Discover more from B2B Wins by Steve Zakur
B2B Wins #4: The Winter of AI Discontent: B2B Marketing Edition
You are the answer! Emerging from the Trough of Disillusionment
It was just moments ago that it seemed AI would solve all our enterprise marketing problems. AI would make content more effective. Maybe content could even write itself. Emails would by persuasive and targeted converting at a higher rate. Marketers would be rich, vital, and stunningly attractive. AI could do it all and we bought whatever was on the truck.
A million flowers bloomed throughout the enterprise marketing tech stack. Some AI promises by sales reps turned out to be nothing more than some rules-based system that didn’t deliver half of the promise of the one-pagers and case studies.
Even the stuff that was “real” AI seemed less than magical. Machine learning models needed human-in-the-loop training, months of data gathering, and even then they didn’t always deliver. When the next vendor called about the latest B2B marketing technical marvel powered by AI our Spidey Sense tingled. We were deep in what Gartner refers to as the Trough of Disillusionment.
But all is not lost. With a keen understanding of how we got here we can figure out what's next. And you're big part of the solution.
The view is great from up here!
“artificial intelligence technologies have just climbed the inflated expectations mountain or have begun to move down to the trough of disillusionment. Organizations have seen these technologies' capabilities in the enterprise, and their expectations have started to become more realistic.“ Gartner, 2020
Back in 2016 or so it wasn't really clear what "AI" was going to do for the business but we were all encouraged to give it a try. Surely value would emerge. How could it not? Many of us found out the hard way.
In 2020 Gartner called out that the market expectations about AI were a bit frothy and would start coming down to earth. Respondents to Gartner's surveys were professing that AI plans were being executed. But when Gartner looked at what was actually getting deployed, whether it was in the vendor space or what CIO’s were building, there wasn't a whole lot of business functionality that was getting powered by AI. While I have no data to support this, I suspect that if you looked at business results even less was powered by the mythical beast. We had reached the very peak of Inflated Expectations. Just over the rise lurked Disillusionment.
It's been a wild ride over the past two years. Lots of external factors--I'm looking at you Covid--caused us to seek digital solutions to even more problems. In general this has been good for business especially those that were digital laggards. Many of the digital solutions promised to put AI to work. For some use cases, especially those related to task automation, AI has proven to be a boom. For example, the classification of content for better content management and personalization is a task that natural language processing (NLP) technologies have gotten pretty good at (though beware the charlatans).
So now we're tasked with figuring out how to make the best use of evolved AI tech. Where do we start? At the beginning.
Start in the right place. Start with he business.
“…[some organizations] start hiring an excellent and massive data science team. In fact, it is the most common mistake and a recipe for a disaster 3-5 years down the road…” -Anna Kostikova
I bet many of you built a Center of Excellence for AI stuff back in the day. AI talent was scarce, so you hired fast. You hired big. You've got lots of bench strength. What to do now? You should start where everything starts. The business problem.
Much of the AI technology that's been deployed in the past was "technology in search of a problem" instead of the other way around. For example, early content generators were just awful. And some rules-based AI personalization were just rules engines that were "trained" on the rules that you already had and didn't do much more than execute those rules. Both those use cases could have benefited more from traditional solutions, better processes, and improved skills of the people responsible. Both those technologies have gotten better over time but most early buyers didn't understand that they were the lab rats in vendor's experiments.
So, start with the problem with high value--where spend is high, value is high, and risk is high--and where AI technology is relatively mature. One of the reasons you've seen AI solutions maturing quickly in B2C use cases is because it's very easy to assess commerce on those three variables. One of the places we struggle in B2B is assessing value. Value attribution of many activities is difficult if note impossible. Anything in digital advertising, email, or social can at least be measured in terms of response. So problems in those spaces may be a good place to start.
For those of you who want to build your own AI tech, they key will be experimentation before committing to large scale projects. You should apply a classic Think Big, Start Small, Go Fast approach. Anna Kostikova wrote a great piece way back in February of 2021 that provided a roadmap for just such AI experimentation. One of the key cautions she makes is to ensure that you're not only budget for the engineering but also for the data wrangling. Important on data, it's not a one-time investment but a long term operational commitment.
Operationalize. Operationalize. Operationalize.
…”to reduce AI project failures, organizations must efficiently operationalize their AI architectures.” - Shubhangi Vashisth, Senior Principal Analyst, Gartner
I won't spend too much time on this one but it does bear saying. The best technology solutions, AI or otherwise, will never matter if they're not deeply operationalized. In my experience, there are two reasons AI solutions don't get operationalized.
The vendor sold you a bag of parts and tools instead of operating software: In the Enterprise Search space every vendor claims AI is deeply infused in their technology. But this means very different things to different vendors. Some vendors have AI technology but it's on you to configure and operate it. It's really a DIY approach to technology. Other vendors have technology that just works. If you have to do all the work then the odds of your tech getting fully operationalized is low. Stuff will get delayed. Resources will leave. Budgets will get cut. The bag of parts approach is fraught with value destroying impediments. Know what you’re buying.
The business is not engaged: If you've chosen the right business problem make sure that all stakeholders are committed. The magic of AI doesn't preclude the need for everyone with a hand in the problem space to be involved. The effort should be lower and the results should be better. Their need to be involved doesn't change.
This need to operationalize is not unique to AI projects but there does seem to be a belief that AI absolves the stakeholders from being deeply involved. It doesn’t.
Things are getting real. Get real.
"Notably, the AI Hype Cycle is full of innovations expected to drive high or even transformational benefits,” says Afraz Jaffri, Director Analyst at Gartner.
Very quickly, over the past two years, things have evolved and I expect that will only accelerate. Some AI tech has matured in a way that it's living up to the promise that we hoped for eighteen months ago. In addition, there are a few technologies that are emerging very quickly and are already demonstrating value.
Some tech that's living up to the promise:
Semantic Search: Most contemporary search engines should be able to not just look up words but also understand the meaning in a query.
AI as a Service/AI Platforms: Let's first understand what I mean by "platform". The best way to understand this is to think about how IT infrastructure has evolved. Twenty years ago if you wanted a bunch of servers, you'd have to buy those yourself and figure out where to put them. All that changed when Amazon Web Services and similar cloud hosting companies came along. Infrastructure--hardware, software, and operations--is now bought as a platform. AI is going in the same direction. There's no need to hire a bazillion data scientists to build your own platform. There are a bunch of these out there including H20.ai, Datarobot, AWS, GCP, Azure, IBM, and OpenAI. You can use these platforms to create your own business applications.
Generative AI: This is AI that creates things. Images. Words. Audio. Video. As marketers we're always looking for both more effective and more efficient content. OpenAI's tools and their DALL-E image generator are places where some of your experiments can take place. For example, the images for each of newsletters is created by DALL-E (I still write the words. Or do I?) APIs are available for your tech team to integrate into MVP features in your tech stack.
Because these core technologies are able to deliver real AI capability the vendor offerings built on them as well as the things your team can build are more likely to add value quickly.
You may be the answer.
You. Yup, you. No, you won't single handedly prevent your company from running down AI dead ends. But set a more reasonable goal. Can you help prevent the organization or function in which you're a stakeholder from making AI errors? That feels like a more reasonable goal. How do you do that?
Continue to get smarter. For example, you can set a Google Alert for topics that are relevant. Following "Conversational AI" will help you understand what's going on in chat technologies. Following "Generative AI" will help you keep track of all things related to text, image, and other technologies that generate content.
Build a network. Find and network with the people in your company who work on AI tech. These folks will probably have their fingers on the pulse of effective AI. Find those people either through the corporate directory or LinkedIn and grab 30 minutes to just chat about what they do, what you do, and pick their brains for ideas. They may also be helpful allies in future discussions and projects.
Lean on governance. I gave you a bit of info. You're going do your own research to keep up with changes. You're going to build an internal network. None of that matters if you're not bringing this into your management system as you assess a technology's ability to solve a pressing business problem. Focus on the problem, not the solution. Assess the solution with a healthy dose of skepticism. If you're buying a bag of parts, know that in advance and budget accordingly.
AI is getting better at solving problems. It also has a long way to go. The best way to prevent the errors of the past and to emerge from the winter of AI discontent without a bad case of frostbit is to continue to be informed about trends in AI that are important to your business.