In the UK, advertising spend is projected to pass £50 billion in 2026 (yes, that’s a letter “B”), mainly driven by the notion of more investment into video campaigns. Despite the massive investment into businesses scaling their ad spend, 60-75% of marketers report that their current measurement systems fail to deliver the transparency required to justify their spend. Been there and done that. My sympathies are with you.
Source: Media Performance
Attribution in itself is still an elusive concept; most of us senior marketers struggle to use our sources of truth as an indicator of what we did right, what we did wrong, and more importantly, what we should do next.
This write-up is about elaborating on the various angles to understand the black box concept of attribution. Having been in situations where certain stakeholders struggled to understand that the data we marketers get isn’t as deterministic as we hope, friction within teams, and how I delivered my output became harder.
It is a fact that 90% of a marketer’s job is in educating internal/external personnel. That’s been the case for me and my work at Proverve - whether that’s for a start-up, scale-up or a corporate household brand like BMW, Meta, Audible or Microsoft.
I’ll close off this intro with a real example. I briefly consulted for a dating app company, and without going into specifics, part of my gig was to explain (at a high-level) how our paid vs organic “verified sign-ups” were trending + explain the ins/outs of the “why”. I’d then go on to channel-specific narratives, and then slowly trickle into explaining the land of Apple Search Ads (ASA). Oh dear.
If you work in the mobile app marketing world, you’ll know where I’m heading next. Essentially, ASA doesn’t help you view post-install events like a “sign-up” or “paid subscription” natively within its dashboard (unless you use a tool like SplitMetrics). This meant that even though install data was pushing through to a Looker dashboard I built, we couldn’t tie the spend to a paid sign-up.
That wasn’t even the problem; the problem was the doubt in my team's value (the growth/paid team) - with the reason being that all the product managers for said company didn’t understand elements like SKAN, the basics of attribution and data fragmentation (more on this later).
A lot of this write-up is 100% inspired by having to redeploy my approach to explaining attribution countless times, which can get annoying and, worse, a morale drain. Not good.
Anyways, let’s squeeze the lemon and drink the juice…if you’re up for it.
Attribution As a Narrative (Not a Deterministic Truth)
**To make this write-up more contextual and additive, I’d highly encourage you to check my video titled “The Accountability Questions CMOs and Senior Marketers Have to Justify When Marketing” >>> HERE
When we talk about attribution, it’s simply the act of assigning a cause or a source to an effect. In marketing lingo, it's how marketing channels and platforms claim an event (or conversion) is happening.
Example: John Doe sees an ad on YouTube for your product > he then searches for your brand name on Google and clicks on your paid search ad > he buys the product > finally, Google Search attributes (claims) this conversion to itself in-platform AND your analytics tool (Google Analytics 4).
The issue with believing the portrayal of your data’s truth is that…it isn’t true. Or at least, not 100%. Here are variables that distort attribution (not extensive).
The Complexity of The Modern Day Customer Journey
The reality is that the process of how customers go from the discovery phase, to the research/comparison phase, to the intent of buying and finally, the buying phase, is incredibly complex.
As of 2026, and likely the next couple of decades, I highly doubt we’ll get to a point where every marketer will be able to see a perfect view of the journey with 100% deterministic accuracy.
We’re not even talking about device and location-based touchpoints, i.e. customers seeing your marketing efforts offline and converting online (and vice versa), as well as web-to-app conversions (and vice versa).
Next up.
Data Discrepancy
If there is one thing that data is good at, it’s inconsistencies. We’ll talk about different sources of truth next. For now, it’s vital to know that the data you may be seeing is affected by external variables, causing you to see a distorted view of the truth.
Data discrepancy is simply us marketers not having a specific methodology in identifying why the data is incorrect in the first place. It’s the same as doing data analysis when the data you're analysing is completely fragmented or low quality, and if you combine this with your own confirmation bias, you may decide on a marketing output that is detrimental to growth.
First step is identifying how the data is ingested, all the variables that may affect its portrayal and deployment attribution data and piece together a narrative that gets you closer to the truth. Speaking of…
Different Sources of Truth
If attribution itself was a friction point already, if we then include the fact that we marketers (and business leaders) view data using different sources, data becomes even more fragmented and again, leading to a distorted view of what’s actually moving the needle when growing your primary KPIs.
For us performance marketers, it’s guaranteed that we will be viewing ad network data (Google, Meta, TikTok). Whereas your CEO and senior managers will be looking at a BI tool of some form (Tablue, Power BI). Then, your data science teams will be creating their own dashboards (God knows how they ingest the data). Finally, your product team will be using specific analytics tools (Mixpanel, Amplitude).
Need I carry on?
A single source of truth (or a central intelligence hub) is a marketer's dream. One I’d willingly pay money for. Anyways, back to reality.
In an ideal situation, having a data warehouse (Snowflake, BigQuery) can somewhat get us to a solution by integrating data from multiple sources (ad network, CRM, MMP, analytics tools) into a unified view. Orchestrating this type of project is a whole other topic, but it’s one worth thinking about if you want to solve your attribution woes and are investing heavily into marketing.
Privacy Changes
Privacy is another factor that skews our view of attribution. Mainly driven by pressures in the EU (where user privacy is a priority), there have been lots of changes made in how businesses can acquire, store and more importantly, utilise data.
Top-level changes are:
- The deprecation of 3rd-party cookies (Safari/Firefox specific). Whereas Google officially retired the "Privacy Sandbox" branding and several core APIs (including the Attribution Reporting API). The current reality is the "User Choice" model, where third-party cookies are supported by default, but users are given clear, persistent controls to block them.
- Apple’s privacy-centric framework, called SKAN (store kit ad network) - now updated to the updated version called “AdAttributionKit” (AAK) model (the successor to SKAN and handles re-engagement tracking). SKAN makes marketing data more anonymous, time-delayed and within limited conversion values, i.e. blocked ID for Advertisers (IDFA) or capped signals.
- Incorrect Pixel and/or Server-side Set Up. This is low-hanging fruit - I’d urge you to qualify the data and events you're tracking immediately, and if they are working correctly. I’ve worked with clients of all kinds, so I know this is an easy one to fix, whether that’s making sure each event has an event_id (to avoid duplicate counting of unique events), making your revenue-based events are tracked, or ingesting lead cycle status as differents conversion in your ad networks via CAPI.
- Delays in reporting and different channel-specific attribution windows (more on this later).
Data Quality (Briding the Confidence Gap)
Another aspect of marketing attribution that distorts our understanding (leading to potentially incorrect next steps) is a grasp of how high/low the quality of the data you’re seeing is.
Low-quality data can misguide your budget allocations, manipulate your forecasted KPIs, mess up your AI and machine learning signals (server-side ad networks), and expose your lack of governance in how KPIs are reported on.
Been there and done that.
Here’s a top-level framework to go about qualifying the quality of your data (good alliteration).
- Accuracy: How accurate is the data you receive? For example, if you get an MQL, does that person’s info match their actual LinkedIn profile? Another example would be the number of events for a specific conversion - what do you see in your database vs in-platform? How far apart or discrepant are they?
- Completeness: Does your customer fill out mandatory sections, and are they qualified? How do the non-mandatory sections affect your view of the data?
- Timeliness: How quickly (or slowly) does your data refresh regarding its attribution and the actual view of the data (these are two separate processes). An example would be that GA4 can take 24-48 hours to fully ingest an event, and your dashboard may only take 12 hours.
- Uniqueness: Does data give you a view of unique customers or does it include duplicates (same record) for the same customer? Does each conversion have its own row in your database or sum?
- Event Match Quality Score (server-side exclusive): Thankfully, most major ad networks can provide this in some form, sometimes called ‘EMQ Score’ or “Event Match Rate”. On Meta, you simply go to your dataset ID’s EMQ score and see where you land out of a score from 0-10, with 10 being the highest quality - ideally aim for above 8. Whereas in Google, the alternative is getting a diagnosis of the conversion event, i.e. excellence, good, needs attention, urgent.
Different Types of Reporting
We’ve already talked about how your data sources can expose different qualitative and quantitative projections; it’s worth mentioning the practical ways we actually view this data, more importantly, what’s available.
Business Intelligence Tools
These platforms (Microsoft Power BI, Tableau, Looker Studio, Thoughtspot) are designed for cross-departmental analysis and function as the visualisation layer on top of a centralised data warehouse.
Platform or Ad Network-specific
These are native environments used primarily for real-time tactical adjustments and monitoring the technical health of platform signals - platforms would be Meta Ads, GA4, Mixpanel, Google Ads, TikTok, basically anything where the data you see is native within the platform.
3rd-party Dashboards from Data Analytics SaaS
These purpose-built platforms (funnel.io, Segment Stream, Triple-whale, Fospha) automate the heavy lifting of data ingestion and normalisation, providing decision-ready insights faster than custom builds.
Different Attribution Models, Types and Windows
For the most part, most of your attribution (how a source gets credit) will either be last-click (internal CRM or custom-made database) OR very likely, data-driven, where the algorithm claims part (or a whole) of a conversion. It’s still worth going over the different attribution models, types and windows, and how they work. You’ll thank me later.
Attribution Model
This is basically the terminology used to describe the mathematical rules for assigning value to specific touchpoints of a customer’s journey.
- Single-Touch: First-click, Last-click, and Last Non-Direct Click.
- Multi-Touch: Linear, Time-decay, U-shaped, and W-shaped.
- Algorithmic: In other words, Data-Driven Attribution (DDA), which uses machine learning to assign credit based on custom account patterns.
Attribution Types (Methodology)
This is specifically talking about the underlying technology a specific platform uses to identify a user and connect their actions to a conversion point.
Deterministic: Relies on a verified 1:1 match using persistent identifiers like hashed emails, CRM IDs, or device IDs (IDFA) - relates to the Event Quality portion of this write-up from earlier.
Probabilistic: Uses statistical inference and behavioural patterns to "predict" the likelihood of a match when exact identifiers are missing due to privacy restrictions.
Attribution Windows + Interaction Categories
Click-Through Attribution (CTA): Credits actions following a direct link click. As of March 2026, Meta narrowed this to link clicks only, excluding social elements.
View-Through Attribution (VTA): Credits actions following an ad impression where no click occurred.
Engage-Through Attribution (ETA): A new category introduced in early 2026 (Meta, Tiktok, YouTube-specific). It credits conversions following non-link interactions, such as social likes, saves, or watching a video for at least five seconds.
Attribution window is essentially the date range (or window of time) a platform gives itself before claiming credit for a conversion, i.e. 7-day click means that if a user clicks on an ad and converts within 7 days, that specific platform will claim it as a conversion. If that same user converts on day 8 instead, the conversion will not be claimed as it has passed the 7-day window.
The Language Application
So now that we’ve talked about attribution as a narrative (as opposed to a deterministic truth), as well as the complexity of the modern-day customer journey, and data quality and discrepancies - all of that knowledge would be useless without having the right language application for each stakeholder or team within your business.
With the ideal goal to break silos and unify the team for the same goal, which in a capitalistic application is called growth 😉
In essence, every person you’ll interact with as a marketer needs a different explanation, point-of-view and post-application (output) of the numbers you claim to have gained, AKA attribute to.
Founder or CEO
In the best-case scenario, your Founder/CEO will be focused on acting as a strategic growth engine. They’ll need to know if marketing is building a sustainable and scalable machine to either sustain or increase the company's total addressable market (TAM) without burning through cash (stage dependent). I’ve worked with Founders who only cared about the primary KPIs like blended CAC and others who cared about every single element. Be prepared for both situations.
Board of Directors/Execs
These people generally don’t place much value on marketing, and generally individual channel performance or KPI is "noise” - they’d rather see overall monthly growth vs your Meta Ads’ ROAS/CAC. They are, however, looking for strategic guardrails to protect their garden and see any blind spots you may be missing. I’ve had a 1-1 with a board of exec and our discussion was always about “what more can we do”, the answer is…plenty.
Chief Financial Officer or Chief Revenue Ops
These personnel are all about capital allocation, forecasting expenditure and risk management; they might as well be called “the risk-averse officer”. They generally view marketing as an expense (rather than an investment), so expect to be rigorously justified against the Profit & Loss - a more modern take on this when it comes to advertising is “Profit-on-Ad-Spend” (POAS).
This is not the time to talk about your creative ideas and incrementality; they only care about where the diminishing returns are from your marketing budget, i.e. what’s the cap of investment needed for the business to survive. You’d be lucky to find a CFO who cares about growth (they're more geared towards reduction by nature).
Brand Marketing
Brand marketers, God bless them, are mostly transitioning into being metric-driven, even though brand growth is an ethereal concept. They are generally inclined to push messaging agendas (brand campaigns), build long-term brand equity and customer demand.
I talk more about how brand marketers can justify ROI HERE - but in short, they’d be desperate for any metric or narrative that can make their campaigns of value, including: share of model, brand search volume, BLS lifts, ad recall or memorability, brand sentiment (surveys) and incremental reach.
Creative (Marketing)
This is one of the bigger shifts that has happened in my work as a performance marketing consultant, which is an emphasis on synergising paid media with creative studio teams. As of 2023, creative teams (creative director, designers, videographers/editors) are such a huge lever for advertising that even they want to start understanding the numbers in detail (due to the pressure of the changing economy).
Thankfully, most of the creative people (or teams) I’ve worked with are generally interested, albeit you have to simplify the narrative. What works? What doesn’t? Why? Why not? What’s next? What do each of the most important KPIs mean?
It’s also worth noting that when you're presenting the performance of your creatives, these are broken down by funnel (TOFO, BOFU), marketing channel, metrics and use case (brand vs acquisition).
The nuances are huge. I also recommend doing these recap/review sessions monthly and aggregating learning into a Creative Learning document. Keep a cadence on this, especially if you’re investing heavily in Paid Social channels.
Growth/Paid Marketing
My people. For the most part, performance marketers tend to be data savvy, although in bigger organisations (or agency environments), the visibility on other sources of truth gets blocked, either due to access issues, compliance, or no one internally pushing for it.
Growth and performance marketers would ideally want to see other sources of truth (and how their metrics differ), and more importantly, the other data narratives different parts of the business are seeing and how they can complement their work.
Product or Product Managers
These are an interesting bunch. In many ways, I see product teams as mini-founders or CEOs, so the narrative they want to know is highly related to their garden (their specific metric, program or product they need to manage and grow).
As product people aren’t necessarily marketers, you’ll likely spend your time tying the metrics to their agendas for their specific sprints. If the product is early, they might want to know what creatives and copy work and why. If the product is mature, they want to know other levers that can scale customer acquisition or retention.
Data Science, Engineering, IT, & DevOps (Techies)
Thankfully, most engineers or IT teams don’t give a damn about marketing (trust me, I’ve spoken to lots). If there was a case where explaining marketing attribution becomes vital, they’d want to know more about the technicality of how the data gets ingested, retrieved, stored and deployed/viewed - which is why you’re reading this write-up!
Devs are a particularly handy team to be on your side, they can help you create proper server-side infrastructures, quality assure broken attribution and help with setting them up in the first place.
For data teams, most discussions I’ve had with them are pretty light touch, surprisingly. They may not care about the specifics of how metrics are trending (unless they are pressured by their managers), but they would want to know how you go about your data analysis conquests.
Data teams care more about the narrative of the data vs the actual numbers in question. Unless they are a data person who leads a growth team…then you’re screwed
If I missed any other business teams, forgive me; I’m only writing about teams I’ve personally worked with.
The conclusion is that every team fundamentally cares about how marketing attribution benefits their own agendas - ideally, you find a synergy with as many teams as possible, so both of you can have a shared agenda and a much-needed value exchange.
BONUS: Going Beyond Platform-data
If you’re still with me, thank you for reading up to this point. Your attribution conundrums are likely a significant ambiguity that needs fixing!
Let’s close off with a few practical ways to go beyond in-platform data (assuming you have at least 18 months' worth of quality data).
Marketing/Media Mix Modelling (MMM)
Marketing Mix Modelling (MMM) is a great strategic investment for any brand that spends over 7-figures/year, and that has a multi-channel media mix (CTV, DOOH, Offline, PR, digital advertising, affiliate marketing).
Unlike digital attribution, MMM provides a top-down view of how all channels drive business outcomes like revenue and brand awareness. You won’t get a super detailed/specific analysis, but rather a narrative that can tie your past actions to the future.
It can also reveal channel scalability that drives diminishing returns, showing you exactly where additional spend will no longer yield a profitable return, even if platform-reported average ROAS remains high.
Worth calling out that you generally need at least 18-24 months of data for the analysis to make sense. You also need to know the ins and outs of what’s happened business and marketing-wise throughout this (part of the narrative piece).
Marketing Mix Modelling platforms worth prospecting are:
- Triple Whale: A Shopify-native "operating system" that unifies cross-channel attribution, financial profitability, and creative analytics for high-growth e-commerce brands. https://www.triplewhale.com/
- Sellforte: A retail-focused Bayesian MMM platform that provides daily, campaign-level bidding and spend recommendations via built-in AI media agents. https://sellforte.com/
- Cometly: An AI-powered attribution platform utilising server-side tracking to provide real-time scaling insights and performance data for digital marketers. https://www.cometly.com/
- Recast: A self-serve Bayesian MMM tool designed for in-house teams to automate media measurement and conduct high-velocity scenario planning. https://getrecast.com/
- Lifesight: A holistic measurement suite that triangulates MMM, multi-touch attribution, and geo-incrementality testing for multi-market enterprise brands. https://www.lifesight.io/
- SegmentStream: An AI optimisation engine that identifies marginal ROAS per campaign and automatically rebalances budgets across platforms on a weekly cycle. https://segmentstream.com/
- Google Meridian: Google’s open-source Bayesian MMM framework that allows data science teams to build highly transparent, geo-level measurement models. https://developers.google.com/meridian
- Meta Robyn: An open-source, ML-powered MMM package from Meta that uses evolutionary algorithms to reduce human bias and automate model training. https://facebookexperimental.github.io/Robyn/
- Adobe Mix Modeler: An enterprise solution that unifies MMM and MTA on a single platform to deliver incremental lift measurement within the Adobe ecosystem. https://business.adobe.com/products/mix-modeler/adobe-mix-modeler.html
- Circana: A global leader in CPG analytics providing store-level, granular MMM and predictive forecasting using one of the world's largest household datasets. https://www.circana.com/
Incrementality (Lift Testing)
A slightly more practical way of analysing data beyond in-platform analytics, incrementality testing takes us from attribution (which touchpoint gets credit?) to causality (did the ad actually cause the sale?).
While standard attribution simply tracks whether a customer saw an ad before buying, incrementality answers the more vital question: "Would this conversion have happened anyway if the ad had never been shown? That’s the golden question.
In short, it helps us isolate specific and unique lifts (conversions) generated by digital advertising (Meta, Google, TikTok, etc.).
The basic core mechanism involves a Randomised Control Trial (RCT):
Test Group: A segment of the audience (or geographic region) that is exposed to the ad campaign as usual.
Holdout Group (Controlled): A statistically identical segment that is intentionally prevented from seeing the ad.
By comparing the conversion rates between these two groups, you (algorithms more so) can calculate the Incremental Lift, filtering out sales driven by organic demand, seasonality, or other marketing channels…at least in theory.
Incremental testing can benefit you by:
Exposing the Retargeting Illusion: Standard attribution often gives a high and optimistic ROAS to your retargeting campaigns. Incrementality testing will frequently reveal that these ads are likely "cannibalising" your organic traffic by targeting high-intent users who would have purchased anyway.
Identifying Saturation Points: These types of tests can help you build "Saturation Curves", which show exactly where the next pound spent yields a diminishing return (similar to how a Marketing Mix Modelling analysis would do). This prevents the common error of scaling a "high-performing" campaign into a zone where its Marginal ROAS (mROAS) falls below your baselines.
Closing off, in terms of implementation, as I’m a performance marketer/paid media specialist, I’ll only highlight approaches regarding Meta Ads and Google Ads, starting with:
Meta Ads (Facebook, Instagram, Threads, WhatsApp)
With Meta ads, incrementality testing is essentially auto-applied - in Ads Manager, you can view under Compare attribution settings, and toggle "Incremental Attribution". When enabled. Meta reported in early 2026 that this feature drove a 24% increase in incremental conversions compared to traditional last-click models.
Google Ads (Self-Serve Lift)
Incrementality testing in Google Ads is called a “Conversion Lift” study, which can be applied under the "Goals" section, then Measurement, then Conversion Lift. The concept is similar to Meta’s view of the metric “incremental conversions”; the only difference is that the output of data is exposed as a conclusive summary - the metric you get is the "Absolute Lift," representing conversions that would not have occurred without the said campaign/ad group
And we’re done. If you’ve found this write-up valuable, I simply request that you apply any learnings and share them with your network.
Author: Jonathan Milanes
Jonathan Milanes is the founder of Proverve (a premium end-to-end advertising consultancy), a multidisciplinary performance marketing coach and consultant and the author of the book Future-proof Marketing.

