Some B2B tech companies support revenue teams by providing back-office insights and guidance. Then there is Conversica, a B2B software company that uses AI for automation to support marketing, sales, and customer success teams on the front end of the customer experience.
The company introduced its first product in 2009 to support sales team activities. Today, there are AI-powered revenue digital assistants for sales, marketing, and customer success. I spoke with Conversica's CEO Jim Kaskade to learn more about these digital assistants, how generative AI comes into play, the ethics of AI-powered digital assistants, and more.
AI plays a supporting role for specific activities
Coversica's digital assistants are very interesting. You put them in place to take on the tasks and activities that can tie up so much of a person's time, and they help you do it at a scale that's impossible for a human.
Kaskade pitches a number of use cases where an AI-powered digital assistant comes in handy:
Instead of marketers creating one-way broadcast campaigns through email where people can't reply to anyone, they can integrate Conversica's AI digital assistant to send emails, respond to replies, and carry on conversations. Kaskade describes it as a "bi-directional, enriched, personalized conversation."
- Driving prospects to digital or physical events and following up with them afterward.
- Nurturing pre-marketing qualified leads
For sales, it's a lot about being that SDR or BDR for a sales leader, helping manage the mid-funnel work required to drive pipeline. Kaskade refers to these prospects as "conversation-qualified:"
- Qualifying prospects that show interest
- Fielding inbound opportunities and qualifying them
For customer success:
Customer Success Managers (CSM) have a very challenging role because one manager has to support many customers. An AI-powered digital assistant can take on many basic tasks like reaching out to schedule QBRs and EBRs, setting up regularly scheduled meetings, and sharing product roadmaps. The CSM can then focus their time on more high-quality and strategic activities.
Other use cases for the CSM:
- Conducting NPS Surveys
- Cross-selling / Upselling
Conversica's digital assistants are not like Alexa or Siri. They are purpose-built with a "skill" or conversational capability to drive specific outcomes for revenue teams. These assistants are triggered by an event, like a download, a first reach out via email, or outreach using a targeted list, and they will continue to perform until the person tells them to stop or says they are interested.
Once the customer/prospect is interested, the AI puts them in touch with a human, notifying them via email or SMS and flagging them in the system (CRM, marketing automation, customer success system) as a hot lead. But the AI isn't done at that point. It will continue to monitor in the background to ensure the human engages, and they'll ask the person if they've been in touch.
I wondered if the AI-powered digital assistant is as annoying as some SDRs with the never-ending emails and phone calls. Kaskade counters that the AI is always 'on message' and in context. With fifteen years of experience, Conversica has been able to train its AI assistants to deliver an outreach cadence that isn't annoying or spam. It knows what you are looking at and knows the value proposition of that product, and it responds quickly when they are told not interested.
Kaskade refers to them as "prompt, persistent, but mindful." He also says that people often come into the client's offices and want to meet the person they were talking to. He even had one client's customer ask the AI-assisted persona out for drinks.
Generative AI for AI-powered assistants
The conversations today on generative AI relate to helping humans create content. But it's different for Conversica. Conversica has removed the human and given generative AI to the digital assistant. By doing so, they reckon to be turning the output into actionable conversations.
That can be scary for some companies because no one is fact-checking or improving the generated content. Conversica eases these fears by building models using customers' domain-specific information. So the AI uses content the customer has already approved.
The firm also gives customers the option of tapping into the public large language models (LLM) for information outside the brand's domain. For example, if someone asks what the AI assistant's favorite drink is. But even this has guardrails around it, Kaskade said, because of the bias built into these LLMs.
The ethics of AI-powered digital assistants
Conversica's perspective is to advocate for transparency, making it clear that people are interacting with an AI-assisted persona. There are even recommended best practices for persona titles. However, Kaskade reveals that 99% of customers elect not to declare it is an AI-assisted persona.
Even though Kaskade and his company's ethics committee advocate for transparency, they also believe it won't matter. It hasn't been an issue for them when people have discovered the truth, he says:
People think that the effectiveness of the exchange will drop. I think the younger generation that's purchasing now/has the purchasing power actually appreciates the fact that they can get their answers quickly and effectively. And they don't really care because they don't want to talk to a salesperson. They actually just want to get the information in by themselves. So there's a lot occurring that I think kind of supports this move to AI automation that is ethically in line with, I think, what the general population would probably start to align to and agree with.
He cites an anonymous enterprise client - a big global brand - that made its AI assistant a VP persona. That, he argues, demonstrates the trust and level of effectiveness of the assistant.
There are two things that he believes make AI-powered digital assistants work well and help ensure the company is ethically responsible.
First, they are purpose-built for revenue teams. They aren't trying to be everything to everyone, which limits the scope of what the AI needs to be proficient:
For any brand, when you're interested in learning about a product and want to eventually buy it, there's certain behaviors around that that are pretty consistent across all buyers. That makes it easy for us to be very accurate at determining intention in that process. And then, over the years, we've learned really well how to take the next-best action based on intent because we're so focused. And so the combination of being good at reading intention in the field that we operate and taking the right action gives us a 98% accuracy level which is unheard of in this AI industry, and it's what caused me to be really excited by the business.
The second thing is that they keep humans in the loop. Conversica has a team called the Training and Audit Desk. If the AI gets confused, it triggers an alert to the training team, who will look at it, determine the intent, and tell the AI the next best action. They will then use this data to help retrain the model in case something will help make the model more accurate.
Then there's the auditing. Every day, the audit team looks at random responses between the AI and the prospect or customer to determine if it was the right intent and next-best-action. Kaskade says they are seeing failure rates of 2 out of 100. However, even these failures are given to clients as leads to review and decide what to do with.
The two activities - training and auditing- help keep the models fresh and accurate, according to Kaskade:
For systems that you want to have a superior exchange experience with, you should always have humans in the loop. Always. I call it supervised AI. But there's never in my mind a case where you shouldn't have that. It may get minimized because the system becomes so effective. It's self healing and self learning, but you always need to have oversight and governance.
My interest in Conversica started because it announced an integration with Salesforce Marketing Cloud last week at the Connections conference. My excitement around AI, including generative AI, lies with its ability to perform marketing activities that empower marketers to spend more time on strategic and creative activities. So things such as setting up campaigns, creating the assets, and automating the workflow.
In the case of the Salesforce integration, Marketers can invoke the digital assistant through journey builder custom actions and deliver offers, product recommendations, and other campaigns using the Marketing Cloud Personalization (Salesforce Interaction Studio).
And it's not just email. Conversica also supports SMS, messaging apps like WeChat and WhatsApp, live voice, and even digital signage (although there isn't a lot of demand for it, so they haven't brought it to market). Kaskade said they support any use case where the client wants to go, but they are focusing much effort on the chat channel because the demand for great chat experiences has exploded since OpenAI announced Chat GPT.
Marketing Cloud is just the latest integration. Last March, it announced an integration with the Salesforce Automotive Cloud. Kaskade says that a significant part of Conversica's success is based on an integration with Salesforce Sales Cloud, which it did many years ago. There are also integrations with HubSpot, Gainsight, Microsoft Dynamics, and others.
Conversica has been playing with LLMs since the beginning of 2020. They worked with Microsoft and OpenAI (GPT2 at the time), Meta, and Google. They even tried to create their own LLM. All the hard, painful work (Kaskade's word) did pay off because it helped them develop LLMs for their customers (on a much smaller scale, of course). If the stats are accurate - 24x return on investment, 40-50% conversion rates, and 10x pipeline growth when using Conversica revenue digital assistants - then conversational AI (with human oversight) may be the path to the future.