Rob May has been a tireless leader for us in the startup community in the Boston area. With his recent post to Inside AI, he is once again leading the charge and inspiring us. At Gamalon, we too have been learning about what works for marketing AI technology to enterprise Customer Experience and Digital Marketing leaders. I was so inspired by Rob’s generosity with his insights at Talla that it made me want to contribute what we have learned at Gamalon as well. Our learnings are slightly different, especially regarding the relationship between SaaS and AI, and I hope that these additional insights contribute to furthering the discussion and the acceleration of AI adoption.
How to market enterprise AI to the customer experience industry
As a quick bit of background, at Gamalon some of the key problems we have been working to solve for customer experience or digital marketing leaders are: “How do I understand how customers are actually talking to my chatbot? Is it saying the right things and providing good customer interactions? … What about our human reps - what are customers really saying to them? How do we quickly discover and understand new or trending customer issues? … and what about other open-ended customer feedback such as from surveys?”
At Gamalon our vision is to enable a deeper level of human connection, and we believe that the only way to improve something is to measure it. We use next-generation machine learning to read what customers say to a company’s chatbot, customer service reps, or open-ended survey questions and provide automated insights in an easy to use visual summary. Our next generation machine learning (developed out of MIT and Stanford, largest DARPA investment in machine learning from 2013-2017, $35M+ development) can automatically find stories in customer messages, no matter how they are worded. (Commercial break: If this sounds useful to you, get a demo.)
We too have been running marketing experiments. This is a complex space and it is evolving rapidly, so in some cases we have arrived at the same conclusion as Rob, and in other areas there is room for different interpretations of the data.
14 16 Lessons Learned
#1. Use content to educate people about AI, generally. Rob suggests to tell people upfront what specific use cases your AI does and does not do, and we couldn’t agree more. We find that the quickest and most direct way to show this to customers, is actual screenshots of real customer conversations, and videos of “ah-ha” moments coming out of the machine learning analyzing huge numbers of conversations.
#2-4. Rob says, "‘Automation’ is the magic word for top of funnel ads.” The best ads we have found address a direct human need that AI can help with. The icon for a phone SMS text messaging app, with 22 million unread messages in the red bubble is one of our favorite ads. What works well on the website? I tend to favor, “show, don’t tell,” letting customers interact live with the machine learning.
#5. Your VCs are wrong. Specific use cases and benefits perform the worst.
Rob wrote, “For us, our worst performing ads were around things like ‘Close Support Tickets Faster’ or ‘Clone Your Best Reps’ or ‘Do Blah Blah in Slack’ or ‘Share Knowledge More Effectively’. VCs will tell you to focus on a specific pain point, but, automation is like this meta pain point that is working way better for us, and for most of the AI companies I know.”
Rob says that “automation” is a trending search term right now. Rob also lists, "knowledge base" and "support platform" and "automated support" as examples of real business buyers looking for a real solution. These search terms sound like people looking for technology, more than expressing a problem they need to solve.
Here’s an anecdote. At O’Reilly AI this September, I asked people to raise their hands if they were, “IT professionals,” and more than half of the audience raised their hands. Then I asked for hands from people who were, “AI professionals,” and virtually all of the same people raised their hands. But what we have found at Gamalon, is that outside of the banking vertical and major technology companies, the majority of these AI professionals are not actually your enterprise budget holders. The buyers are business people.
Rob’s successful search terms all feel like people searching for technology. Maybe some very early adopter business buyers will do that. But overall, I think what we are witnessing is a slow migration from innovation technology teams interested in technology to business leaders focused on solving a specific problem in a specific use case. Gamalon continues to expand our work with the Global 500, many of them technology leaders in their own right, who could appreciate the differentiation of our core machine learning technology from the beginning. But simultaneously we want to participate in the gradual and noisy evolution towards addressing business leaders in the Global 2000, and not just AI professionals. My perception is that this is a pretty routine crossing the chasm activity, even as it is a bit harrowing while you are in it. The entire AI market hasn’t quite crossed the chasm yet, as evidenced by Rob’s search terms.
That said, there are business people today who are not AI professionals and would not search for a technology, but who have clear problems that machine learning AI can solve, and are generally the kinds of people who would buy from a startup. For example, we meet digital marketing leaders every day who know that they want chat on their website in order to convert more leads, and that more than half of today's customers expect companies to be available 24/7. What is their problem? They are concerned about choosing and maintaining the right chat solution, and they are often trying to choose between human powered chat (which can be expensive, higher variance, and less responsive) or chatbot (which is generally rigid and inaccurate and limited in capability).
They usually do not realize that they can use a combination of humans and bot. Similarly, they usually have not thought about how they are going to monitor and analyze what customers are saying in these conversations to the human rep or chatbot in order to continuously improve customer experience and conversion rates.
So customers are educating themselves, at the same time as the landscape of solutions is changing just as fast. At Gamalon, we have begun to converge on a classic challenger sales approach, where we ask questions that we know the customer will need to be able to answer. Questions like, “Have you thought about how you are going to know where your human and/or bot powered chat needs improvement?” We are still in a market where we need to educate buyers about what pitfalls or trade-offs they might encounter, and a challenger sale is a good way to do this. We need to coordinate this with challenger based marketing as well.
#8. Free trials don't work well, because the product may not perform well.
Rob wrote, “We had a free trial for several months. What we learned was that all our paying customers did not come in that way. In a demo, you can show the intelligence of an AI product on a data set - we can even copy in data from a company's website before a demo, to show them info in their own language, but free trials haven't closed very many deals for us.”
I agree with Rob that people have been trained by SaaS that, “every tool should have a free trial, and be super easy to setup…,” and that can be difficult for machine learning based products that need to be trained on data before they perform well. But sampling is a time tested marketing strategy. If you sell M&M’s and you know that people will generally love M&M’s once they have tasted them, then you just send out small free packets of M&M’s. The free software trial is just the time honored marketing strategy of sampling a great product to customers.
It was a ton of work to develop, but at Gamalon we made it a focus to develop a free demo environment for our machine learning where customers can experience our system. Virtually all of our customers have come from a free demo followed by a free evaluation. Mind you this is *not* a free usage in the SaaS style with a self-service evaluation period where the customers try the solution by themselves. This is a short engagement (e.g. a week) to help a customer who has expressed sincere interest evaluate Gamalon’s solution, and we help them. It’s really a standard Proof of Concept (PoC), borrowed from the standard enterprise software sales process.
That said, this may be a function of the maturity of the technology. With more mature technology, a more mature market, and having learned how to repeatedly get customers to a fast win with the technology, AI should scale to free trials in the future. We think of our PoC as concierging a future self-service trial.
#10. Events work well because you have more time to explain yourself.
At Gamalon, we do a lot of our selling through events right now. We are most successful when we actually bring our demos and let people play with them at the booth. It is difficult to automate something before you have done something manually first. Events give us a way to manually operate our early funnel so that our team learns what works and what doesn’t.
#11. Bot companies have the dual challenge of marketing their bot, and the thing their bot does
I think this is one of the most interesting points that Rob makes. His basic premise, which I find inspiring, is that just as everything is now SaaS and we take it for granted, by, say, 2028, everything will be bot and we will take that for granted. Buyers will just search for the function or problem and assume that the solution will include machine learning and bot capabilities. That seems right to me.
At Gamalon, we have recently done another round of the customer discovery process from Steve Blank’s Four Steps to Epiphany. We get out of the building and ask customers specifically what problems they need to solve. For customer experience automation with machine learning, the pay dirt in customer discovery is something similar to, “My boss really wants me to deploy chat, but if customers start complaining about the solution I chose because I couldn’t monitor and continuously improve it well enough or quickly enough, it could cost me my next promotion or worse. I am worried that either my human-powered chat or my new chatbot will do a bad job talking to customers. I need a better way to monitor, and when there are issues to quickly drill in and find out why customers are having problems.” We continue to find ways to translate these real problems into search terms for lead generation.
At the end of the day, one ultimately has to find a bunch of folks with an unsolved business problem. That’s done with good old-fashioned get-out-of-the-building customer discovery and market research. This kind of research zooms in on the market segment, buyer persona, and pain point.
#15 Here is one of my own. Prepare separate collateral to sell to both IT/AI professionals and functional business leaders.
Rob wrote,“Now SaaS has trained buyers that every tool should … be hyper targeted to one key use case. AI isn't like that. If a system learns, the more use cases you can give it, the more it can learn if you expand it into more parts of the org, more business processes, or more pieces of the technology stack. Systems that are more integrated are going to perform better. SaaS broke apart big systems to build best of breed workflows. Now AI is re-integrating those workflows to be more intelligent across them all because it sees the data across them all.”
Ultimately, will enterprises want personal AI software, departmental AI software, or enterprise-wide AI software? I agree with Rob that the more an AI platform knows about your business, the more value it can produce. In Gamalon’s case, we see a future where every channel in which a company communicates with customers via text (social media, webchat, customer service, surveys, etc.) will be connected to the same AI. If we could hire one ultimate all-in-one marketing, sales, and customer service employee to learn about and interact with each of our customers throughout their lifecycle, wouldn’t we want to do that? That brilliant empathic employee would be much better at serving our customers than a group of separate departments who are not as coordinated with one another. Because the enterprises that use the AI technology in this integrated fashion will be more successful, we anticipate that the market will likely evolve in this direction of AI integration. As evidence, our current customers are rapidly increasing the number of data integrations that they attach to Gamalon’s machine learning.
All true, BUT... Today in 2018, because of the legacy of SaaS and the way enterprise software buying will continue to work for the next couple of years, you can’t have the enterprise AI hub until you win a first key use case. Then you can expand from there. We can’t fight it, and we shouldn’t even try.
So we need to focus on a single use case. Once we do, how will the buying process work? Who will be involved? Software used to be purchased by a centralized IT organization. Since the advent of cloud and SaaS and the shrinkage of IT budgets, the buying power for SaaS has mostly moved outside of the IT organization to functional business leaders. But those same IT organizations that got almost entirely out of the business of buying software, are now very interested in leading the enterprise assessment of AI, and they correctly see that an enterprise AI hub is a 1+1=3 proposition.
In banking, the wallet for AI purchases is still very strongly held in these IT and Data groups. Outside of banking, in the Global 500 across B2C manufacturing, insurance and other verticals, the budget comes from the business leader, but IT/data groups are key influencers, and may even sponsor the software and walk it into the business leader’s budget.
In somewhat smaller companies lacking a large centralized data science or IT group, the business leader will be the buyer. These folks are used to buying self-service SaaS software for their department. We need to find ways for AI to fit within that framework. That said, as startups race to address this buyer through a more SaaS style offering, these same buyers are simultaneously moving to add a centralized AI buying expertise to their organizations, even if it is just one person who is an “AI leader.” I predict that in the miedium-term, the market will meet somewhere in the middle. I am told by those who lived it that software buying in the 1990’s looked like this: you would meet your business sponsor and they would get excited, and then they would send you to go have lunch with their CIO, who would farm you out to one of their associates in the org to evaluate you, and then you would be sent back to your business sponsor to close the sale. I think it is likely that this is what the AI sales process will look like for the next few years in the Global 2000.
#16 Sample the Product
A centralized IT/AI may want to get involved in the buying process, and that can be good because it may lead to land and expand opportunities. One way to speed up the evaluation, however, is again through easy product sampling. Like Talla, Gamalon gets a lot of questions from customer experience and digital marketing executives about how they can evaluate our machine learning versus others. As Rob says, “Most SaaS gives you binary outputs, a yes/no, an action, but AI gives you ‘I'm 94% sure…’'. How do you deal with that, or evaluate whether that 94% number is a good number? It's like figuring out if a meteorologist is a good one.”
This is a big frustration for enterprise buyers evaluating AI solutions. Potential customers were always essentially saying to me, “Help me help you. You need to be able to articulate to me how your machine learning technology is better or smarter.” That’s a little like a car buyer requiring the car salesperson to explain why the car’s fuel injection system leads to greater gas mileage. Does either the salesperson or the buyer know enough to appreciate the science and how it supports the claims of differentiation? Given that most people do not have a deep technical background in machine learning, it’s really hard to explain why one machine learning system might be able to understand things with less human help, or learn faster, or scale to more foreign languages more easily. So what does work? Benchmark white papers or data sheets do not seem to help much. But we have found that if the system gives the right answer more than 90% of the time, people can tell. Executives can pretend to be a customer and interact with the system for 15-30 minutes, and they will either be satisfied or they will not be.
This might be a basic lesson - if your machine learning technology is good, why not just let people interact with it directly in a demo environment before they buy it? Again, this is the time honored marketing tactic of product sampling. Essentially, show, don’t tell. And do it in such a way that the business leader can get it and drive a decision faster. Clearly for Talla and Gamalon setting up a fast easy demo environment has been a core strategic focus, and this may be important for machine learning startups more generally.