An inside-out perspective on the development of MarTechBot, and its implication for marketers.
We all have experienced the unprecedented pace of AI-driven change in the last six months. The catalyst for that change was “access.”
AI’s inflection point was OpenAI’s decision to provide free and unfettered access to ChatGPT — the result: 100 million users in less than two months.
As martech and marketing operations leaders, this open access is both a blessing and a challenge. It dramatically changed our 2023 plans and priorities.
That’s where MarTechBot entered the picture approximately two weeks ago. Thanks to Marc Sirkin and the team at MarTech for allowing me behind the curtain of MarTechBot, providing insider access to how it’s being trained, the underlying tech, and the real-time learnings.
Sirkin and I discussed the implications of the contextual “MMM” prompt that he posted about. That experiment demonstrated that training MarTechBot with this site’s content would result in customized answers for marketers. The result was expected but impressive nevertheless. And led to further reflection. Here are some of the insights I walked away with.
- Start now. Learning how to train an AI bot using a company-specific language model should be at the top of your 2023 to-do list. It may not be released to the public, but the potential benefits demand that we all start taking tangible steps now.
- Echochamber effect. Watching MarTechBot respond within the bubble of marketing and martech was awesome — like a two-week-old baby already knowing how to say “mom” and “dad!” However, the implications are serious. Biases may creep in just as quickly. In the world of marketing tech, would MarTechBot soon conclude that the only solution to each marketing problem is to add a new tool to the stack?
- New marketing ops roles. We discovered that training a bot comes with all sorts of new guardrails. One example is operationalizing GPT token limits. While word counts are a rough metaphor, they are not exact equivalents given the predictive feedback loops that are the foundation of large language models (LLM). Another example would be new content ops roles to edit audio/video text transcriptions. Previously, slight inaccuracies produced by real-time closed captioning would’ve been overlooked. Those inaccuracies are consequential when text is fed into training bot models.
- Pivots. If a bot can be trained so quickly to take on a tone of voice, can it be retrained instantly? What if a brand has trained a bot on messaging and tone that’s now obsolete because of a new product direction or rebranding?
But wait, there’s more! The following are just the tip of the iceberg when it comes to new MarTech and MOps challenges (e.g., unanswered questions!) that MarTechBot prompted.
- New stack without a how-to guide. Those creating generative AI systems admit that they don’t understand exactly why and how they respond the way they do sometimes. How does a marketing ops professional explain that to customers, the executive team, shareholders, etc.?
- Balancing speed and responsibility. The race to innovate will unearth thorny legal, copyright and ethical issues. Will new content tags such as #train_on_this (or #do_NOT_train_on_this) be honored?
- The potential rekindling of marketing-IT “infighting.” Over the last 10 years, we have established some norms in the role/responsibility splits between marketing and IT. But AI tools will be used by the entire enterprise. Will marketers need to renew their cross-functional partnership with IT, or risk losing access to important datasets that IT will and should always control for the enterprise?
- Rapid infusion into marketing automation. As I wrote and spoke about in March, these capabilities also drive reinvestment in core CRM and marketing automation platforms as the foundation of the martech stack. I’ll cover the impact on data management in part 2 of the series in June. How much will change again or be introduced between now and then? (I’ve already modified my outline three times!)
In the past, vendors and/or consultants could usually help us identify where something was awry in our stacks. That won’t be the case with the AI bot stack for the next 6-12 months. We have to be the operator behind the curtain. Start today.
The post What’s behind the MarTechBot curtain? appeared first on MarTech.
About the author
Milt is currently a strategic consultant (Mission MarTech LLC) and supports clients through a unique combination of strategy, operations, and technology experiences. He enjoys being a cross-functional translator and can provide advisory services in parallel with hands-on implementation and support. With 30 years of leadership experience, Milt has focused on aligning service, marketing, sales, and IT processes around the customer journey. Milt started his career with GE, and led cross-functional initiatives in field service, software deployment, marketing, and digital transformation. Following his time at GE, Milt led marketing operations and customer experience teams at Connecture, HSA Bank and MSI Data, and he has always enjoyed being labeled one of the early digital marketing technologists. He has a BS in Electrical Engineering from UW Madison, and an MBA from Kellogg School of Management. In addition to his corporate leadership roles, Milt has been focused on contributing back to the marketing and regional community where he lives. He serves on multiple boards and is also an adjunct instructor for UW-Madison’s Digital Marketing Bootcamp.
(10)
