AI Agents for Small Business:
Winning in 2025

AI Agents for Small- and Mid-Sized Companies

Michael Krigsman: What does it mean to democratize AI? Andrew Wesbecher, founder of Powered_By, reveals how SMBs can deploy 24/7 AI agents.

Andrew Wesbecher: Powered_By builds AI agent solutions for small to medium-sized businesses across several agent modes, including voice, phone, text, email, and digital avatars. AI agents automate tasks that historically have been performed by humans.

Michael Krigsman: Can you give us concrete examples?

Andrew Wesbecher: A hotel was answering phone calls at their various properties to book reservations from their clients, check on reservation status, or request concierge service.

That is done by a human today, who is picking up the phone and handling that interaction with those customers or prospective customers.

Today, based on the advances made in voice agent technology where the state of the art is such that it’s almost indistinguishable between speaking to a human and speaking to an AI agent, a hotel could front all of their customer facing phone interactions with an astonishingly human like AI agent that could answer the phone, ask customer quest or answer customer questions, book reservations on specific dates for specific rooms, upgrade activities, or offer concierge services.

Or, if the organization or the hotel wanted an AI agent to perform in-room dining activities, the hotel guest inside the room could dial a number. Instead of speaking to a human, they would speak to an AI agent who would take their order, process it, and deliver it.

Michael Krigsman: Everybody wants to know how well this works.

Andrew Wesbecher: The sophistication of AI agents across the modes that I mentioned—voice AI, email AI, text AI—is to the point where it’s about 85 to 90 percent away from being indistinguishable from how a human behaves.

Specifically in voice, and on our website, we have a couple of examples where you can sample this, the voice quality tricks your brain into thinking as to whether or not, and questioning yourself, whether you’re speaking to a human, or in fact speaking to an AI agent.

Michael Krigsman: I tried the demo on your website. I was impressed. You get engaged in this conversation with this person that’s not a person.

Technical Architecture of AI Agents

Andrew Wesbecher: Our intellectual property is using the foundational AI LLM models and building the sophistication around text-to-speech, speech-to-text, real-time translation, and real-time transcription to make the interactions with a user as human-like as possible.

Michael Krigsman: You begin with the LLM, and you’re layering various other technologies on top of it to make it useful as an agent for your customers.

Andrew Wesbecher: Architecturally, here is a quick summary: At the bottom of our architecture is the LLM.

We dynamically switch from whatever L LLM best suits our needs, whether low latency, high quality, best voice quality, lowest cost.

We can dynamically switch from GPT 4.0 Mini to Claude Sonnet 3.7 to Meta Llama 2, dynamically.

On top of that, you have a proprietary prompt layer that allows us to train the agent based on their specific use case.

We build a training module that gives the AI agent in whatever mode it’s operating in, whether it’s voice, email, text, or digital avatars, the instructions that guide its behavior.

We give it a knowledge base, a series of articles, spreadsheets and other documents that we use a technology called RAG that the agent in real time, in addition to the prompt, can inspect this knowledge base to inform the types of responses that it has to customer questions and the type of behaviors it carries out with the lowest latency possible.

On top of the prompt layer, we implement the natural language processing layer, where we have a variety of open source and in-house or proprietary tools that we use for text-to-speech, speech-to-text, real-time translation, currently across 15 languages, and real-time transcription.

On top of that, we have the action layer. That takes the underlying prompt model, the knowledge base, the text to speech, the speech to text, the natural language processing capabilities, and we automate the behavior of the various agent through the modes that we want it to operate in, whether it’s, as mentioned, voice, phone, email, text, or digital avatars.

Business Benefits and Implementation

Michael Krigsman: What are the benefits and the value for small to mid-sized companies?

Andrew Wesbecher: It’s a variety of things. We focus today on customer engagement.

Our wedge into our customers is reducing your total cost of engaging with customers through the various means that you engage with customers today, while also improving the customer experience.

Our primary value proposition is that today, you employ humans to do menial, repetitive tasks that can otherwise be automated through AI agents in voice, phone, text, etc.

That allows you to reallocate your human staff members to higher priority tasks or higher value priorities you currently don’t have the bandwidth for.

You can allocate those humans to focus on those activities and have the lower-level or repetitive tasks managed by the AI agents.

All that can be done 24/7 in the most human-like manner.

Michael Krigsman: What is the process for implementing these agents, as you described?

Andrew Wesbecher: We approach each customer delivering them a turnkey AI agent solution.

We do the design, the building, the systems integration, where the customer has little effort in the implementation and management of these AI agents.

Our key differentiation is that there are vendors in the AI agent marketplace which today primarily sell dev tools to their customers for them to build their own AI agents.

For small to medium sized businesses that don’t necessarily have the internal development expertise or AI know how to use and onboard those developer tools, we take that out of the picture, and we build our solutions custom and deliver it as a turnkey solution based on the customer’s requirements and business needs.

We do that for you, and we take your requirements, we take your workflows, we take which modes you want to operate in and build that and deliver that to you without you needing to hire additional staff members, additional AI/ML engineers that a small to medium sized business wouldn’t otherwise have.

Michael Krigsman: There’s a configuration process. You speak with the customer. You spend time. You understand their business, and you configure your solutions to adapt to what will be most beneficial for them?

Andrew Wesbecher: Exactly.

Michael Krigsman: You mentioned hospitality earlier, a hotel chain, as an example of where these agents might be useful. Can you give us other use cases?

Andrew Wesbecher: Take an auto dealership. Hypothetically, they have 35 locations across the state of Florida. They employ over 200 sales reps responsible for selling the various automobiles.

That dealership also has 2,500 leads in their CRM of the individuals who have either come into the dealership or visited their website and have opted in to receive marketing and sales information from that dealership.

We have an Outbound AI module that instantaneously allows you to contact all 2,500 of those leads in your marketing database through phone calls, emails, and texts to promote a spring sales event at the dealership.

That dealership has 15 percent off MSRP on their latest Mercedes models. They can have a voice call where, in the most human like, astonishingly human like manner, me, as a customer who I’ve opted in to receive marketing information from that dealership, will pick up the phone and on the other end of the phone will sound like a human and say, “Hi, this is Dave Frankel from Mercedes of Tacoma, Washington. I understand that you’ve been in the dealership before, and wanted to let you know that we’re offering a spring sales event where we’re having 15 percent off of our MSRP. I’d like to see if you’d like to schedule a time to come into the dealership and look at the latest models.”

That agent will handle the booking, reservation, and scheduling for that human to come into the dealership to drive any vehicle or the customer’s vehicle of choice.

That same workflow also operates in email and text.

Now, the biggest caveat of using this today is that the AI agent cannot call on individuals not opted in to receive sales or marketing information from the business.

Michael Krigsman: How do you ensure that you are not spamming customers?

Andrew Wesbecher: We have built the Powered_By Outbound AI module to adhere to the privacy regulations associated with the 1991 act, the TCPA Act, which prohibits bots or AI agents from cold calling on individuals who haven’t opted in to receive marketing and sales information from a given business.

That is not permitted use for our product here at Powered_By.

Michael Krigsman: Andrew, it sounds like the key piece here is the dynamic, interactive aspect because personalization, by email or, less so, by phone calls, but even to some extent by phone calls, has been around in CRM systems for a long time. But you’re going far beyond personalization alone.

The Evolution and Capabilities of AI Agents

Andrew Wesbecher: Exactly. The reason for that, it’s not only personalized, but the textbook definition of an AI agent is that it interacts with its environment and learns how to become better at its task with the more interactions that it has.

With memory, with AI agent memory, we can establish a long tail of that AI agent knowing who that individual they are speaking to or emailing with or texting with. Like a human would, the AI agent can have context into the conversations that it has had previously.

Michael Krigsman: With this memory and this engagement, if you have a natural-sounding voice, that conversation from the point of view of the customer will seem comfortable and intuitive.

Andrew Wesbecher: That’s the intent.

Michael Krigsman: The packaging of these elements together creates the unique experience you’ve been describing.

Andrew Wesbecher: One unique element is that we can stack the agents. You can have a voice conversation with an astonishingly human-like voice agent that can carry over to the same person in email, nick@whatevercompany.com, who emails you to schedule an appointment, to ensure that you’ve got the right documentation signed, and carry that into a text for further correspondence and confirmation of appointments.

It’s not isolated over the voice chat. We can stack the agents for a single vertical conversation or single, vertical engagement with a given customer.

Michael Krigsman: It goes way beyond the traditional notion of personalization, which, compared to this, is extremely simplistic.

Andrew Wesbecher: Exactly, and that’s because the textbook definition is that the AI agent learns over time. The foundation of these LLMs is machine learning and machine intelligence, and the more interaction it has with a given customer, the more it can improve and tailor its interactions with that given customer.

Use Cases and Practical Applications of AI Agents

Michael Krigsman: Can you give us another example or use case of AI agents in practice for small and medium companies?

Andrew Wesbecher: Certainly. I’m a health clinic and I have… I’m a pediatrician, and I have 15 locations across Texas.

Today, I book appointments either online or over the phone. A customer is going to have a variety of questions, whether they’re an existing patient or a new patient.

When they pick up the phone and call the pediatrician’s office, they want to know, “What availability do you have? What insurance do you take? What’s the mechanism for getting insurance approval to book an appointment or a medical doctor’s appointment for my child?”

All of that today is human-intensive, and if you multiply that out across a pediatrician’s office, a large pediatrician’s office that may have 10 or 15 offices, they may have five or six or seven receptionists or reservations managers that are responsible for fielding these calls.

Today, that can be automated completely in a HIPAA-compliant manner, such that the AI agent can ask the individual’s name, their child’s name, ask what type of insurance they have, ask for the insurance ID, and ask for when they want to schedule a booking appointment.

All of that can be done and stacked from a voice call to an email that sends a calendar invite to confirm the booking, and a text reminder a day before to confirm or reconfirm that the appointment is occurring.

Andrew Wesbecher: Let me give you another example. Yesterday, I had an unfortunate experience trying to get a refund from a major airline that changed something on my reservation, for which I felt I was owed a refund.

I tried to do so on their website through their chatbot. Their chatbot is script-based, so you can’t ask questions with context. It only follows a script.

Additionally, when you want to move that to a phone call, I was in a phone queue, an IVR phone tree, for about 20 minutes before I could speak to somebody to obtain my refund.

With an AI agent, you could do that instantaneously, 24/7, with phone and voice quality that is remarkably human-like, that can act and process sophisticated requests on behalf of the customer.

Michael Krigsman: We touched on this earlier, but again, how well does it work? That’s the magic question from the customer’s standpoint.

Andrew Wesbecher: The more information we get from the customer on their workflow…

If it’s a voice agent, we often ask for recorded calls that they’ve had with customers in the past to diagnose the example in the workflow of a good customer interaction.

We also ask for examples of where you’ve had perhaps unfortunate interactions, where a customer has hung up or been angry with the service that they’ve been provided.

We can extract that information, sanitize it, and anonymize it to train how the AI voice agent can operate to optimize the good and de-optimize the bad.

Additionally, we also ask for FAQs, and is there a certain number of questions that you’re commonly asked? What are the responses commonly given to those questions? We build that into our knowledge base.

We also ask for documents, whether it’s a menu, a catalog of services or products, an inventory that cannot… it can be a dynamic inventory in a spreadsheet or some service that we can put an API hook into.

It could be a static menu or static inventory. The more information that we have that is accessible today by the human staff member that is currently handling those calls, emails, or texts, if we get that information, we form the AI prompt and our knowledge base to make the AI agent as capable or even more capable than a human because its knowledge base is limitless.

Michael Krigsman: You can adopt the style or tone or cultural approach, we could say, that a particular organization maintains towards its customers.

Andrew Wesbecher: We’re working with a stylish, hip, millennial, or Gen Z-focused hotel chain. We want the voice interaction, the email interaction to be more representative of that brand, of that style of being hip and cool.

However, if we’re dealing with a staid insurance company that’s by the book, no humor, we only do what we say we should do in the instructions, we can tailor the behavior of the agent based on the brand or the guidance of the company that is using the…

ROI and Democratization of AI Agents for SMBs

Michael Krigsman: Agent. What about ROI measurement? How can an organization measure the results of an AI agent project?

Andrew Wesbecher: The first area is productivity increases, and the reason for that is an AI agent operates 24/7 tirelessly, with no breaks, no coffee breaks, no lunch, and no stepping out for a walk.

You can do things with full utilization of an AI agent’s time 24 by 7.

We can spread what is performed by humans in a 9:00 to 5:00 or an 8:00 to 4:00 environment to 24 hours.

You’re able to perform customer engagement activities in a given day, almost twice the amount of customer engagement activities that you would otherwise be able to do with a team of human staff members.

The first point of ROI is that productivity increases in terms of the time those activities can be performed in a given day.

Andrew Wesbecher: The second is revenue growth. This unlocks new opportunities to sell, market, and upsell your customers as evidenced in the auto dealership example, where you can launch phone calls, emails, and texts automatically and simultaneously to all 2,500 leads in your auto dealership lead opt in database to convince them to come into the dealership to visit for their new spring sales event.

Andrew Wesbecher: The third is cost savings, which is you can reallocate your human staff members to other priorities, or you have a shortfall in certain areas of your business, and you can’t hire more people to make up for that shortfall.

You can reallocate those individuals that are currently doing these menial, labor intensive tasks on the phone, on email, on text.

You can reorient those individuals to those unserved or lower-served, lesser-served areas and have the AI agent perform the labor-intensive, menial tasks on phone, voice, email, SMS, texts, etc.

Michael Krigsman: You’re saying the measures are not much different from the ROI analysis that you would place on people doing their jobs.

Andrew Wesbecher: 100%.

Michael Krigsman: Andrew, you mentioned earlier that you’re trying to democratize AI agents. Can you elaborate on that? What did you mean?

Andrew Wesbecher: The magnitude of SMBs' importance on the US economy is enormous.

We believe that they should have the right to access the same innovative AI agent technology that the largest Fortune 100, Fortune 500 companies are currently implementing, but are out of reach to a small auto dealership, real estate agency, law firm, or accounting firm because they don’t have the AI engineers.

They don’t have the know-how regarding AI agent technology, and most importantly, they don’t have the budget.

We build custom solutions designed to fit SMBs’ staff resources and their understanding and knowledge of AI agent technology. Most importantly, we operate and deliver solutions that are affordably tailored to the budgets of small- to medium-sized businesses. That’s our mission.

Michael Krigsman: Can you offer practical advice for business owners who want to become more involved using AI agents in their organizations?

Andrew Wesbecher: You have to start small, and we recommend that you look at where your primary use cases would be across the modes we operate in—voice, phone assistance, email, text, digital avatars.

We want you to become a fan of AI agent technology, and we want to delight your customers, and we want your customers to come to you and say, “Wow, that was a unique experience with that voice agent that I spoke to.”

Or, they say, “I had an enjoyable conversation with your assistant on the phone, not knowing that it wasn’t a human.”

Michael Krigsman: Andrew Wesbecher, founder of Powered_By, thank you for taking the time to chat with us.

Andrew Wesbecher: Thank you, Michael. Appreciate it. We’ll see you next time.