AI. It’s on everyone’s mind—and marketers are no exception. You’ve likely heard about it from co-workers, vendors and peers, and if you had a nickel for every AI mention you heard … well, you get the point. With the release of ChatGPT late last year, OpenAI supercharged the conversation around large language models (LLMs), marking 2023 as “the year of AI.” As of February 2023, three months after launch, ChatGPT is estimated to have surpassed 1 billion page views and over 100 million users.
You’ve probably used ChatGPT yourself out of curiosity if nothing else. Or as a marketer, you may have considerable knowledge about its marketing applications. Ultimately, though, a lot of it still feels pretty fuzzy and buzzy, and the countless martech solutions that highlight the latest and greatest Generative AI (GenAI) and LLM capabilities, still lack broad adoption. That said, in some cases there is early tangible value being offered by the marketing ecosystem. All of these solutions have a common denominator with today’s modern organizations: a robust, transparent and scalable data strategy, and the prerequisite to AI is the heartbeat of modern marketing: customer data.
The protection of the brand’s customer data is paramount – customer data privacy, compliance, and governance should be the bedrock of a brand’s modern martech stack. Regulations like GDPR and CPRA are becoming more common across the globe. A data breach, or even not complying with Data Subject Requests (DSRs), can heavily damage the brand, destroy customer loyalty and may result in significant fines. As marketers grapple with the clutter of information about GenAI and LLMs, how can they be equipped to identify which marketing vendors are truly prioritizing the security of their data? To help you navigate this increasingly busy space, we’ve outlined three questions, as well as recommendations for how to probe each question, that can help marketers evaluate AI solutions, and decipher how customer data is collected and used to power AI offerings.
Question #1: What kind of data are you using to train your AI models? And how are you collecting it?
For robust AI solutions, a marketing vendor may need (or want) to use your customer data to train sophisticated models that are tailored to your exact business needs and help you deliver personalized marketing. Customer behavioral data from your website and digital applications is foundational to AI-powered personalized experiences in the marketing space. It’s no surprise that some solutions that currently reside in your stack are already collecting customer behavioral data on your behalf (web analytics, CRM platforms, customer data platforms, product analytics, etc.). While behavioral data is important, it’s rarely the only type of data needed to properly train an AI model for marketing purposes. If your behavioral data is siloed, organizations may be forced to build data pipelines to support AI model training on a comprehensive corpus of required data. This can result in more complexity, and often lead to increased costs and inefficiencies. If the marketing vendor under evaluation is new to your stack, not only do you have to add yet another SDK to your website and apps, you also run the risk of spinning up yet another data silo for your customer data.
Another way for marketing vendors to use a brand’s data to power differentiated customer experiences is to pull it from an existing SaaS tool, such as the ones mentioned above. For instance, if your data is in your CRM system today, why not leverage that data to build and train your AI models? While the rationale is sound, this leads to increased overhead and added complexity of building a new, bespoke integration for a different point solution. It’s also worth highlighting the risks of lock-in that come with this type of bespoke integration. For instance, in a scenario where the evaluated marketing vendor integrates with a brand’s current customer engagement platform, future changes or migrations may prove to be more challenging and expensive, effectively locking you into that platform.
Lastly, if the marketing vendor solution under evaluation requires access to data using traditional file transfers, this results in additional, potential risks. In this scenario, besides sacrificing governance and security of their data, potentially eroding consumer trust, brands also suffer from the inefficiencies of adding a new pipeline, working with a disparate data silo, and having to pay to store your data in yet another platform. Chances are this approach doesn’t offer brands the frictionless path to AI modernization that they seek.
Recommendation: Probe how your marketing solution captures and uses customer data
More than ever, marketers need to capture digital signals and operationalize their organization’s data through tailored AI solutions. However, the way that data is captured is the differentiated factor to consider, and a core enabler of trust. Vendor solutions, across any domain including marketing, shouldn’t be a separate data silo that require one-off integrations; instead, they must be able to leverage the full breadth of your data from your enterprise source-of-truth, while simultaneously eliminating the complexity of integrations, the multiplicity of data stores and the friction associated with data collection. If marketing vendors aren’t using the full breadth of a brand’s enterprise data to develop tailored and holistic AI models for the brand, then you are likely missing out on value.
Question #2: How can we trust your models?
For marketing and advertising use cases, brands trust AI models to generate copy, content and audience segments that are helpful—not harmful—to their brand’s marketing efforts. They need to fully understand what kind of controls the solution has to keep hallucinations out of your content (hallucinations are essentially a response by an AI model that is not justified by its training data). There are plenty of ways hallucinations can be harmful, ranging from minor typos in content creation, to something as potentially harmful as promoting an incorrect or competing brand. By fine-tuning models specifically on the brand’s data, they mitigate the risk of hallucinations in production by owning and governing the models’ underlying data sets.
A common point of resistance that a brand faces when adopting AI/ML technologies are internal teams responsible for ensuring data security of your organization. According to a Harvard Business Review report, 79% of senior IT leaders expressed concerns that AI technologies raise security risks. Any marketing vendor solution chosen will require approval from these internal stakeholders and possess proper governance features to ensure only the right people and processes have access to PII and sensitive data.
Recommendation: Ensure access and governance are foundational for all providers in your marketing tech stack
While a marketing vendor’s solution should be able to use all of an organization’s data to maximize value delivered, brand’s should provide access only when necessary. Specifically, marketers should probe vendors about how they will be using the brand’s data, and whether they need to persist that data for the solution to work. Modern marketing solutions that can offer trusted AI should be built natively, connected seamlessly to a brand’s data platform, or leverage secure data sharing that eliminates data copies. These approaches not only simplify IT and security’s jobs by enabling full data governance control, but it also makes trust the foundation of the partnership between the brand and the martech vendor.
Question #3: Will my purchase lock me into AI in its current format, or will I be able to leverage the latest innovation as it evolves?
The pace of AI/ML innovation is extremely fast. Even OpenAI, Meta and Anthropic have made significant strides forward in the quality of their LLM models. In a typical SaaS contract, brands sign a one-year or maybe even a three-year deal aligned to a fixed fee, seat-based or monthly active-user subscription model. Problems can arise here when expectations from the sales process don’t quite match reality in production. In some cases, this type of business model can even make brands feel locked into a specific solution because of the investment they made in it.
Recommendation: Choose marketing vendors with modern, future-oriented business models
Brands should ask about consumption-based pricing, or “paying for what they use.” Shifting from a licensing model to one that charges based on usage offers a number of advantages for organizations, the main one being flexibility. This type of pricing structure allows organizations to more easily test—and potentially even switch between—multiple models from competing solutions. Maybe some models are better for some use cases than others. Whatever the case, knowing what comes under the hood with the price tag can both save you a lot of money in the long term as well as give you flexibility to back out from a bad solution.
How Snowflake helps brands find the right solutions
The Snowflake Data Cloud can help you retain ownership of your customer data while giving you access to best-of-breed AI/ML application partners. In the future, many of these AI/ML solutions may be built and executed in conjunction with a Data Clean Room as a Native Application, which Snowflake is powering today. If you’re a Snowflake customer interested in how Snowflake can help with AI, reach out to your account manager. Otherwise, let us know how we can help you own your brand’s customer data today.