Labelling products with AI is common at present, with many technology industry observers scratching their heads to find a platform, service, or piece of software that doesn’t use AI, machine learning or one of the other, often ill-defined variants like ‘deep’ or ‘cognitive AI’.
A possible reason for so many instances of AI being applied to products stems from AI’s many different forms – reflecting its multimodal nature that works on a variety of data, including text, video, and imagery. Additionally, there is a range of disparate terms, such as agentic AI (where AI can interact with other software via APIs) and generative AI (in which AI can create new data based on sources it’s accessed).
Software is an industry that’s fond of new acronyms, with the very latest buzz-phrase being MCPs, a fork of APIs designed specifically for AI models to interact with each other. But the core label of ‘AI’ is now commonplace on cloud services, software, hardware, networking, storage, and every marketing app under the sun.
The specialisation problem
Digital marketers are trained in the detail of market variation, messaging, broadcasting methods, and statistical analysis. Like many specialists, marketers lack the training in computer science and technology; those are the purview of software developers, database experts, systems architects, cybersecurity specialists, and a dozen more sub-genres in the technology space.
That lack of specific knowledge in technology makes marketers as prone to any other group outside pure-play technology to un-provable claims about the underpinnings of the tools and platforms they use.
As an example, we could take a software SaaS platform that allows marketing functions to correlate their various digital campaign (paid and organic) and e-commerce elements, and see their results in a series of dashboards. Here, marketing professionals can, for example, examine statistics, create reports on individual campaigns, and undertake competitor analysis. Our imagined platform might also capable of scenario modelling and forecasting possible revenues from e-commerce sales, with predictions based on historical data and the performance of marketing activities.
If users could pose their queries to such a platform in natural language, such as “show me the best performing campaigns on mainland Europe”, then a large language model would parse their query and produce the desired statistics. Right there is an implementation of AI, acting as an intermediary between users, and the statistical data held in the platform’s database.
But does that function justify a strap-line for the platform that describes it as “AI-powered”?
The black box
The problem marketers have is one experienced by the vast majority of users of AI platforms: what’s under the hood in proprietary software is just that – proprietary, and closed off from external examination. It’s impossible to verify any claims about AI “power.”
There are perhaps dozens of solutions on the market at the moment that are, in whole or part, just like the platform described above. Almost without exception, they claim AI elements, and none of those claims are open to outside scrutiny to a level that would satisfy a specialist in IT. Is there really an AI ‘engine’? What are any AI elements’ biases or shortcomings? What sources were used to train the AI?
If we take competitor analysis as an example element of our imagined platform, we see it will likely present available metrics (traffic, lighthouse scores, share-of-voice statistics, and so on), and it’s unclear whether there is an AI acting internally that affects or processes any data. Perhaps the attractive dashboard shows results based on, not AI, but classical algorithms working on rule-based aggregation.
Similarly, many platforms offer predictive functions: which campaigns will work well, what might be the income generated if we take a particular action?
Predictive functions in many platforms are based on time-series regressions, processes that could be defined as machine-learning functions, but are generally much closer to classical statistical modelling than to any revolutionary ‘deep AI’.
In the latter case, there is likely a software element that could be used to justify a claim of ‘AI under-the-hood,’ but its implementation might be fairly basic, and predictions may not be better than those derived from a combination of human experience in the sector, gut instinct, and good old-fashioned metrics.
(It’s also worth noting that machine learning algorithms are pretty terrible at maths. Ask for financial advice from AI at your peril.)
Seeing is believing
If marketing software companies were to offer AI-powered platforms that they wanted to differentiate from the hundreds of other similar tools, they could do so by publishing details of feature sets, model families, hyper-parameters, and validation splits, to name just a few details that would justify their AI claims. To most of us, such information is impenetrable. But to a data scientist, even one fresh out of college, such information might act as validation that there are things happening inside the closed, black-box platform that make it different. Verification by means of some degree of openness would let objective third-parties (and organisations thinking of signing up to an ‘AI-powered platform’) be sure they weren’t being sold snake oil.
The adage to ‘follow the money’ is, at present, not particularly good advice. Investors are throwing money at any company with an AI element in its (self-published) description, and there appears to be little to discern between products that will or won’t bring value to their users. Checking out which ‘AI-powered’ marketing platform is getting funding from investors and venture capitalists isn’t particularly useful as a yardstick of quality at the moment.
Marketers’ experience, specialisation, knowledge, and gut-feeling remain the most valuable commodities a team can have, especially when so many claims of AI are un-verifiable, if not specious. What’s surprising is it’s the marketing function that seems to have gone all-in on AI. After all, which type of business specialist should be most immune from unverifiable marketing hype?
Software in its classical, algorithmic forms has and will continue to bring massive benefits to the marketer. It may well be enhanced, in some measurable ways, by machine learning. But platform vendors should be willing prove, to a suitably-qualified specialist, just what effects AI has on outcomes, and how AI enhances a platform beyond what’s achievable by ‘traditional’ algorithmic software.
(Image source: “Snake Oil shading and coloring” by opacity is licensed under CC BY-NC-ND 2.0.)
See also: TikTok pushes AI ads that look just like real creators
GIPHY App Key not set. Please check settings