top of page

The challenge of marketing attribution – where did you come from?

Updated: Sep 26, 2023

What is Marketing Attribution?

Marketing attribution is the process of determining how different marketing activities contribute to customer acquisition and retention. It plays a critical role in measuring the effectiveness of marketing efforts and helps businesses to optimise their budgets and strategies for maximum return on investment (ROI) and minimum waste.

Purchase journeys starting today are more likely than not to start online, whether they are B2C or B2B (the stats are from 2018 and I think it safe to assume that the %s would be even higher today). They are also incredibly complex, with Google claiming that the average purchase journey has between 20 and 500 touchpoints.

Regardless of the specific number of touchpoints, it is easy to understand how they can quickly accrue - customers interact with brands through a multitude of channels, such as social media, search engines, review sites, email campaigns and in person at stores or events. These interactions can occur across multiple devices and may span several days or even weeks. Furthermore, in complicated B2B purchases, multiple individuals within a purchaser’s organisation might be involved in the decision-making process over many months.

As a result, a Chief Marketing Officer must understand the cause-and-effect relationship between their marketing initiatives and customer behaviour. Without a consistent approach to attribution, you can find every channel coming up with their own ‘measures of success’, often provided by media owners – which understandably tend to overstate performance.

This is an important step in creating sustainable competitive advantage through adopting a ‘test and learn’ approach to customer acquisition, something I’ve talked about previously. It is also a key step in shifting from marketing operating as a cost centre to a profit centre, with the ability to measure and improve Return on Investment (ROI).

How to create a Marketing Attribution model?

The topic of attribution ‘modelling’ can provoke a host of reactions from marketers, often based on their own flavour of marketing – some think that attribution modelling systematically undervalues top-of-funnel brand marketing, others will obsess on the precise allocation of credit between different touchpoints to maximise the accuracy of their models.

My take is that the clue is in the name, an attribution model is just a model – a tool that should help you make better commercial decisions. It will almost certainly be wrong in some respects – you are trying to get as close as you can to replicating real-world customer behaviour, but it is unrealistic to expect to achieve 100% accuracy. My main suggestion to anyone spending time on attribution is to not let perfect be the enemy of good enough – this is something you should iterate over time.

One important qualifier before you invest meaningful time in analysing attribution – ask yourself whether you have a high enough volume of customers that you couldn’t just interview each customer to understand why they purchased (for example, if you only have 10 enterprise clients, this could be a better approach). Regardless of whether you go down the attribution modelling route, it is always interesting to ask your customers how they heard about you, but this isn’t a substitute for a robust attribution model.

I follow four steps to build an attribution model, and I’ll talk through each in turn:

  1. Building a base dataset for each unique prospect journey;

  2. Applying a model to ‘attribute’ credit between each touchpoint in a unique prospect journey;

  3. Start making decisions based on your model and test if it increases ROI & growth; and

  4. Supplement your model with additional measurement techniques if needed e.g. incrementality testing and/or marketing mix (econometric) modelling.

1. How to build an attribution dataset?

The cornerstone of your attribution model is to create a proprietary dataset of unique prospect journeys. Think of this as a data table where each row represents a touchpoint between a potential customer and your business, sorted in order from first to last. In the columns are a variety of details about each touchpoint for example a date-time stamp, the type of interaction, details of the device and browser for website visits, a unique identifier such as GCLID from Google Analytics, traffic source details and any flags related to your customer journey e.g., whether the individual logged into your website.

The types of touchpoints you can include at an individual user / prospect level will vary based on your current marketing activities and your business model, but might include:

  • Online interactions:

    • Website visits

    • Formfills and content downloads on platforms such as LinkedIn/Meta

    • Email opens

    • Mobile app downloads / sign-ins

    • Display ad views

  • Offline interactions:

    • Outbound calls attempted/completed

    • Inbound calls (for example using specialist call tracking software)

    • Booked sales calls & demos

    • Event attendance

    • Partner referrals

    • In-store purchases

Some types of touchpoints are very difficult or impossible to track at an individual user/prospect level – and you may be better to assess their impact through a different approach such as incrementality testing, which I’ll talk about later in this post.

I think it is incredibly valuable to build this dataset for yourself. Relying on third party tools can limit your ability to inspect the underlying data. Being able to explore this dataset for yourself is powerful – to prompt questions and to check that the results make sense. Relying on channel-level attribution exposes you to the challenge that media owners often overstate their own impact. Just try adding up the ‘conversions’ claimed by Google Ads, LinkedIn, and Meta for example – you will get a number that is often greater than your total actual conversions.

I’m not going to get too technical in how you go about doing this as the specifics will depend on your business model and existing data stack – I’m happy to discuss if you want to contact me. However, the default I follow is to start with exporting data from Google Analytics for ecommerce / B2C businesses and combining marketing automation and CRM data for B2B business. The tools you need to do this e.g., csv downloads vs APIs, excel vs SQL/Python, will depend on the size of your business and nature of your customer journey.

An important step for any advertiser is to be able to join marketing and customer journey data to sales data / conversions at a customer level. You will need to use a common key or unique identifier to do this (for example collecting Google Analytics Client ID – GCLID – at point of conversion into your own systems such as a CRM). This join allows you to understand the value of a specific conversion more precisely at a profit level (or even better customer lifetime value prediction).

Now some of your touchpoints will contain personal identifiers such as an email address or perhaps your own unique identifier that allow you to join them together. Others will effectively be ‘anonymous’ so you will need to look for other ways to combine them. The main approach I use is to use GCLID which for most online visits will be persistent for any given device over time. This is placed in a cookie by GA so has been somewhat impacted by changes to cookie consent, but provided your Google Analytics is configured server-side will survive the upcoming deprecation of third-party cookies.

This approach doesn’t help to join different devices that belong to a single user however – for this I will look for any combinations of GCLID with your own unique identifiers or personal information to ‘join’ two or more GCLIDs together. For example, I supported an ecommerce business to track GCLID from their email click-throughs and ‘my account’ visits, which had a high mix of mobile visits. By combining with their own unique identifiers, they could then join mobile and desktop sessions in the purchase journey.

For B2B journeys where there are potentially many individuals involved, I will often use email domain to group individuals from the same prospect. You might also be able allocate individuals to accounts within your marketing automation platform and/or CRM system.

When joining together touchpoints into journeys you will need to think about some rules such as – ‘how long should I allow between touchpoints before we are really looking at a new journey?’. The answer to this should be common sense based on your product/service and typical sales cycle – 28 days is too long for grocery delivery, too short for enterprise software and probably just right for high-value holidays.

2. How to define an attribution model?

Now you’ve got your list of interactions for each journey, you need to decide how to allocate the ‘value’ of each conversion against the contributing touchpoints. There are a few standard models that you will probably have heard of:

  1. First touch/click – all the credit is allocated to very first touchpoint in a journey, a proxy for how the customer heard about you in the first place. To me this normally offers the right balance of common sense and simplicity.

  2. Last touch/click – all the credit is allocated to the very last touchpoint in a journey. The default for several tracking tools such as Google Analytics. Tends to overstate the value of both navigational channels such as branded Paid Search and affiliate channels such as discount sites.

  3. Decay & divide evenly – less common but basically credit is divided across multiple touchpoints, with decay rewarding the most recent touchpoints and divide evenly doing exactly what the name suggests. I think these are arbitrary and have never used them.

  4. ‘Data driven’- a term you will hear a lot, as the new GA4 will default to ‘data-driven’ attribution. This will mean something different in every context but beware any black box where the rules or model principles are not explained. I would personally steer clear of this unless you are able to get a very clear understanding of what any ‘data-driven’ model is trying to do.

Which to choose? Well, the real answer is that you should test all of them to understand which most closely matches the true incremental impact of each aspect of your marketing activity. But that isn’t really a very practical approach!

My advice is to start by keeping it as simple as possible. I’ve seen lots of people agonise over this decision and often 80% right is good enough. My default is first touch/click – it makes the most sense to me and will often skew attribution in a logical way towards upper funnel and generic marketing activity. In a recent example of a model that I built in the travel sector, non-brand Paid Search saw 60% more profit allocated to it with a first touch/click model compared to a last touch/click model.

I should say that adopting a first touch/click model does not mean that the channels, content and creative that sit in the middle and the latter stages of the customer journey are not contributing – in fact they are normally critical to successful conversion. However – allocating significantly more of your budget to these activities is unlikely to attract brand new prospects to your business, notwithstanding the point I’ve made before about the role of continuous improvement throughout your customer journey.

The right answer will also differ for each business – when I worked in the super competitive travel industry, I adopted a custom data-driven model where I was able to understand the underlying principles – but for many of the SMEs I worked with as an investor, this would have been massive overkill.

3. How to use an attribution model?

So, you now have a model that gives you a view of marketing spend and ROI at a channel, campaign, and potentially keyword/creative level. The first step in using your new attribution model is introducing it anywhere you are making decisions about marketing spend – whether big picture strategic discussions around the board table, or granular bidding decisions in performance marketing channels such as Paid Search or LinkedIn. You should start to see whether it leads to (a) better quality discussions and (b) higher ROI and profitable growth.

A good place to start is to focus on anywhere that your attribution model suggests is losing you money: whether that is whole channels, campaigns, or individual keywords with very low ROI. One thing I always look at is search terms which haven’t generated any revenue for a while e.g., the last 3/6/12 months. You can initially be more confident when using the results of your model in respect of digital, click-based media – you may need to supplement with some of the additional analysis I explain below for offline and impression-based media.

You should adopt the same ‘test and learn’ approach with your attribution model as you would do with other aspects of your marketing and customer journey. Make some changes and monitor the results carefully. Did the results match what your attribution model suggested? Don’t be afraid to make changes if not.

Finally, it is imperative to communicate to your stakeholders. Many times, I’ve seen a marketing team start to measure performance in a new way and a finance team continue to use the old approach, as they don’t understand the reasons for the change and are not confident in its rigour. This is another one of the many reasons to keep things simple and ‘auditable’.

4. How to supplement your attribution model?

Depending on the nature of your marketing activity and your customer journey, you may be able to refine your attribution model with some additional data sources and analytical techniques, for example:

Incrementality testing: for marketing activities which you cannot measure directly, but which you would expect to generate an immediate response from your target audience (e.g., Paid Social, Direct Mail, TV/Radio in some cases). The approach involves defining target segments who will receive the advertising and comparing the outcomes of the following days/weeks against a control group who were not. These groups are often defined geographically (e.g., only in the South West, or only in certain postcodes), but you can really use any criteria where you are confident in achieving a robust control group and where your tracking will allow you to analyse the results (e.g., don’t segment on postcodes if your customer journey doesn’t collect postcodes). You need to make sure your target audience is exposed enough to the advertising so be sure to allocate sufficient budget and narrow your target audience if necessary.

Marketing mix modelling: an econometric modelling technique for larger advertisers and those who don’t ‘own’ the customer journey such as FMCG brands. This approach looks for correlations between marketing spend and growth in traffic/leads/conversions. This approach does not rely on understanding individual customer journeys so in my experience tends to produce an aggregated view of channel ROI rather than super granular results e.g., at keyword or creative level. One of the limitations to keep in mind with any attribution model is that it will skew towards short-term marketing activities which drive an immediate response – marketing mix modelling can offer a longer-term perspective. I’m planning to cover this in more depth in a future post, as it is something I’ve had less personal exposure to given the businesses I’ve worked with.

Hold-out area: this may sound like a simple one, but if possible, I always try to keep a geographic area exposed to as little paid advertising as possible – perhaps a city/county/state. Like the incrementality test, it is a helpful way to understand the impact of organic channels such as SEO and word of mouth referrals.

Primary research: asking your customers ‘how did you start the process of researching your purchase’ and ‘where did you first hear about us’ is a very worthwhile exercise. An attribution dataset can never capture the importance of word-of-mouth referrals and the reputation of your business. Don’t assume customers will have perfect recall, but qualitative interviews can be insightful alongside an attribution model. Across the many times I’ve run this type of research, for both B2B and B2C advertisers, word-of-mouth is invariably the single most common source of leads, accounting for up to 30% of purchases.

Some conclusions

In case you haven’t realised by now, building a robust attribution model is a never-ending process. There will always be ways you bring in additional datapoints, your marketing activities will be constantly changing, as will your customer journey. There are two principles that I try to keep in mind:

1. Attribution can never be perfect – it just must be good enough to make decisions each that will help you improve marketing efficiency. Given this, getting a first version of your model up and running then quickly iterating will create more value than spending months obsessing about a vastly complex data-driven attribution model, or joining every single touchpoint in a long enterprise sales journey together.

2. You are trying to understand human behaviour and your ability to influence it through data – so always ask yourself whether what you are seeing makes sense. All those journeys that start with a ‘Direct’ traffic source weren’t just people waking up one morning and spontaneously typing your website address into their browser. Make sure you can explain the human behaviour behind what your attribution data is telling you.

There are several tech providers out there who can help with some aspects of building an attribution model – I appreciate that the approach set out here may seem quite technical or require access to development resource that you don’t have. However, understanding the analytical process end-to-end is still important so that you can see both the strengths and limitations of any tech solution you may consider. In a future post I’ll share some views on the providers that I’m aware of.

Finally, if you need motivation to tackle your own attribution challenges, when we put in place an attribution model at CarTrawler we were able to reduce the Cost Per Acquisition in non-brand Paid Search by 60% at the same time as increasing volumes by 50% - as we could see which search terms and devices were driving the most valuable customers and use ROI to guide decision making for the first time.

If you’d like to discuss how you can build a view of marketing attribution in your business, please Contact Me.

All views expressed in this post are the author's own and should not be relied upon for any reason. Clearly.



Subscribe to get my ideas as they are published

Thanks for subscribing - I normally publish ideas at the start of each week.

bottom of page