For all the talk of artificial intelligence, there is big gap between the widespread excitement about AI, and genuine understanding of how it will become practically useful for marketers.

Granted: AI features are widespread in modern marketing tools. LinkedIn now offers to rewrite your posts with AI. Other individual pieces of tech for customer profiling, analytics, ad targeting, content management and creative production, etc. come with AI features which enhance the effectiveness of the software.

These, however, are piecemeal applications.

Truly exciting developments that transform business performance, on measures of efficiency, resilience, profitability and customer loyalty, would include things like…

• personalised brand experiences that customers genuinely value – rather than perceiving personalisation efforts as creepy, intrusive or irrelevant
• the ability to manage dynamic offers and pricing in a coherent way across all brand touchpoints, including in third-party settings such social commerce and other marketplaces
• predictive analytics that allow marketing strategy and campaigning to respond, in real-time, to factors as wide-ranging as the economy, the climate and geopolitical events.

Improvements of this type go beyond individual pieces of software, to the combination of insights from across the business, and the many pieces of technology it uses.

Brands already carry out such data processing for the purposes of analytics and reporting, but the utility of these efforts is limited by how quickly and efficiently data moves around the martech stack.

The expansion of AI-led marketing, there, ultimately hinges on how quickly brands can transform their management of data.

In this article we will:

• look at how the poor management of data currently leads to unintelligent marketing – and how many brands do not yet have sufficient command of their data to evolve beyond the status quo
• investigate how composable commerce architecture will pave the way to the widespread use of marketing AI – since modern martech solutions are specifically designed for their data to be easily shared with other tools
• explain how the modern data stack will be crucial for AI-led marketing at scale, since brands’ legacy data architectures currently restrict innovation with AI.

Digital transformation does not play out equally from one business to the next; some brands are 10+ years ahead of others in terms of the sophistication of their technology and data. This gap will only widen as AI enables the more tech-savvy brands to extend their lead against competitors.

There is a clear-cut advantage, therefore, for marketers who understand the limitations of their existing setups, and the technology buying decisions needed to capitalise on emerging technology.

Unintelligent marketing, and the data problem holding back AI

Brands have a chequered history of data-led marketing innovation. There have been significant efficiency-savings and improvements in ROI – but also, a lot of poor customer experiences, and occasional PR disasters.

This example from Target, an American retailer, is very well-known, but is an excellent example of the risks of how technology can make poor marketing decisions when relying on insufficient data.

By 2013, the brand had all the data to accurately show that a customer was pregnant. It lacked, however, the intelligence layer to determine whether marketing, based on that data, was appropriate for individual customers. This led to an unfortunate interaction between a father and his teenage daughter, who hadn’t yet broken the news to her parents.

More broadly speaking, customers have been shown to be resistant to unintelligent personalization and automation efforts.

63% of the 1,000 UK consumers surveyed by CM.com in May 2023, a customer service vendor, said they did not consider purchase history to be ‘acceptable data for brands to draw on when providing a personalised customer experience’.

Of course, ‘purchase history’ implies first-party data that the brand has legitimately collected and processed. So why is this regarded as unacceptable?

The problem, I would argue, is that such efforts are based on overly limited datasets, which is undermining customers’ perceptions of automated/personalised marketing.

Compare the idea of a friend buying you a gift that you already have, versus one that they correctly predicted you’d love. Predicting what someone will love depends on having an incredibly detailed understanding of the individual – an understanding made up of many different data points.

By contrast, many automation efforts backfire, because – as in the Target example above – they are based on a relatively small number of data points: a single purchase, or even just a couple of clicks on a brand’s website.

Improving on this state affairs implies taking a number of steps:

• connecting datasets from across the business – rather than only 2 or 3 closely connected martech modules – to greatly increase the precision and quality of data insights
• making that data actionable by processing it at scale. AI will be of great value here, because it implies a volume of processing that exceeds conventional technologies
• feeding those insights back into executional martech, including tools for campaign management, creative production and content management.

All these steps are currently limited by the brands’ technical architectures, which restrict the way that data moves around the business, and how effectively that data can be processed.

Before you can consider pathways to AI adoption, therefore, you need to understand the need for digital transformation in brands’ martech stacks.

Unlocking marketing data through composable architecture

It is not the purpose of this article to explain the concept of composable commerce architecture as this has been extensively covered elsewhere (including in our recent ebook, ‘Major marketing trends hinge on composable architecture’).

But it is necessary to understand how the nature of brands’ technology affects their ability to manage data, since effective data management is the foundation of AI.

Composable architecture is an approach to assembling technology stacks out of multiple different pieces of software, from different vendors, who designed their software for this purpose.

By contrast, legacy technology vendors built expansive software platforms (known as ‘monoliths’) to capture as much of a business’s technology budget as possible.

Bundled in with the monolith is a central data stack. This allows a professional to query the database in order to extract reports from across whichever parts of the business uses part of the monolithic software platform. Those reports, known as ‘business intelligence’ (BI), are then used to inform decision-making.

Of course, no single monolith ever accounted for the whole of a company’s technology needs – so over time, businesses have integrated other tools with the monolith. Depending on the nature of that integration, the business should then have been able to include those new modules in its data analytics, and combine data insights across old and new modules.

Cygnet Digital’s Peter Ross explains…

“To take a hypothetical scenario: say you wanted to analyse the effect of your email marketing on in-store transactions.

These are two entirely different systems, but at some point over the years, the company has probably found a need to manually integrate its POS software with its monolith – so all the data is exposed in the centralised data lake. That would enable you to run your queries without any additional technical development.

The data processing may be slow, and you might need a degree in data science to extract the intelligence – but ultimately the centralised database gets you the business intelligence you need.”

Though effective for the purpose of periodical reporting, however, this approach is not suitable for AI processing of the type that’s needed for most brands’ marketing ambitions.

This is due to the slow pace of data processing in a monolithic data stack, and the rapid adoption of new, composable martech modules by brands.

1. Legacy data stacks are too slow for real-time AI processing

One of the core use-cases for AI in marketing is the automation of real-time decision-making, so that hundreds of thousands of individual marketing decisions can be made per day without human oversight.

To return to the Target example above: there is a huge volume of data which might inform the wisdom or likely ROI of sending certain pregnancy promotions to any individual customer.

This data is spread across a very wide range of different software modules. It might include:

• the customer’s age or income bracket (stored in your CRM or CDP [customer data platform])
• the supply of your pregnancy/babywear range, and how that supply is forecast to change (inventory management & demand planning systems)
• interactions with customer service or on social media – since if the customer has recently had an argument with a customer service rep, you might prefer to switch off all marketing for a month or two – or issue more attractive offers to regain their trust
• feedback loops from your CMS, so that specific customer actions on your website or your app can inform the timing or content of messaging
• your pricing and promotions engines – which, in combination with the CDP, might help to calculate the likely ROI of sending that customer different types of offers, and tailor the offer amount accordingly.

AI has the potential to combine these data insights in real-time in order to determine what offers are issued and when – and to optimise performance throughout the campaign by machine learning.

Real-time AI processing, across such a wide range of modules, simply wouldn’t be possible with total reliance on a traditional, monolithic data stack.

Even at Shopify, a relatively modern enterprise, their legacy BI solution left analysts, “starting reports and actually walking away from their desks to let them finish.”

Denis Zgonjanin, then of Shopify, said:

“Because it took a while to load, our reporting got in the way of hundreds of people trying to do their jobs.”

Clearly, technology which is too slow for manual business analytics is going to be of limited use for real-time marketing automation.

Unfortunately, replacing this technology is easier said than done.

2. Brands’ pace of technology adoption is still relatively slow

As mentioned above, brands have gradually customised their legacy monolithic platforms over the past 15-20 years by carrying out hardcoded integrations with additional tools.

Commonplace integrations of this type include the addition of modern CRMs and automation platforms, and more recently, modern content management systems.

Many of these individual integrations would have been costly and time-consuming. A 6-month timeframe and a bill of $80,000 would not be uncommon for an integration between a modern pieces of software and the legacy ERP.

This has been partly alleviated by the age of composable commerce, and the arrival of modern alternatives to nearly every martech module. Examples include new tools for data, payments, checkout, order management, supply chain, product information, search, and new marketing touchpoints such as mobile apps, marketplace integration, social networks, POS, etc.

This is valuable for the utility of AI, in that modern marketing solutions are ‘API-first’ in most cases. This means that they are designed in anticipation of the need to share data via APIs, including with the brands’ database(s) and analytics tools – which makes them highly useful for AI processing.

Indeed, AI will be increasingly needed to handle the necessary level of processing across this growing quantity of data, from an ever-greater number of SaaS tools.

The trouble is that, despite the best efforts of modern solution vendors, there remain significant restraints on how quickly many enterprises can implement new pieces of composable martech.

Though these modern tools are designed to be quick to deploy, and though many popular solutions offer out-of-the-box integrations, these integrations still typically need to be customised at individual brands. Furthermore, new martech solutions are coming to market at such pace that, in practice, integrations still occupy of a lot of development time.

This is leaving many enterprise brands exposed to competition from more modern businesses, who have built their businesses around a modern data stack – and so are much better placed to capitalise on AI.

How the modern data stack unlocks AI marketing

There are essentially two ways to achieve a technical architecture suitable for the widespread application of AI to a brand’s marketing setup.

One way is to build a modern technical architecture from scratch. Established enterprises cannot do this, as the scale of the development challenge is prohibitive – but analysing the startup brands that have done this is useful for understanding the competitive standard in modern technical architecture.

The second way is to migrate towards the modern data stack. This will play out over years at most enterprises– but along this path, brands will acquire various modern data tools which will unlock significant ROI as they are deployed.

The key point is that either of these approaches makes it easier for data from across the business to be exposed for AI processing, at sufficient speed for AI-powered marketing at scale.

1. Building a modern data architecture from scratch

Wayfair, an online furniture retailer, claims to enjoy 30% larger cart sizes, thanks to personalisation achieved through the a highly evolved data setup.

This is not actually a case study of the use of AI in marketing, but it’s extremely helpful in showing the level of marketing sophistication that can be achieved by getting your data in order.

This presentation on Wayfair’s data stack is worth a watch. It’s intended for technical audiences, but from around 7:30 to 9:00, it clearly shows the sheer number of different pieces of software that exchange data with the core database in order to personalise customer experiences.

The key component of this stack is the technology used to handle customer identity. In the case of Wayfair, this solution is Aerospike, a purpose-agnostic data platform. In Wayfair’s stack, Aerospike processes various data sources in order to personalise customer experiences as they’re taking place.

Aerospike draws on a customer’s past purchases, and also factors in data such as browsing behaviour trends collected from across the customer base. This allows the technology to make calculations of the probability that a certain message, search result, offer, etc. will resonate well with the customer and drive further purchasing.

Building this stack from the ground up has allowed Wayfair to select Aerospike as its own best-in-class data platform, and position it between the various customer touchpoints, in order to have excellent visibility and control of the customer data.

2. Building a modern data stack around legacy architecture

The question for enterprises is how to ‘be more Wayfair’, despite the presence of legacy data infrastructure.

The end-goal is a setup where the company’s legacy database is just one of a range of other data solutions, which sit around a cloud data lake licenced separately by the brand. The monolithic database is left in situ, along with the legacy ERP – whilst the modern data stack spans the entire business, and can easily connect to new martech modules as they are acquired.

The below diagram by Altexsoft, a digital agency, is a useful representation of the various components in the modern data stack.

Not every marketing professional needs to understand this in detail. Many non-technical professionals, however, will recognise some of the example vendor logos in this diagram – such as Salesforce on the left, and on the right, Tableau (a very widely-used data analytics & visualisation tool).

The crucial things to understand from the marketing perspective are:

• the data sources (left) – which could be the technology that runs your webstore/website, your app, your platform, any business software used by your teams, or any other marketing touchpoint
• the data uses (right) – the software which is used to turn the data you’re storing and collecting into actionable insights

…since these are the data inputs and outputs that marketers are likely to use or rely on for campaigning.

With more software components working alongside each other, each highly specialised for certain purposes, the entire platform becomes faster and more responsive. Processing times also fall, because the modern data stack is cloud-based – spreading the load across many different vendors’ cloud-based servers, as opposed to one company’s servers.

Importantly, a modern, API-first, cloud-based database in the centre of the stack exposes data across the much of the business (including from the legacy ERP), and allows for processing at sufficient speed for AI-driven marketing.

Steps towards adopting the modern data stack

Brands cannot simply switch overnight from a monolithic/traditional data stack to a modern data stack; the diagram above implies years’ worth of development and integration time.

Rather, they will evolve towards the above setup by adopting modern data solutions which are, themselves, components of the modern data stack. This gradually becoming easier, since dedicated data platforms are becoming widely available for specific marketing purposes.

A solution of this type which has increasingly been in the spotlight in recent years is the CDP (customer data platform). CDPs are often bundled with CRM platforms, but recently, the need to do more customer data has led brands to licence more advanced CDPs as separate modules.

The CDP is intended to draw on the various sources of customer data from across that business, process that data into rich customer profiles, and push that data to any given CX or customer engagement tool. That could include to new customer touchpoints and channels that the legacy ERP vendor had not anticipated – from a mobile app, to extended reality (XR) experiences –  which could be handled by the modern CMS solution. It could also include the media advertising solution that you use to advertise on social networks (i.e. Smartly).

AI can enhance this module in a number of ways, such as…

• ‘filling in the gaps’ about that customer, where there is limited data, but where the data you do have is recognisable from customers that you know more about
• predicting the customer’s next move(s) with a range of probability, and allocate marketing budget based on the likelihood of them taking certain actions, and the predicted ROI.

Combined with a composable CMS or campaign management tool, that data could then be used to tailor ‘content’ – words, images, shapes, colours, sounds, products, offers, advertising, etc. – based on the customer profile. If the CMS/campaign management system was also AI enabled, it could measure and optimise its effectiveness through machine learning as campaign data builds.

Historically, implementing CDPs (and other powerful data modules) has been a major challenge. A Forrester study of 313 CDP users in the USA, carried out in 2022, showed that…

“…only 10% say their CDP meets current needs and only 1% think it can handle future ones.”

The difficulty with this module is that it traditionally came with bundled with an additional database – requiring a major data implementation exercise to pull all the company’s customer data into the CDP’s data platform.

Older CDP vendors may have viewed this as necessary due the fragmented state of brands’ data; by contrast modern data tools often avoid introducing a new database to the business. Instead, they simply connect to the existing available data sources and operate as a processing and activation layer on top. One such CDP vendor is ac company called Hightouch.

Data vendors with similar approaches are available in other areas of the stack.

In supply chain and fulfilment: a leading AI-powered inventory solution is Invent Analytics. The vendor claims to be able to generate 1.5-5% improved profitability at any enterprise currently relying on legacy demand planning software – which could be millions of dollars a month at an enterprise brand.

Speaking on the Martalks Podcast (disclosure: this podcast is produced by Navigate B2B for The Rosenstein Group), Invent Analytics founder Gurhan Kok explained…

“Perfect is the enemy of better, so we don’t wait for perfect. We wait for acceptable quality data: 70-80% in a month rather than 90-95% in a year.”

Similarly to Hightouch, Invent Analytics is a pureplay processing and analytics layer which makes use of the brand’s existing database(s) – which drastically reduces implementation and training times. It compensates for any gaps in this data with its advanced algorithms and AI functionality.

In one case study, Academy Sports + Outdoors, a retailer was able to go live in in only 3 months despite having over a million SKUs. That’s compared to the 12-18 months implementation time that would be typical for a traditional inventory analytics solution.

*

As implied by Kok’s comment above, the value that brands get from these novel solutions is influenced by the quality of the data inputs – so the benefits of these implementations will build over a time.

And, significant obstacles to the modern data stack remain.

Though modern tools are a lot easier to implement, this is relative. Saving a year on implementation is attractive, but a 3-month implementation project is a still a major commitment.

And as the modern data stack evolves, brands will find costly gaps appearing between their evolving composable architectures, and tools which have been ‘left behind’ in the legacy data stack.

Giving the example of the POS, Cygnet’s Peter Ross explains…

“Companies transitioning to a modern data stack are often not prioritising moving the POS as part of their digital transformation.”

In such cases, running analytics that combines POS data with data in the modern data stack…

“…would require either a series of complex queries, or a large data export and mangling or exporting everything to a BD platform – which in itself might not be able to query by this without setup.”

This is a phenomenon known as ‘technical debt’: where the significant value of historical technology investments produces a persuasive business case against digital transformation.

Ultimately, this is a question of opportunity cost.

In the short term, in certain areas of the stack, it may make business sense not to evolve. But over the next 3-5 years, failing to expose certain data silos for AI processing will ultimately leave you at a competitive disadvantage to those companies which took the plunge. 

In marketing, AI is a means to an end

Currently, the majority of businesses today have a hybrid setup, with some modern-cloud-based data management tools, a centralised database in their monolithic ERP – and varying connectivity across the whole stack.

Ultimately, the degree to which brands can harness AI will depend on the speed at which they can break these legacy dependencies, and unite the majority of touchpoints and tools under a modern data stack.

Much of this pathway is being laid through the adoption of composable martech – but this is a process that will play out over several years, due to the significant difficult of developing on or replacing existing software.

Granted, there are various pieces of martech which can be implemented in only weeks or months, but other, more complicated modules take much longer.

Implementing Commercetools – an ecommerce platform, and a leading advocate of composable commerce architecture – frequently takes brands 1-2 years from project go-live.

Even the work of implementing a single, enterprise CRM can run beyond 12 months, as different teams re-evaluate processes, databases and integrations around the contingencies of the new technology. As a result, many brands still operate multiple CRMs with little or no data passing between them.

Meanwhile, customers increasingly pass freely between digital touchpoints and marketing channels, generating data which can either be turned to the brand’s advantage – or, which risks leading to more irrelevant, impersonal marketing if improperly managed.

The brands which overcome these legacy constraints will be those best-placed to win or retain market share. That will accelerate, as AI extends the rewards that brands can reap for those efforts.

But it needs to be understood that AI is not the strategic ‘north star’ here; rather, it’s better-quality marketing, enabled by a better grasp of customer data, enabled in turn by modern technology.

That, of course, has been true now for a couple of decades – regardless of any considerations of AI.

About Navigate B2B: SaaS marketing specialists

Navigate B2B is a content agency that specialises in highly differentiated, often technically complex businesses.

We collaborate with business and technology leaders to produce creative media, digital UX and thought leadership that engages and educates their target audience.

By hiring and training the cream of writing talent, we produce content that founders and technology leaders are proud to put their names to – enhancing your network and building your reputation, with the minimum demands on your time.

And with the rigorous marketing & reporting that you’d expect from a full-service agency, we ensure your content publishing efforts are driving sales, and helping you to achieve your wider business goals.

Visit NavigateB2B.com to find out more.