Beyond Attribution

If you're a founder running a business that does a couple of million a year or more, and your marketing is mostly being measured through attribution, this piece is for you.

In this post, I'll set out why I believe attribution measurement is limiting the growth of your business, how MMM works, and how it can deliver such an enormous impact. The numbers cited throughout are industry standard and, if anything, on the conservative side.

Why we measure at all

Nobody is in the business of having perfect measurement. You certainly can't take good ROAS to the bank.

What people are trying to do is allocate capital better. If you can move marketing money around your business more effectively, you generate more sales. More sales unlock more investment. More investment grows the business. Measurement is upstream of all of that. It is an enabler, but not an end in itself. The true end point is the decision you make once you've measured.

Like plans, all measurement is wrong, but some of it is useful. This is a really important framing. The question isn't is this measurement accurate. The question is does this help me make a better decision.

What attribution does well

Attribution has a pretty bad rep. But this belies the fact it's the cornerstone of thousands of multi-million-dollar businesses. These businesses make money and grow, and would no doubt contest, quite rightly, that attribution is a key part of that.

There are many things attribution is very good at. It will tell you which keyword drove a checkout in the last hour. If you want to optimise a campaign in flight, attribution would be a good place to start. If you sell low-cost, high-volume products with a single sales and marketing channel (and you're happy staying this way), attribution will do just fine.

The issue is that attribution is used above and beyond this very narrow set of short-term and limited use cases. Because it's tangible and immediate, it gets elevated beyond them. It's easy to understand (money in, money out) so even the CFO gets it. They may even like it.

But there are some very real limitations to this type of measurement.

8 limitations of attribution

1. Last-click is fiction.Let's start with the low-hanging fruit. Last-click attribution, where all credit for a sale is given to the ad clicked immediately before the sale, is dangling the lowest. The issue with this methodology is that it has very little basis in reality. Nobody (or extremely few) sees an ad once and immediately buys. What actually happens is a consumer is influenced by many different ads before they purchase (11 on average across 4 different platforms, actually) before they convert.

2. Multi-touch isn't much better.OK so LCA is BS, moving on to number two. What about multi-touch attribution? This methodology spreads the credit across all tracked ads along the path to purchase. Seems like a sensible approach. But, leaving aside the fact that there are about 20 different ways of doing this, there is a more immediate issue. Most users now opt out of being tracked that way (Apple's ATT framework alone takes 60%+ of iOS users out of the picture). So the tracking can only see a minority of users. You're measuring an iceberg from the bit sticking out of the sea.

3. It only sees clicks.The tracking only tracks ads that get clicked. What about ads that do not involve clicks? Many of these are not à la mode (TV, OOH) but also include podcast ads, Spotify ads, and influencer campaigns, none of which can be counted via attribution. (A further disclaimer: I am a classically trained marketer; I will argue till my final breath that the funnel is still as relevant as ever. This is backed up by the data; a buyer is 2.8× more likely to click an ad from a brand they know. Brand-building is important, and many of the best channels for doing this are non-clickable.)

4. It can't see the rest of the world.There's a long list of things that influence sales beyond marketing (you can see where this is going by now), and attribution can't track them. British people love talking about the weather. It's also one of the best indicators of buying behaviour — a 1% shift in UK temperature moves retail sales by about 1%. You can expect to sell more beer when the World Cup is on. On top of this there are things like price, distribution, competitor behaviour, press, stock levels: real-life things that impact your sales completely above and beyond marketing.

5. It counts demand that was already coming.Chances are there's a portion of people who were going to buy your product whether you advertised or not. Repeat customers, organic discoverers, renewals, etc. They may click on the ad but they were going to buy anyway. Attribution makes no distinction. The latest figures show that only 40% of attribution-measured conversions are incremental.

6. The plumbing is fragile.Setting it up and keeping it running is a royal pain in the never-you-minds. Pixels break, cookies expire, tags misfire, marketers perspire and eventually expire. There is literally a whole industry dedicated to auditing and fixing broken tags, and yet around 20% of tracking tags are firing incorrectly at any one time.

7. Platforms mark their own homework.You're asking the platforms, not universally admired for their honest and objective business practices, to mark their own homework. Oh and obviously, they'll only mark their own homework. So those 11 touchpoints over 4 platforms, that's 4 separate reports with 2.75 touchpoints in each. It is not a full picture.

8. It's correlation, not causation.Attribution is just counting. It's an observation. This person saw this ad, and this person also bought your product. It counts the number of times that happened. That should not be confused with that person saw your ad so they bought the product, merely that these two events both occurred. The analytics term for this is correlation. The problem is that past performance is not a good indicator of future performance. So just because March's number was X, it does not mean April's number will be X, nor that next March's data will be X. It just means that March's number was X. Going back to the point I made at the beginning, useful measurement is a means of enabling effective decision making. On this basis, attribution is not good measurement.

Enter MMM

Marketing Mix Modelling was developed in the late sixties and seventies. Like low-rise jeans, it has been in and out of fashion ever since. It enjoyed a big boom after iOS killed tracking, which has quietened somewhat, but interest is still up around 300% in the last three years. The fact is, every single one of the fastest-growing, most valuable brands in the world uses MMM in some capacity.

Data

MMM uses existing data, such as marketing (spend, performance, etc.) in a weekly format, of which it needs two years' worth. You can also layer in other factors like temperature, press, Google Trends, stock shortages and so on. These are your input variables. Then you layer in the thing you want to measure (the output variable). Most commonly this is revenue, but it could be app downloads, video views, whatever your business runs on. Most importantly, MMM does not rely on ad tech, pixels or tracking. It uses existing data that is already within your business.

Model

MMM is a statistical model — regression analysis, to be precise. It is built using layers of data. Two years of weekly data equals 104 (52 × 2) observations, meaning there are 104 times we can see your business data at different levels. How many levels? Well, if you have 15 input variables and 1 output variable, that is 104 × 16 = 1,664. These are the data points. These data points also create unique pairs, the total number of combinations that all the variables can be at. In the example above, this amounts to 1.4m. It is from these that we can start to identify patterns; they are the building blocks of the methodology.

Mechanics

If you plotted one of these pairs on a chart, you would get a messy scatter diagram. It would be possible to draw a straight line that was closest to the most number of those dots. This is called a coefficient. The coefficient is the relationship between that pair. We are not trying to find the relationship between one pair though, we're trying to find the relationship between 1.4m pairs. This is what the model does. It finds the coefficients that explain the overall performance in the fullest possible way. It does this by first trying to establish what the baseline figure is, and then layering on coefficients that correctly predict what the actual sales were. This is rarely perfect, but a good MMM can work backwards from the coefficients and baseline to predict sales performance within 5–10% error. If it can't, there is something else driving sales that is not captured within your variables.

Once you have the master explanation, you can then take it apart to isolate the individual impact of the variables above and beyond the baseline. What's really cool is you can also see what the optimum level of investment should be, whether you are under- or over-invested, meaning your ROI is being compressed because of audience availability.

This last exercise is typically the most valuable. Most brands are materially underinvested in marketing. They could increase their budgets profitably (and therefore grow their businesses and make more money) if they just invested more.

What it doesn't do

What I'm not trying to do is present MMM as some fix-all silver bullet. It has its limitations. It won't tell you which creative is performing best, which keyword to bid on, or what audience to target. It can feel quite intangible because of this. It's hard to explain, it can be hard to implement, and the cost and speed traditionally associated with it are inhibitive (but more on that later). It is not useful for tactical questions about campaigns and assets. It models recent past to inform near future and it requires refreshing, annually as a minimum. It is overkill below a certain spend, and useless if you spend the exact same amount on the same channels week in, week out.

Like any measurement, it is only as valuable as the actions that are taken from it. Bad practitioners tend toward the kitchen-sink approach, overwhelming brands with statistical jargon and biblical-length slide decks, equating this with value. The impact is constrained by many human and structural factors, such as leadership buy-in, organisational agility and open-mindedness to new approaches. The absence of these things is where most engagements fall down.

What the upside looks like

But enough of that doom and gloom, let's consider what happens if it goes really, really well. Saying that, the example below is a typical case based on the lower end of industry averages. I have seen these impacts many times but they are not my numbers; this is the lower average of what MMM could do for your business.

Take a brand spending £3m a year on marketing at a 3:1 ROI. So £9m of return as it stands, who deploy MMM.

The first thing MMM offers is the chance to reallocate budget from low-ROI channels to high-ROI channels. Nearly all brands find at least one channel they should pull money out of and at least one they should put money into. Doing only that, and changing nothing else, typically lifts ROI by about 20%. On £3m of spend that's an extra £1.8m of revenue. The work has paid for itself many times over before you spend an extra pound. Hurray!

The second lever is scaling. Once you can see where you're under-invested and at what point each channel saturates, you can grow spend into those channels with confidence that the marginal return justifies it. A reasonable medium-term outcome is increasing spend by around 25%, from £3m to £3.75m, without diluting ROI. That delivers another £2.7m in revenue. Oof.

The third lever, which is where MMM gets genuinely interesting, is unlocking channels you previously couldn't justify. Top-funnel and brand-building work becomes investable, because for the first time you have a credible read on what it returns. Over a three-year window this could reasonably add around £2m of incremental revenue. Cha-ching!

Stack the three. The brand goes from £3m of spend returning £9m, to £3.75m of spend returning around £15.5m. That's roughly 73% more revenue from your marketing, and a 38% lift in ROI, on a budget that's grown by a quarter. The incremental £6.5m wasn't sitting in better creative or smarter targeting. It was sitting in a smarter allocation question that attribution couldn't answer.

They work together

I want to be clear that I'm not saying attribution is bad and MMM is good. They do different jobs and most mature businesses end up using both.

Attribution is the right tool for short-term, tactical, channel-level decisions. Which audience inside Meta. Which keyword cluster inside Google. Which creative variant. The questions you're asking weekly. MMM is the right tool for the strategic question of how much should sit in which channel in the first place, and the questions you're asking quarterly or annually.

There's also a useful feedback loop between them. Better attribution discipline at the channel level — cleaner tagging, smarter event setup, more honest measurement of what each channel is doing — feeds cleaner inputs into the MMM, which gives you a sharper read on the strategic question. The two get better together.

Where do I fit into all this?

Here's the pitch. I run Hanya: AI-Accelerated MMM. Traditional MMM costs between £150–300k and takes 6 to 12 months. We have been able to reduce this dramatically by developing a proprietary AI data layer that increases speed and decreases complexity by about 80%. Not only that, we derisk MMM for brands by saying: if we don't find at least a 20% improvement in 12 weeks, we will refund you the full price of the engagement (the 20:12 Guarantee).

What is that price?

• Partnership Plan — £150k a year, including unlimited refreshes within a 12-month period.
• Flexi Plan — one-off, £80k for the first build (£50k onboarding + £30k insight generation).
• Early Adopter Rate — £25k all-in for our first two clients, onboarding free. We'll refund the full £25k if you let us use your anonymised results as a case study.

Why do all this? The 20:12 Guarantee is there because we are confident in our solution and the value proposition. In the unlikely event that your marketing is already perfect, then you are not our target customer, and there is no opportunity for either of us in a long-term partnership; frankly your goodwill is worth more to us than your money.

The case-study buy-back exists because peer recommendations are worth more than any ad campaign. We'd rather reinvest into our customers' businesses than pay for ads.