Misguided expectations are one of a number of growing pains that marketers face as their AI strategies mature. CMOs have struggled to advance the technology’s use cases beyond low-hanging fruit, such as using it for social media content creation. Also, anti-AI sentiment—which has gained steam in the industry’s creative circles—continues to be a thorn in the side of CMOs.
High costs associated with AI are making matters more difficult, especially because many CMOs appear to incorrectly assume that returns will quickly offset their investments. Consider a popular AI use case among marketers—using generative AI to create personalized content. Implementing the software behind this capability could cost $750,000 to $1 million, according to Gartner. And keeping the system updated with data, human training and other resources leads to an insidious investment, requiring between $790 and $1,200 per user, per year, according to Gartner.
The same Gartner research shows that other generative AI deployment approaches are even more costly, such as creating a virtual assistant, which can call for between $5 million and $6.5 million in upfront costs and $8,000 to $11,000 per user per year. Virtual assistants are typically chatbots that can perform a variety of support roles, including for customer service and internal use cases. Brands such as Volvo and Walmart have experimented with virtual assistants, although the specific costs of these activations are unclear.
Also read: 5 AI marketing activations to know about right now
Further, developing one’s own custom AI model can cost between $8 million and $20 million upfront and $11,000 to $21,000 per user per year. These are AI platforms that are built entirely in-house, as opposed to relying on foundations developed by OpenAI, Google or other providers.
Several agencies are designing custom models because they can play roles that are ultra-specific to their businesses, as well as be trained on mostly proprietary data. Edelman, for example, has built a custom large language model (LLM) that is meant to help marketers cultivate brand trust.
Greene admonishes marketers to remember that AI is not a one-and-done implementation—it requires a continuous flow of updating to not only stay operational but also to scale to a company’s ambitions.
More from Ad Age: 9 AI tools agencies are embracing right now
Aside from having unrealistic investment expectations, CMOs often demonstrate misunderstandings about how AI actually works, said Adam Birkenhead, chief operating officer at Digitas North America. For example, CMOs have a desire to craft a marketing message that is virtually the same every time, but that’s not how AI works, Birkenhead said. Variations are noticeable between almost any two outputs generated by AI, even when drawn from the same input within the same system during the same AI session. For example, asking a chatbot to write ad copy, using identical parameters, will likely return nonidentical results.
The work required to smooth out an AI system’s outputs and find more enriching use cases requires many resources, one of the most important of which may be time, Birkenhead said. Programming AI to write alt-text, for instance, may seem undaunting for a single image, but applying that same task to scores of images—in a way that delivers a consistent output—requires building additional infrastructure that changes the process altogether.
