BI or Tableau
Everything on your cloud, documented and yours
Your decision
We keep running it for you, or your own team takes over.
The data we work with
User-level ad exposures and clicks (ad server, web analytics or clickstream)
Campaign cost data
Conversions and the journeys that did not convert: both are needed
TV and radio spot logs (GRPs, timestamps)
CRM and order data
Offline response data: codes, calls, store sales
History: ideally two years or more, so seasonality can be learned
Your data stays yours
We work with pseudonymised identifiers, run the pipeline on your infrastructure, and sign a data processing agreement under Art. 28 GDPR wherever we can see personal data. Details on our Trust & Data page.
Why ProDataAI
Built by people who have run this in production
20+ years in data and AI, 100+ projects delivered.
Developed and productised user-level attribution, including offline channels, for a SaaS marketing attribution platform.
Refined over many engagements with advertisers in retail and fashion, telecom, travel, pharma and healthcare, industrial and manufacturing, financial services, and online marketplaces.
Delivered cross-media budget optimisation for large telecom and retail advertisers.
Senior roles at a global technology leader, and Director of AI Engineering at a European retailer.
Hands-on modelling combined with C-suite advisory: the person who builds your model is the one who explains it to your board.
Same budget. Better performance.
Same budget, better performance. See which channels really earn their share, move money away from the ones that only collect credit, and get more from the spend you already have.
Your contact from the first call to the live system. 20+ years in data and AI, hands-on in every project.
For your data team
Your data team gets the full picture: the methodology documentation, the validation report (stability of each channel's credit, agreement between independent checks, holdout results) and the code. We walk them through it in the assessment, under NDA, before anything is built.
FAQ
Frequently asked questions
How is this different from GA4 or platform-reported attribution?
Platform-reported numbers come from the platforms themselves, and each one tends to credit its own channel. Google's data-driven attribution in GA4 is a real step up from last click: it also compares converting and non-converting paths. But it only covers Google's own digital channels and the touchpoints Google can track. We cover every vendor and network, add offline channels such as TV, radio, out-of-home (OOH), catalogues and stores, and connect attribution to budget optimisation. The model is transparent, and it is yours.
Can you measure AI assistants and AI answers in search?
Partly directly, partly by inference, and we are open about which is which. Visits that arrive with an identifiable AI referrer become a channel of their own instead of disappearing into direct traffic. Much of the influence never produces a click, so we add post-purchase survey answers and measure the lift AI visibility leaves in brand search and direct traffic against a baseline, the same way we measure a TV spot.
Why do non-converting journeys matter?
Because influence is a comparison. If a channel appears just as often in journeys that end without a purchase as in those that end with one, it probably did not persuade anyone. Only by looking at both can a model estimate how much a channel actually raises the probability of conversion. Rule-based models that look only at converted journeys cannot tell a channel that persuades from one that is simply present.
Isn't one strong channel enough?
Rarely. A striker without playmakers gets very few chances. Brand search looks strong because it closes, but much of its volume is created earlier in the journey by TV, display, catalogues or an AI assistant's recommendation. Cut those and the closing channel dries up a few weeks later. The model measures these combinations, so you can see which channels work as a team.
Why not just use a rule like linear, position-based or time decay?
Because the rule decides the answer before looking at your data. Linear assumes every touchpoint adds the same value, position-based assumes the first and last matter most, time decay assumes the most recent matters most. None of these weights is ever estimated, and all of them use only converted journeys. They are hypotheses. A data-driven model tests them against your own customers' behaviour.
How do you know the model is right?
We check four things. Stability: the probabilistic model is calculated directly from your journeys, not trained, so we recalculate it on many resampled sets of journeys and measure how much each channel's credit moves. Agreement: it has to tell the same story as the stabilised regression model, which is fitted on your data and tested on users it has not seen. Response models: the TV and radio baselines are trained and tested on airings they have not seen. And in the pilot we validate the budget recommendation with a holdout (incrementality) test in the market.
Do we need user-level data for TV and radio?
No. For TV and radio we measure the response per spot: the traffic and conversion uplift in the minutes after each airing, compared with a baseline. That uplift is then assigned probabilistically to the journeys that were active in those windows. Voucher codes, call tracking, catalogue IDs and loyalty IDs add further offline signals where you have them.
How does it relate to media mix modelling?
Media mix modelling is a valid tool, and we use its strengths: it covers offline channels and it forecasts well. But it works on aggregated weekly channel data and models each channel separately before adding them up, so it cannot see user journeys or how channels interact. We attribute at the journey level and combine that with a multivariate forecast for budget planning, so the budget reflects how channels influence each other.
Who owns the model and code?
You do. The model, the code and the pipeline run on your cloud (GCP, Azure or AWS), are documented, and are handed over to your team. If you prefer, we keep running the monthly optimisation for you as a managed service.
Find out what your channels are really worth
Start with a free 30-minute call. Tell us your channel mix and how you measure today. We tell you honestly whether data-driven attribution would change your budget decisions, and what an assessment would involve.
