?> Architecting Marketing for AI | Snowplow
Snowplow Outperforms GA4 in GigaOm Comparative Evaluation.
Read More
  1. Home
  2. Blog
  3. Data Insights
Data insights

Architecting Marketing for AI

Architecting Marketing for AI
Jump to

I am delighted to welcome Tom Affinito, Head of Cloud Marketing at LiveRamp, to the Snowplow blog to share some emerging thinking around what we are calling ‘Architecting Marketing for AI’. Tom has been a great collaborator as we have been exploring how marketing (and martech!) needs to adapt in the face of generation-defining developments in AI.

We want to introduce seven guiding principles for organizations that have recognized the need to align their marketing efforts with AI. These principles are the result of our learned experiences at Snowplow and LiveRamp, where we have helped marketing, data and analytics teams accelerate towards AI and machine learning.

Image courtesy of Dall-E

Unfortunately, the “now comes with AI!” claim that almost all marketing software vendors make, doesn’t work in practice. Aligning your marketing organization with AI requires a significant rethink of the way that data is collected, segments are built, creatives are designed, campaigns are run and performance is measured. Above all, it means direct lines of communication with your AI and machine learning experts, whether in-house or in an agency.

Organizations are now faced with a real ‘sliding doors’ moment for their marketing strategy. Door one: the more qualitative, story-oriented marketers sticking with ‘Fonts & Colors’ marketing apps and bundled AI features. And door two: the more quantitative, tech-forward marketers embracing ‘AI Optimization’ and re-architecting their people, processes and products for AI and machine learning. Door one is easy to open but has diminishing returns over time. Door two, on the other hand, is the gateway to sustainable growth for your organization.

As organizations self-select for ‘AI Optimization’, we see them trying to navigate this complex landscape, asking questions like:

  • Is the cookie dead? What does that mean for identity?
  • How can I experiment with AI without jeopardizing my brand?
  • What is a customer data platform (CDP)? Can it deliver my AI strategy?
  • How can I work more effectively with our data science team?
  • What role can third-party data play in a GDPR/CCPA world?

These questions contain important differences of opinion and pivot points for our industry. But organizations need answers now. CMOs have already sold their Boards on an ‘AI everywhere’ strategy and are expected to be heads-down executing on this strategy in 2024.

What follows in this post, then, is a set of seven guiding principles. These are based on both mine and Tom’s experiences – but I should stress that they don’t represent the views of LiveRamp or Snowplow. Nonetheless, they are based on a clear worldview of how AI, customer data and marketing should all interlink. We hope you find them useful.

Without further ado, here are our guiding principles:

  1. AI assistants aren’t enough
  2. Embrace first-party data
  3. Turn your customer data warehouse into your data science workbench
  4. Choose technologies that inter-operate
  5. Give your data scientists a seat at the table
  6. Own your own customer identity strategy
  7. Adopt a “crawl, walk, run” approach

The rest of this post will introduce and detail each of the principles. Let’s get started:

AI assistants aren’t enough

The genie is out of the bottle – AI assistants, chatbots and copilots are now embedded in marketing teams’ daily workflows, as the latest incarnation of ‘shadow IT’.”

TOM AFFINITO | HEAD OF CLOUD MARKETING, LIVERAMP

2023 was the year in which marketers embraced AI assistants – whether they were asked to by their managers or not. There is now a dizzying array of AI productivity tools, and marketing teams are using them for everything from campaign brief creation to copywriting to video production. In many cases, marketers have even managed to ‘teach’ OpenAI and other large language models (LLMs) about their brand, by uploading their campaign briefs, brand docs and similar, using techniques such as mega-prompting and context stuffing.

AI assistants like ChatGPT hint at the potential for AI for marketing, but they are not nearly enough for a marketing team that wants to architect itself for AI. AI assistants like ChatGPT are generic, being trained on the open internet, and operating in a prosumer or ‘desktop’ paradigm. Out of the box, these LLMs don’t understand your brand voice, they aren’t trained on your personas and they don’t understand what ad campaigns, copy and creative have worked for you in the past. Workarounds like context stuffing are time-consuming, error-prone and incomplete, as this post from an AI startup explains.

The solution is to evolve beyond AI assistants and copilots and embrace AI that is trained off your own first-party and proprietary data, living in your own cloud. This isn’t OpenAI’s AI any more – it’s your company’s AI, powered by your company’s proprietary data and derived insights. The six remaining guiding principles are all in turn informed by this framing of data and AI ownership. Let’s continue.

Embrace first-party data

First-party data acts as the baseline for getting to know your customers because it comes from genuine engagement with your company across a wide variety of consumer touchpoints. When you connect the constellation of touchpoints per customer and multiply that by millions, you can easily see how powerful first-party data can become for your business as a whole.”

LIVERAMP | FIRST-PARTY DATA: WHAT IT IS, HOW TO USE IT, AND WHY IT MATTERS NOW MORE THAN EVER

At Snowplow, since our founding in 2012, we have focused entirely on helping retailers, publishers and other brands to generate first-party behavioral data from their own customers. First-party data is collected with the permission of customers and is extremely rich because it is generated by directly observing customer behavior on a brand’s own digital estate.

In contrast, second-party and third-party data is typically demographic or transactional data that is purchased or otherwise sourced from intermediaries, brokers and clearing houses.

Why is first-party data so valuable for AI? There are three main reasons. Firstly, it records not just the decisions that people make, but how they make those decisions, and the environmental factors that were true when they made those decisions. This makes it incredibly predictive. Secondly, it records what people are doing now – which as a marketer is super important because how well a user responds to a message, offer or outreach is very dependent on what they are doing at this particular moment in time. And lastly, first-party data provides a huge amount of context for brands – behavior at the airport check-in desk will be different from behavior inside the airline’s mobile app, and both will be different from a search on the travel aggregator website.

Imagine a car dealership with a digital platform for the whole customer journey, including test drives, car configurator, leasing and financing. In this example, behavioral data from the car configurator can help a dealer identify what aspects of the car specification are important for the buyer. If a buyer is engaging with content related to financing, or booking a test drive, that would suggest that they are very serious about making a purchase. And if a dealer wants to make an offer to the prospect, doing so whilst she is engaging with the car configurator (so is thinking hard about the purchase) is likely to be far more effective than sending an offer when she’s doing something else.

This principle of embracing first-party data is not meant to detract from second- and third-party data, which can play an important role in AI-centric marketing. Getting the Socioeconomic Score (SES) for a buyer’s zipcode (third-party data) is helpful and predictive. But without a clear first-party data strategy, especially one incorporating digital behavioral data, your data scientists will be challenged for breadth and depth of training data.

Turn your customer data warehouse into your data science workbench

You can’t have missed the largest data-related trend outside of AI: organizations are rapidly building a ‘central brain’ – or at the very least a central source of truth – in the public cloud, leveraging cloud data warehouses and data lakes. This is disrupting almost all of the incumbent marketing software vendors.”

ALEX DEAN | CEO & CO-FOUNDER, SNOWPLOW

With the move to the public cloud, organizations have brought their customer data workloads with them and are taking advantage of the cost-effective, highly-scalable technology from Databricks, Snowflake, Google and others.

The organizational goal is to use this customer data – including Snowplow’s first-party customer behavioral data – to develop and exploit their own proprietary intelligence. Storing this data in a centralized data warehouse or data lake is the first step. From there, this customer data is unified, modeled, analyzed and activated.

For marketers, this centralization helps solve the ‘split brain’ problem of having multiple different marketing systems, each with their own data store, each containing incomplete, inaccurate and/or outdated sets of customer data. Rather than trying to maintain a spaghetti nest of products and integrations, marketers can bring their analytics and activation workflows to the data, working off a rich, fresh and unified Customer 360.

This central source of truth is now evolving into a ‘central brain’ that enables the new data science workloads. Machine learning toolkits like Databricks, Snowflake’s Snowpark, Google Cloud’s Vertex AI, Azure AI and Amazon SageMaker are providing a ‘data science workbench’ for your technologists inside the customer data warehouse or lake itself. Rather than having to work on data extracts, your data scientists can now bring their algorithms and expertise to the Customer 360 itself – hugely empowering for your AI Optimizers!

Choose technologies that interoperate

In the age of data privacy and security, composability is synonymous with data sovereignty and autonomy. It allows you and your teams to own and manage the data the way you want. It starts with data creation and extends throughout your data stack to modeling, activation and analytics. It will not only make your engineers happy, it will make your marketers happy as well. Most importantly, it will make your customers happy with more relevant experiences.”

JONATHAN MENDEZ | FALLING IN LOVE: CTOs AND THE COMPOSABLE CDP

As marketers, we are used to selecting products and platforms based on the core capabilities of the technology – ‘how can it help us on the job-to-be-done?’. But in the ‘AI Optimization’ paradigm, we need to focus on how well these technologies can interoperate with each other

Marketing apps with bundled AI features will not enable your data scientists to do their day jobs. Your experts need to be able to use their own AI and machine learning tools and then integrate their segments, decisions and scores into your wider marketing estate.

Work with your IT and engineering teams to re-orientate your technology stack and product choices towards composability and interoperability. They can interview current or prospective vendors to understand:

  • Does the product have well-documented integration points, such as APIs or native access to the cloud data warehouse?
  • Are there published examples or case studies of the product being integrated with other relevant solutions?
  • Does your data science team believe they can integrate this tool into their workflows?
  • Is your product able to easily import data and intelligence from your ‘central brain’ (the data warehouse or data lake) to activate?
  • Is your product transparent about what decisions (and therefore intelligence) is developed in it – and it is straightforward to export a record of those decisions and intelligence to your central source of truth so that it can be audited and built on?

AI is disrupting traditional categorizations and market positions for software. Work with your technology peers to take a more fluid and adaptable perspective that prioritizes interoperability over individual features or market share. Until things stabilize in our new AI Optimization normal and winners again emerge, composability will be king!

Give your data scientists a seat at the table

The AI strategy of your company is not the AI strategy of your martech vendor. The right question to ask a vendor is not, ‘what cool new AI-powered features do you have on your product roadmap?’ The right question is: ‘How are you going to open up your platform so that my data scientists and MLOps engineers can be productive?'”

ALEX DEAN | CEO & CO-FOUNDER, SNOWPLOW

Maximizing outcomes from AI is now a Board-level imperative. Whether real or imagined, companies see themselves in an ‘arms race’ to exploit AI and machine learning faster and more effectively than their competitors. In a difficult economic environment, investments in AI (from hiring data scientists to renting GPUs) are being waved through.

There is nothing wrong with your CDP vendor offering, say, experimental generative AI features to help with 1:1 personalization – give these features a spin, and hopefully they’ll increase conversions. But packaged ‘black box’ AI features are not going to empower your own machine learning experts – they are not designed for these technical personas.

Your data scientists, MLOps engineers and others have their own experiences, tools and methodologies. You need to include them in the conversation about your marketing challenges and let them get creative on how they can apply their unique skills and workflows to your problem.

Own your own customer identity strategy

Some marketers have come to a kind of imperfect acceptance in the area of identity, balancing convenience against more durable solutions. But AI upends all of that: you can’t model a customer and predict that customer’s future behavior without a deep understanding of who that customer is, and that starts with a strong identity foundation that can consolidate customer touchpoints accurately. That’s the key for modeling propensities, affinities and interests far more accurately than our data science could do in the past.”

TOM AFFINITO | HEAD OF CLOUD MARKETING, LIVERAMP

Data modeling, feature engineering, propensity scoring – in a marketing context, all of this essential machine learning work is centered around the notion of a customer. But who is this customer, and how do you recognize them? Is the customer:

  • Known – they have been personally identified, with at least some personally identifiable information (PII) associated with them. This may also include anonymous identifiers 
  • Unknown – they are being tracked against anonymous identifiers (or at least non-PII) and have not yet been personally identified

And how does customer consent (from the basis of collection through to the right of deletion) overlay into this?

Your customer identity strategy doesn’t have to be particularly complicated (see the “crawl, walk, run” principle below), but it does need to be deliberately designed and explicit. Any data engineer, data scientist or business user working with your customer data, or BI or AI outputs from it, must be able to ask: “But how exactly do we define a customer?”.

Adopt a “crawl, walk, run” approach

In the ever-evolving landscape of Artificial Intelligence (AI), businesses are keen to integrate this technology to enhance efficiency, innovate, and gain a competitive edge. However, the journey towards successful AI implementation in the workplace is often complex and challenging. This is where the “Crawl, Walk, Run” approach becomes crucial. It’s a methodical strategy that ensures a gradual, well-managed, and sustainable integration of AI into business processes.”

MICHAEL L. WOODSON | THE VITAL STEP-BY-STEP APPROACH: “CRAWL, WALK, RUN” IN AI ADOPTION AT THE WORKPLACE

Introducing AI into your marketing is a major endeavor and one fraught with execution risk. You can reduce this risk by opting for a gradual rollout through a “crawl, walk, run” approach, involving:

  1. Crawl: Build your data foundation, identify your initial use cases and run small-scale pilots
  2. Walk: Put your pilots into production as ML apps, integrate your AI into your existing systems and upskill your workforce
  3. Run: Broaden out your estate of ML use cases, create thought-leadership around bleeding edge applications and attract and develop top talent

The other paradigm we like to use to accelerate and de-risk projects like this is the concept of the ‘vertical slice’, which we have adopted from the games development industry. In this approach, just enough of each ‘slice’ or layer of the problem is built to achieve the final deliverable. This gets you to a demonstrable outcome faster and, most importantly, it spares you from over-investing in a specific part of the solution before you really understand the problem domain.

Epilog

The waves of AI innovation are crashing over all areas of modern business with astonishing speed – and marketing is not spared. Those organizations that successfully harness these technologies to grow, retain and upsell their customer base will outperform for the next decade or more.

The six guiding principles in this article are an attempt by Tom and myself to move beyond the topical martech debates and give marketing teams actionable guidance now. We hope you find these principles valuable.

If there is a common theme running through all of this guidance, it’s one of communication: aligning your marketing with AI requires close collaboration between marketers, IT and data scientists. Find your ‘AI translator’ – a person or team that can speak multiple ‘languages’ and can interpret across departmental or functional boundaries in your organization. Leverage these people to bring your first small-scale AI projects into production. Good luck!

This article is adapted from Alex Dean’s Substack Source of Truth: A blog that charts the ‘Great Reset’ that is coming in the adtech-martech solution space driven by data warehousing, first-party data and AI.

More about
the author

Alex Dean
Alex Dean

Alex is the Co-founder and CEO of Snowplow.

View author

Ready to start creating rich, first-party data?

Image of the Snowplow app UI