At Talla, we have an automation platform for support teams that automates much of the support process away. It's a combination of a very different knowledge base (although we work with other KBs) and a chatbot. What is unique about our product is that you don't have to script out anything. You don't have to build bot decision trees. The bot learns from your Support documentation like a human would. Then it connects to your other Support systems to take action. Instead of a search engine, it's an action and automation engine that you can control from a web, or chat interface. (Plug: If that sounds cool, get a demo.)For 2 years, as we have been building towards this, we have run hundreds of marketing experiments. The purpose of this post is to chronicle those and share what we have learned with the bot / AI communities. We are doing this because we believe the biggest barrier to adopting AI tools is that the buyers are still figuring out what they want, how to use the tools, and how to buy them. Most of the time, buyers are evaluating AI like you would evaluate SaaS. That's wrong. VCs are not helping, because they too usually treat AI like SaaS in terms of metrics and expectations, so they are giving their portfolio companies bad advice. If we can educate the rest of the AI market on what we have learned, then we hope more buyers will understand AI, buy more of it, and that helps us all, including Talla.
So my attempt here is to show that the way enterprise buyers think about AI at the top of the funnel is different than SaaS (and companies need to adapt) but that the way buyers think about AI at the mid and bottom of the funnel is similar to SaaS (and buyers need to adapt / companies need to educate).
I've split this post into sections in case you want to skip around. They are:
1. Quick Talla History. (this will explain why we ran some tests we ran)
2. What AI Buyers Are Getting Wrong.
3. 14 Specific Lessons We've Learned At Talla.
Quick Talla History
The foundational idea behind Talla hasn't changed from Day 1. We set out to build digital workers. The way we do this is by turning every business process into a supervised learning workflow, so that we gather data to automate tasks. The path to making this vision a reality has had 4 distinct turns.
In the beginning, we launched a Task Assistant bot on Slack to learn about the Slack ecosystem. What we learned was that bots can be problematic because you can get sucked into focusing all your resources on the interface - how the bot communicates - rather than the functionality of the bot (what it actually does). People loved it, but we never thought it would be a paid product.
After that we built a ServiceDesk bot for HR and IT. This was a cool product that people liked, but a bad business model. That industry charges per helpdesk rep, and we had a product that reduced the number of helpdesk reps you needed. We thought we could move the buyers off of that. We couldn't. So we were going into mid sized companies and making $99/mo for a handful of reps. It would have taken forever to scale revenue. We killed that product.
We realized then that we had stayed out of the Support space because it was competitive, but, now we had a differentiator. As part of the helpdesk work we had stumbled across some cool ideas for what I call accelerating annotation of data using customer input. By building those ideas into a Knowledge Base, this KB became the "brain" of a digital worker. As we built actions onto our KB, it turned into a full fledge Support Automation platform, (the industry is toying with the name CX Automation). And so what follows in these marketing lessons consists of things we learned at various stages along the way.
What AI Buyers Are Getting Wrong
AI is not SaaS, just like SaaS isn't packaged software. In the early days of SaaS, buyers asked dumb questions like how much it cost for maintenance updates. VCs asked questions like how many downloads we got this month. Neither of those matters in SaaS.
Now SaaS has trained buyers that every tool should have a free trial, be super easy to setup and deploy, and 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.
AI has to be trained, either initially, or ongoing, or both. Buyers don't have workflows setup for that. Free trials don't work well for most AI use cases, because AI software learns like a person learns. New employees take a few weeks to get up to speed, and so does AI software.
And finally, AI generates probabilistic outputs. 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.
So with that, here are 14 lessons we have learned marketing to these buyers.
14 Lessons Learned Marketing AI Bots
#1. Use content to educate people about AI, generally.
The biggest problem so many AI companies have is this conversation where they ask a prospect what they are looking for and the response is "my boss said we need to explore AI." The rep says "great, AI for what use case" to which the prospect says "I don't know. What does your AI do? Tell me and we can see if we need that." Ugh. This is a big waste of time if you have to educate every prospect this way. Create some top of funnel content to point them to instead. (You can point them to a post I wrote in 2017 about the PAC Framework for deploying AI, or better yet, add to and expand that concept)
#2. Focusing on "new and innovative" doesn't work well in ads, but does on the website.
AI companies face the challenge of whether they should anchor to old things and well known buying concepts (for us, this was a knowledge base) or new things. I pushed a lot of "New Kind of Knowledge Base" messaging when Talla was in that phase. It performed very poorly in Ads, but reasonably well on the website. One possibility here is that people are looking for a thing, and you catch their eye with a different message (see #4), then the second message they see as they hit the website is something about being new and innovative.
#3. "AI Powered (Whatever)" is a mediocre ad with a slow sales cycle.
Most companies will tell you that using AI in top of funnel ads doesn't work well. It generates tire kicker leads and people researching new ideas, without BANT. It's worth a small experiment, but focus on other things.
#4. "Automation" is the magic word for top of funnel ads.
When we look at specific ads and how they relate to people who actually request a demo, "Automation" is the word that vastly outperforms everything else.
#5. Your VCs are wrong. Specific use cases and benefits perform the worst.
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.
#6. Get written about in Venturebeat and MIT Tech Review.
These were not our top traffic sources, but they were our top sources for people who actually became qualified leads.
#7. Retargeting works well.
AI sales are complex. Retargeting users with different messages gives them a chance to absorb a bit more. It's our #2 source of demo requests.
#8. Free trials don't work well, because the product may not perform well.
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.
#9. Bot app stores aren't very useful.
Discovery is a huge problem, and most people installing from the stores seem to be tire kickers. We are in both the Slack app store and MS Teams app store. We get about 3x the installs from MS as from Slack, but, the Slack ones are slightly higher quality. I know both MS and Slack are continuing to work on this, and it's admittedly a hard problem to lower friction and ease discovery, while still preventing misuse and junk bots filling up your channels. Keep an eye on this but don't rely on it for now.
#10. Events work well because you have more time to explain yourself.
We've done several things like the the MidMarket CIO Forum. They can seem pricey compared to CPC ads, but you get 30 minutes in front of a room of buyers to really walk them through your product, rather than 6 seconds on an ad or a website hero statement. Most AI products are a bit complex and in that evangelical phase of the market development, so getting small group time can pay off.
#11. Bot companies have the dual challenge of marketing their bot, and the thing their bot does.
This is tricky. There is a segment of the market that comes to us looking for "bots" or "support bots" or "enterprise chatbots." In addition, we get leads for "knowledge base" and "support platform" and "automated support" related terms. Your bot (if you have one) is a differentiator, and you need to play it up early, but, if you focus purely on bot marketing, you may end up getting confused with chatbot creation platforms, bad leads who don't have a use case. But at the same time, some business buyers who have a real use case are looking for a bot solution to that use case. This won't last forever - just like people in 2010 might be looking for a solution like "SaaS accounting software", they don't do that as much now. The "bot" piece should fall away as everything gets botified.
#12. AI solves a Lot of imaginary problems. Set high qualification bars.
Before SaaS, much of software buying was "we are the experts, let us tell you how to solve your problem" Then with SaaS, we saw the rise of customer development - "just listen to customers and they will tell you!" But customers are often way way wrong when they tell you what they want in new technology. I've seen a bunch of companies and leads, for instance, who insist that this HR benefits Q&A problem is a big one - "we have this problem where people ask a ton of benefits questions and it gets annoying answering them over and over." What people miss is, if you only interact with HR a couple times a year, you forget to go to their bot. A bot has to be your everything bot first, to get you to go to it for HR questions. I have spoken to soooooo many entrepreneurs who have tried this space and are flailing. It doesn't work. You need content to educate your buyers about the workflow change required to make AI work. And you have to stay away from marketing imaginary problems that buyers think they have. Some barriers to your sales process can actually be helpful here. Whereas SaaS was about low friction, think about some marketing and sales process friction - for better qualification purposes.
#13. Use your marketing to challenge conventional wisdom of buyers.
Lots of people are worried about doing this, but buyers have a lot of incorrect assumptions about AI. I've had people on calls a number of times tell me "I just want a tool that works like Google does for the web - plug into my data sources and then answer the questions I ask." It never occurs to them that web data is structured, so that is much easier for Google to do. And that if they want to make this happen for all their business data, they may need to structure that somehow too. AI doesn't work very well on unstructured text, for example, but buyers expect it to. They anticipate magic at times when they shouldn't. Push back on this.
#14. AI doesn't solve new "direct" problems in most use cases. It solves old problems better, and meta problems often above the direct problems, but companies should still be buying now so they don't fall behind. Find key trigger points.
I get a lot of "smart home" emails and ads. I probably should do some more smart home stuff. It's way better than what I have. Currently none of my lamps are smart, for example. But when I move next, I'll probably think more about a smart home - not because I won't have lamps and thus have a burning need for some, but because it's a good time to think about whether I should upgrade my lamps.
Accounting software users are unlikely to buy your AI accounting software because of its cool AI prediction capabilities or whatever. "Better prediction" isn't a need they have. It's a cool thing they will consider if they happen to be switching software for other reasons. I think that's why automation, cost savings, and ROI based approaches work best for us at Talla. We lead with things like "we can save a 1000 person company $1M per year." That gets more attention than focusing on the increased accuracy you can get from using our AI compared to the status quo.
I'm a big fan of trying to map your marketing to the "buyer's journey" concept. The trouble is, in an AI world, the buyer's journey is full of misinformation from different previous journeys. But the good news is, it is getting better faster. We have seen way more leads in the back half of 2018 that are educated and prepared. Part of that is probably because we've tried so many things we have learned better who to target and how to help them through the journey, but I believe a large part of it is the market waking up to buying AI.
I hope this post was helpful. If you are a company with more than 10 people in Support, I hope you will check out a demo of Talla. And if you are also building an AI company, good luck, and reach out if you have any interest in co-marketing.