草莓社区

By Rick Chavie
Founder and CEO, Firefly Kinetics 

Democratization of AI

AI鈥檚 potential makes for a decision that all businesses will have to face: how do I surf this latest wave of creative destruction? Offload routine work to one of the many generative AI toolsets? Tackle complex, high-value decisions with AI where other business intelligence tools are lacking? Bring expensive talent in-house to plot a path with AI or rely on external experts? 

AI tools are now commonplace: tools based on large language models (LLMs) such as ChatGPT (and its derivative in Microsoft鈥檚 Bing) or Jasper AI and HubSpot for marketing; generative AI e-commerce tools from Amazon for producing product descriptions and eBay for content derived from characterizing photos; Shopify and Adobe with multipronged GenAI for e-commerce and marketing; and Bard from Google鈥檚 which also leverages a user鈥檚 account data.

More, Better Boilerplate

Given the pervasiveness of AI offerings, major companies are making investments in AI, with three in four business leaders rating generative AI (GenAI) as the top emerging technology. These leaders see the top value for AI in improving efficiencies, with 97% of leaders agreeing that deploying GenAI AI will allow employees to be more thoughtful and creative by relieving them from manual processes.¹

But seriously? A consensus goal of AI use focuses on efficiency improvements for routine manual processes, e.g., better boilerplate content. However, this consensus choice of GenAI for routine use in content generation also risks LLM 鈥渆rrors with confidence鈥� (or 鈥渉allucinations鈥�). OpenAI鈥檚 CEO Sam Altman acknowledges these quality issues associated with LLMs: 鈥淚 probably trust the answers that come out of ChatGPT the least of anybody on Earth.鈥�²

Commerce Complexity

As digitalized commerce continues to expand, along with AI-generated digital ad clutter, marketers face rising complexity in personalizing offers to consumers. Among the key issues:

• The proliferation of touchpoints: social media, texts, mobile, short-form video, QR codes, email, direct mail, websites, kiosks, POS, Smart Voice Assistants (SVA), etc.
• Overcoming the deprecation of third-party cookies, upon which 85% of them have relied to identify and track target customers across commerce touchpoints.³
• Identifying the permutations of product attributes preferred by individual consumers.
• Measuring consumer behavior across digital and physical channels.

For marketers, the complexities do not stop there as they look to AI: 鈥淚t turns out there鈥檚 a fatal flaw in most companies鈥� approach to machine learning, the analytical tool of the future: .鈥�⁴

The resources and priority companies are placing on GenAI efficiency initiatives will improve the odds of getting to production and taking out costs. However, using AI for scalable demand generation that is differentiable from competitors (with the AI same tools) is a higher hurdle.

Models Breaking Badly

Marketers are creative in evolving 1:1 or persona-based personalization models, turning the 鈥渇lawed鈥� creativity of GenAI into a source for new ideas and finding new ways to track consumers in the cookieless commerce era. With the deprecation of third-party data, they now embrace zero-party data (e.g., direct consumer surveys), first-party data from their own websites (own cookies can still be used), and second-party data from partner or broker sources.

The increased effort to procure and synchronize these new data sources in service of 1:1 personalization suggests that the use of personas (i.e., representative customer segments typically based on demographic, psychographic, and even biometric data) will be a preferential strategy for aligning customized AI content and offers to segments.

Soon, even the persona approach may be a challenge: 鈥�90% of online content will be synthetically generated by the end of the decade,鈥� maintains Nina Schick, author of Deepfakes: The Coming Infocalypse. What could go wrong in marketing to fictional individuals (personas) with generative technology that is trained on legacy and synthetic LLM training data and is prone to hallucinations? And in responding to changes in highly dynamic digitalized markets where competition and new products emerge frequently and are just a click away?

Chasing Perfection Routinely

Vince Lombardi of the Green Bay Packers said, 鈥淧erfection is not attainable, but if we chase perfection, we can catch excellence.鈥� Perfection in this case is avoiding the pitfalls of AI while mining the amazing potential.  But excellence in AI should lift both the top and bottom lines to differentiate and grow in competitive, digitally enabled markets.

Observation: Game-changing upsides will come from using AI to change our decisions: replace mediocre choices with routinely better decisions and content having predictably better outcomes. Successful structuring of AI use can move company productivity to the right of today鈥檚 bell curve of outcomes. Not perfect outcomes, but ones that routinely accelerate market share gains.

The following are the keys to achieving AI success:

• Implementing guardrails for GenAI. The guardrails come in several forms, including 鈥減riming the pump鈥� with concepts to keep it on message, forcing certain precise content into the output (e.g., for a product description, there can be legal liability if content inaccurately describes food ingredients, pharma instructions, etc.)
• For LLM applications, incorporate internal training data in producing content that is better aligned with your brand and your customers. This training data should include touchpoints to better pinpoint how and when customers receive offers and shop.
• Incorporate collaborative approaches with AI tools to unite their processing power with human creativity. Such collaboration produces greater results than either alone. In the book, the Second Machine Age by MIT鈥檚 Erik Brynjolfsson and Andrew McAfee, the authors make a compelling case for the superior outcomes of such an approach
• Don鈥檛 be satisfied with efficiencies. Beyond the obsession with content generation, AI-enabled decision-making for demand generation offers great upside. Lifting marketers, product managers, and sales teams from average to good or excellent is in play with these technologies and should be pursued aggressively to differentiate a company鈥檚 outcomes.
• Deploy 鈥渁tomic level鈥� measurement of results. Systems today can process enormous amounts of data yet effective measurement requires investing in high-quality data. Great granular data with off-the-shelf AI can outperform a great AI tool with poor data.
• Massively scale test and learn. 鈥淏lack box鈥� AI obscures insight into the 鈥渉ow鈥� of results, much less attribution to specific marketing tactics. Yet, precise measurement of changes in variables, such as which touchpoint is used and when 鈥� or altering specific words - in offer design, can divine the 鈥渨hy鈥� with robust feedback loops in live testing.

So, enjoy the chase on your way to AI excellence!

[1]  June 2023 update of 200 US business executives across industries in organizations with $1 billion or more in revenue on their views of generative AI.

[2]  Fortune, Matt O鈥橞rien, and the AP. August 1, 2023

[3] The State of Marketing Trends in 2023, from AI to Z. HubSpot.

[4]  Jack Flynn. Jun. 13, 2023

[4]  Venture Beat, July 19, 2019

[Blockquote 2]  NielsenIQ. Dec 5, 2019

[Blockquote 2]  NielsenIQ. 23 June 2015

Back to AI Edge: Q4 2023