Latest / CXOTalk / AI Agents for Small Business: What You Need to Know | CXOTalk #878
Transcript
- 0:00What does it really mean to democratize AI?
- 0:03Andrew Wesbecker, founder of Powered By, reveals how SMBs can
- 0:08deploy 24/7 AI agents. Powered By Builds AI agent
- 0:14solutions for small to medium sized businesses across several
- 0:19agent modes, including voice, phone, text, e-mail and digital
- 0:26avatars. AI agents automate tasks that
- 0:30historically have been performed by humans can.
- 0:33You give us some concrete examples.
- 0:36Let's say, for example, a hotel was answering a bunch of phone
- 0:41calls at their various different properties to book reservations
- 0:44from their clients or to check on reservation status or
- 0:49concierge service. That typically is done by a
- 0:52human today that is picking up the phone and handling that
- 0:55interaction with those customers or prospective customers today.
- 0:59Based on the advances made in voice agent technology, where
- 1:03the state-of-the-art is such that it's almost
- 1:06indistinguishable between speaking to a human and speaking
- 1:09to an AI agent. The hotel, for example, could
- 1:12front all of their customer facing phone interactions with a
- 1:17astonishingly human like AI agent that could, for example,
- 1:22answer the phone, ask customer question or answer customer
- 1:24questions, book reservations on specific dates for specific
- 1:28rooms, upgrade activities, or offer concierge services.
- 1:32Or if the organization of the hotel wanted to have an AI agent
- 1:37perform in room dining activities, the of hotel guests
- 1:40inside the room could dial A number.
- 1:42Instead of speaking to a human, they would speak to an AI agent
- 1:45that would take their order and process the order and have it
- 1:48delivered. I think everybody wants to know
- 1:51how well does this work. The sophistication of AI agents
- 1:56across the modes that I mentioned, voice AI, e-mail AI,
- 2:00text AI, is to the point where it's about 85 to 90% away from
- 2:07being indistinguishable from how a human behaves specifically in
- 2:12voice. In on our website, we have a
- 2:14couple of examples where you can sample this.
- 2:17The voice quality tricks your brain into thinking as to
- 2:20whether or not and questioning yourself whether or not you're
- 2:23actually speaking to a human or in fact, speaking to an AI
- 2:26agent. I tried the demo on your
- 2:28website. I was really impressed.
- 2:30You sort of get engaged in this conversation with this person
- 2:37that's not really a erson. Our intellectual property is is
- 2:40using the foundational AI LLM models and then building all of
- 2:45the sophistication around text to speech, speech to text, real
- 2:49time translation, real time transcription to make the
- 2:53interactions with a user as human like as possible.
- 2:58So you begin with the LLM and then you're layering various
- 3:04other technologies on top of it in order to make it useful as an
- 3:09agent for your customers. Yeah, Architecturally, just a
- 3:12quick summary, at the bottom of our architecture is the LLM and
- 3:16we dynamically switch from whatever LLM that best suits our
- 3:20need, whether it's low latency, high quality, best voice
- 3:23quality, lowest cost. We can dynamically switch from
- 3:27GPT 4 dot O Mini to Claude Sonnet 37 to Meta Llama 2, all
- 3:33very dynamically on top of that. Then you have a proprietary
- 3:36prompt layer that allows us to train the agent based on their
- 3:41specific use case. We build a training module that
- 3:44gives the AI agent in whatever mode that it's operating in,
- 3:48whether it's voice, e-mail, text, or digital avatars, the
- 3:52instructions that guides its behavior.
- 3:56And then we give it a knowledge base, a series of articles,
- 3:59spreadsheets and other documents that we use, a technology called
- 4:03RAG that the agent in real time, in addition to the prompt, can
- 4:08inspect this knowledge base to inform the types of responses
- 4:13that it has to customer questions and the type of
- 4:16behaviors it carries out with the lowest latency possible.
- 4:20On top of the prompt layer, then we implement the natural
- 4:24language processing layer, where we have a variety of open source
- 4:28and in source or proprietary tools that we use for text to
- 4:32speech, speech to text, real time translation currently
- 4:36across 15 languages and real time transcription on top of
- 4:40that. Then we have the action layer
- 4:42that takes all of the underlying prompt model, the knowledge
- 4:46base, the text to speech, the speech to text, all of the
- 4:49natural language processing capabilities.
- 4:52And then we automate the behaviour of the various
- 4:55different agent through the modes that we want to it to
- 4:58operate in, whether it's, as mentioned, voice, phone, e-mail,
- 5:02text or digital avatars. What are the real benefits and
- 5:07the real value for small to mid sized companies?
- 5:13It's a variety of things. We really focus today on
- 5:16customer engagement are wedge into our customers is all around
- 5:23reducing your total cost of engaging with customers through
- 5:28the various different means that you engage with customers today
- 5:31while also improving the customer experience.
- 5:36And so our primary value proposition is today you employ
- 5:40humans to do a lot of menial repetitive tasks that can
- 5:45otherwise be automated through AI agents in voice, phone, text,
- 5:49etcetera. That then allows you to
- 5:52reallocate your human staff members to higher priority tasks
- 5:58or higher value priorities that you currently don't have the
- 6:02bandwidth for. You can then allocate those
- 6:04humans to focus on those activities and have the lower
- 6:08level or repetitive tasks managed by the AI agents.
- 6:12All that can be done 24/7 in the most human like manner possible.
- 6:19What is the process for implementing these kinds of
- 6:26agents as you were just describing?
- 6:29We approach each customer, delivering them a turnkey AI
- 6:33agent solution. We do all of the design, the
- 6:36building, the systems integration where the the
- 6:38customer has very little effort in the actual implementation and
- 6:44management of these various different AI agents.
- 6:47And that's sort of our key differentiation is that there
- 6:50are a lot of vendors in the AI agent marketplace, which today
- 6:56primarily sell dev tools to their customers for them to
- 7:00build their own AI agents. For small to medium sized
- 7:04businesses that don't necessarily have the internal
- 7:07development expertise or AI know how to use and onboard those
- 7:13developer tools. We take that out of the picture
- 7:16and we build all of our solutions custom and deliver it
- 7:19as a turnkey solution based on the customer's requirements and
- 7:24business needs. We do that for you and we take
- 7:27your requirements, we take your work flows, we take which modes
- 7:31you want to operate in and build that and deliver that to you
- 7:35without you needing to hire additional staff members,
- 7:39additional AIML engineers that that a small to medium sized
- 7:42business would wouldn't otherwise have.
- 7:45So there's a configuration process, you speak with the
- 7:48customer, you spend time, you understand their business, and
- 7:51then you configure your solutions to adapt to what will
- 7:55be most beneficial for them. Exactly.
- 7:59You mentioned hospitality earlier, hotel chain, as an
- 8:03example of where these agents might be useful, can you give us
- 8:08some other use cases? Let's take an auto dealership.
- 8:11Hypothetically, they have 35 locations across the state of
- 8:14Florida. They employ over 200 sales reps
- 8:18responsible for selling the various different automobiles.
- 8:22Let's say that dealership also has 2500 leads in their CRM of
- 8:28all of the individuals that have either come into the dealership,
- 8:33visited their website and have opted in to receive marketing
- 8:37and sales information from that dealership.
- 8:41We have a module called Outbound AI that instantaneously you
- 8:46could contact all 2500 of those leads in your marketing database
- 8:51through phone calls, through emails and through texts to
- 8:56promote, for example, a spring sales event at the dealership.
- 8:59Let's say that dealership has 15% off MSRP of their latest
- 9:04Mercedes models. They can have a voice call where
- 9:07in the most human like astonishingly human like manner.
- 9:11Me as a customer who I've opted in to receive marketing
- 9:14information from that dealership will pick up the phone and on
- 9:17the other end of the phone we'll, we'll sound like a human
- 9:20and say hi. This is Dave Frankel from
- 9:22Mercedes of Tacoma, WA. I understand that you've been in
- 9:26the dealership before and wanted to let you know that we're
- 9:28offering a spring sales event where we're having 15% off of
- 9:31our MSRP. I'd like to see if you'd like to
- 9:34schedule a time to come into the dealership and look at the
- 9:36latest models and that agent then will handle all of the
- 9:40booking reservation and scheduling for that human to
- 9:45come into the dealership to drive any vehicle or all of the
- 9:50vehicles of the customers choosing that same workflow
- 9:54operates also in e-mail and text.
- 9:58Now the biggest caveat here of using this today is the AI agent
- 10:05cannot call on individuals that are not opted in to receive
- 10:12sales or marketing information from the business.
- 10:15How do you ensure that you are frankly not spamming customers?
- 10:22We have built the Powered by outbound AI module to adhere to
- 10:27all of the privacy regulations associated with the 1991 Act,
- 10:32the TCPA Act, which prohibits bots or AI agents from cold
- 10:38calling on individuals that haven't opted in to receive
- 10:43marketing and sales information from a get in business that is
- 10:46not permitted use for our product here at Powered By.
- 10:51Andrew, it sounds like the key piece here is the dynamic
- 10:56interactive aspect because of course, personalization by
- 11:03e-mail or less so by phone calls, but even to some extent
- 11:06by phone calls has been around in CRM systems for a long time.
- 11:10But you're going far beyond personalization alone.
- 11:14Exactly. And the reason for that, it's
- 11:16not only personalized, but the textbook definition of an AI
- 11:20agent is that it interacts with its environment and learns how
- 11:25to become better at its task with the more interactions that
- 11:30it has with memory. With AI agent memory, we can
- 11:34also establish a long tail of that AI agent knowing who that
- 11:41individual they are speaking to or emailing with or texting
- 11:45with. Just like a human would.
- 11:47The AI agent can have context into the conversations that it
- 11:51has had previously. And so with this memory and this
- 11:55kind of engagement, if you have, for example, a natural sounding
- 12:00voice, that conversation from the point of view of the
- 12:04customer will seem very comfortable and very intuitive.
- 12:08That's the intent. So it's really the packaging of
- 12:11all of these elements together that creates the unique
- 12:15experience that you've been describing.
- 12:17Yeah. And one unique element is that
- 12:19we can stack the agents, which is you can have a voice
- 12:22conversation with a astonishingly human like voice
- 12:27agent that can then carry over to the same person in e-mail,
- 12:32you know, nick@whatevercompany.com, who
- 12:35then emails you to schedule appointment to then ensure that
- 12:39you've got the right documentation signed and then
- 12:41carry that into a text for further correspondence and
- 12:45confirmation of appointments. So it's not just I isolated over
- 12:48the vote voice chat. We can stack the agents for a
- 12:52single vertical conversation or single vertical engagement with
- 12:57a given customer. So it goes way beyond the
- 13:00traditional notion of personalization, which compared
- 13:04to this is extremely simplistic. Exactly.
- 13:07And that's because textbook definition is that the AI agent
- 13:11learns over time, and the foundation of these LLMS is
- 13:16machine learning and machine intelligence.
- 13:18And the more interaction it has with a given customer, the more
- 13:22it can improve and tailor its interactions with that given
- 13:26customer. Can you give us another example
- 13:29or use case of AI agents in practice for small and medium
- 13:35companies? Certainly.
- 13:37So let's say for example, I'm a health clinic and I have, I'm a
- 13:41paediatrician and I have 15 locations across the state of
- 13:45Texas. And today I book appointments
- 13:49either online or over the phone. There are a variety of questions
- 13:55that a customer is going to have, Either they're an existing
- 13:58patient or a new patient. When they pick up the phone and
- 14:01they call the pediatrician's office, they want to know what
- 14:05availability do you have? What insurance do you take?
- 14:08What's the mechanism for getting approved by insurance to, to
- 14:12book an appointment or book a, a, a medical doctor's
- 14:16appointment for my child? All of that today is very human
- 14:20intensive. And if you multiply that out
- 14:22across a pediatrician's office, a large pediatrician's office
- 14:25that may have 10 or 15 offices, they may have 5 or 6 or 7
- 14:31receptionists or reservations managers that are responsible
- 14:34for fielding all of these calls today, right now, that can be
- 14:39automated completely in a HIPAA compliant manner such that the
- 14:43AI agent can ask the individual's name, their child's
- 14:48name, ask what type of insurance they have, ask for the insurance
- 14:51ID, ask for when they want to schedule a booking appointment
- 14:56for. All of that can be done.
- 14:59And then stacked from a voice call to an e-mail that sends a
- 15:02calendar invite to confirm the booking and then a text reminder
- 15:06a day before to confirm or reconfirm that that appointment
- 15:09is occurring. Well, let me give you another
- 15:11example. Yesterday, I had an unfortunate
- 15:13experience trying to get a refund from a major airline that
- 15:18changed something on my reservation that I felt that I
- 15:20was owed a refund. I try to do so on their website
- 15:24through their chatbot. Their chat bot is all script
- 15:27based so you can't ask any questions with a lot of context.
- 15:31It only follows a script. Additionally, when you want to
- 15:34move that to a phone call, I was in a phone queue, an IVR phone
- 15:38tree for about 20 minutes before I could speak to somebody to
- 15:43obtain my refund. With an AI agent you could do
- 15:46that instantaneously 24/7 with phone and voice quality that is
- 15:53remarkably human like, that can act and process very
- 15:58sophisticated requests on the behalf of.
- 16:01The customer, we touched on this earlier, but again, how well
- 16:06does it work? That's the magic question from
- 16:10the customer standpoint. The more information that we get
- 16:14from the customer on their workflow, if it's a voice agent,
- 16:19often times we ask for recorded calls that they've had with
- 16:23customers in the past to diagnose where a good call,
- 16:28what's the, what's the example in workflow of a good customer
- 16:31interaction. We also asked for examples of
- 16:34where you've had perhaps unfortunate interactions where
- 16:38customers hung up or been angry for the service that they've
- 16:41been for. Provided we can extract that
- 16:43information, obviously sanitize it and anonymize it to train how
- 16:49the AI voice agent can operate in a manner to optimize the good
- 16:55and and de optimize the bad. Additionally, we also ask for
- 17:00FAQs and is there a certain amount of questions commonly
- 17:03that you're asked? What are the responses commonly
- 17:06to those questions? We build that into our knowledge
- 17:09base. We also ask for a bunch of
- 17:11documents, whether it's a menu, a catalog of services or
- 17:16products, an inventory that cannot you know, it can be a
- 17:19dynamic inventory in a spreadsheet or some sort of
- 17:22service that we can put an API I hook into.
- 17:25It could be a static menu or static inventory.
- 17:29The more information that we have that is accessible today by
- 17:33the human staff member that is currently handling those calls,
- 17:37emails or texts, if we get that information, we form the AI
- 17:42prompt and our knowledge base to make the AI agent as capable or
- 17:48even more capable than a human because it's knowledge base is
- 17:53limitless. So you can adopt the style or
- 17:59tone or cultural approach. We could even say that a
- 18:04particular organization maintains towards its customers.
- 18:09Let's say, for example, we're working with a stylish, hip,
- 18:12millennial or Gen. Gen.
- 18:14Z focused hotel chain. We want the voice interaction,
- 18:18the e-mail interaction to be more representative of that
- 18:21brand, of that style of being hip and cool.
- 18:25However, if we're dealing with a very staid insurance company,
- 18:28that's by the book, you know, no humor.
- 18:30We only do what we say we should do in the instructions.
- 18:33We can tailor the behavior of the agent based on the brand or
- 18:39the guidance of the the company that is using the.
- 18:42Agent. What about ROI measurement?
- 18:44How can an organization measure the results of an AI agent?
- 18:52The first area is productivity increases and the reason for
- 18:56that is an AI agent operates 24/7 tirelessly.
- 19:01No breaks, no coffee breaks, no lunch, no stepping out for a
- 19:05walk. So you're able to do things with
- 19:08full utilization of an AI agents time 24 by 7.
- 19:13So what typically is performed by humans in a nine to five or
- 19:18an 8 to 4 type of environment, we can spread that to 24 hours.
- 19:24And so you're able to perform customer engagement activities
- 19:29in a given day almost twice the amount of customer engagement
- 19:33activities that you would otherwise be able to do with a
- 19:37team of of human staff members. So the first first point of ROI
- 19:41is productivity increases in terms of the time those
- 19:46activities can be performed in a given day.
- 19:49The second is revenue growth. This also unlocks a series of
- 19:53new opportunities to sell, market and upsell your customers
- 19:58at as evidenced in the auto dealership example where you can
- 20:03launch phone calls, emails and texts automatically and all
- 20:07simultaneously to all 2500 leads in your auto dealership lead opt
- 20:13in database to convince them to come in the to the dealership to
- 20:17visit. For their new spring sales
- 20:19event. And the third is cost savings,
- 20:21which is you can reallocate your human staff members to, you
- 20:25know, other priorities or let's say you have a shortfall in in
- 20:29certain areas of your business and you can't hire more people
- 20:34to make up for that shortfall. You can reallocate those
- 20:37individuals that are currently doing these menial, I should say
- 20:40labor intensive tasks on the phone, on e-mail, on text.
- 20:44You can reorient those individuals to those on server
- 20:48or lower, serve lesser served areas and have the AI agent
- 20:52performed the labour intensive menial tasks on phone, voice,
- 20:58e-mail, SMS text, etcetera etcetera.
- 21:01And in fact, what you're really saying is the measures are not
- 21:06much different from the ROI analysis that you would place on
- 21:11people doing their jobs. A. 100%, 100%.
- 21:14Andrew, you mentioned earlier that you're trying to
- 21:18democratize AI agents. Can you elaborate on that?
- 21:23What did you mean? The magnitude of the importance
- 21:26that SM BS have on the US economy is enormous.
- 21:30We believe that they should have the right to access the same
- 21:36innovative AI agent technology that the largest Fortune 100,
- 21:40Fortune 500 companies are currently implementing right
- 21:43now, but are out of reach to a small auto dealership or real
- 21:48estate agency or law firm or accounting firm because they
- 21:52don't have the AI engineers, They don't have the know how in
- 21:56terms of AI agent technology and most importantly, they don't
- 21:59have the budget. We build custom solutions
- 22:03designed to fit SMBs staff resources, their understanding
- 22:09and knowledge of AI agent technology and and most
- 22:13importantly, we operate and deliver solutions affordably
- 22:17tailored to the budget of small to medium sized businesses.
- 22:20That's our mission. Can you offer practical advice
- 22:23for business owners who want to become more involved using AI
- 22:29agents in their organizations? You have to start small and we
- 22:33really recommend that you look at where your primary use cases
- 22:40would be across the modes that we operate in voice, phone
- 22:43assistance, e-mail, text, digital avatars.
- 22:47We want you to become a fan of AI agent technology and we want
- 22:51to delight your customers and we want your customers to come to
- 22:57you and say, wow, that was a really unique experience with
- 22:59that voice agent that I just spoke to.
- 23:01Or they say I had a very enjoyable conversation with your
- 23:04assistant on the phone, not even knowing that it wasn't a human.
- 23:08Andrew Wesbecker, founder of Powered by Thank you so much for
- 23:12taking time to chat with us. Thank you, Michael.
- 23:14Appreciate it. We'll see you next time.