Turning Ineffective Chatbots into High-performing Revenue Generators
Autodm AI’s road to quality
At Autodm AI, we are motivated by the firm belief that conversational AI will revolutionize how people interact with companies online, particularly when making purchases. Our conversational AI platform enables businesses to boost sales and marketing in travel, real estate, sporting goods, beauty and other industries with sophisticated and premium product offerings.
After many years of dealing with complex industrial AI at C3 AI, we founded Autodm AI in 2023. Buoyed by the launch of ChatGPT, advancements in LLMs, and growing interest in AI development and adoption from company boards, we anticipated smooth sailing.
Our hopes were quickly tempered. Achieving 50% quality — measured by satisfactory interactions — is straightforward and easily achievable. However, to gain companies’ trust in a product, one must meet a higher standard of excellence, a standard that requires an in-depth evaluation of what makes a conversational AI great as well as the resources and expertise to build an innovative, user-friendly product.
In the last six months, we have developed a multi-agent orchestration with robust AI processing layers, leading us to consistently exceed 90% in quality customer interactions. A year in, our vision has materialized: initial large-scale deployments have not only yielded a 170–240% increase in leads but also resulted in a 35–55% boost in conversion rates.
How the Typical Gen AI RAG Approach is Inadequate for Conversational AI
Our first MVP aimed to enable users to discover a company’s products and services conversationally to facilitate purchasing decisions.
We naively thought that we could crack this problem combining the well-known Retrieval-Augmented Generation (RAG) approach and a smart combination of calls to Large Language Models (LLMs).
As development progressed, the initial AI-generated responses seemed promising. However as we went into customer meetings to demo our product, our design partners were not impressed. They immediately noticed issues: inaccuracies, language inconsistencies, irrelevant content, lack of interaction proactivity. At best, only 50% of the AI responses were satisfactory.
There were also challenges on the commercial front — chatbots have existed for decades, and frustration with their quality is widespread. People have little patience for subpar conversational experiences. If a certain threshold of quality is not reached, there is zero commercial interest.
Surprisingly, most AI companies rely on this “good enough” approach. This popular strategy is a non-starter for customer-facing, enterprise-grade conversational AI.
What about Autodm AI? We went back to the drawing board to develop a better product.
Conversational AI must achieve Human-Level Interaction Quality to Succeed
What makes a quality conversational AI? Over the last year, we have worked with our design partners and clients to provide a more precise definition: quality in conversational AI is defined as the encapsulation of multiple concepts at once:
- Response accuracy and latency
- Resourcefulness
- Knowledgeability
- Proactivity
- Hyper-personalization
- Empathy.
Online customers expect human-level interaction performance, similar to conversations with company representatives in physical stores, via call centers, or through messaging apps.
Conversational AI must do more than just answer questions about products and services. It should provide recommendations, ask clarifying questions, capture intentions, and convert prospects into customers — seamlessly and naturally.
To be effective, customer-facing conversational AI needs to meet 90% of quality standards while leveraging all the aforementioned features — anything less and it’s back to the dark ages of disappointing chatbots.
Unfortunately, reaching such a high bar is difficult, even for top tech companies like LinkedIn, as illustrated in their recent post describing their challenging journey with Generative AI.
The team achieved 80% of the basic experience we were aiming to provide within the first month and then spent an additional four months attempting to surpass 95% completion of our full experience — as we worked diligently to refine, tweak and improve various aspects. We underestimated the challenge of detecting and mitigating hallucinations, as well as the rate at which quality scores improved — initially shooting up, then quickly plateauing.
Linkedin Tech Lead — Juan Pablo Bottaro
But why is it so hard? We identified two key challenges:
- You cannot control the messages customers will send, yet they want and expect a decent response. This means you need to handle a broad range of customer message types, gracefully manage those that are out-of-scope and successfully identify subtleties in phrasing.
- Conversational errors have a compounding effect: the conversations we have seen typically consist of 5–8 interactions (i.e. a message from the user followed by a response from the AI). For a conversation to be satisfactory all its interactions need to be satisfactory, meaning that to reach 90% quality on the overall conversation you are only allowed 1–2% failure on each interaction, a much higher bar.
Mastering Quality Unlocks Sales Conversion and Capturing Rich Insights
To reach 90% minimum quality we spent the last months re-architecturing and re-building the entire product, while at the same time laying the foundations to handle the power of generative AI technology.
At its core, the Autodm AI platform now offers a multi-agent orchestration platform that turns our conversational AI into a semi-independent entity that can have two-way conversations while drawing on various expert skills (e.g., recommend a product, ask questions to clarify intents, answer inquiries) to guide prospects in their purchasing decisions and convert them into customers. Our platform now includes built-in layers to overcome hallucinations, manage scalability, security, data privacy, moderation and meet other critical customer-facing requirements.
This novel approach of combining multiple agents with an enterprise-grade conversational AI has enabled us to broaden the range of capabilities offered by Autodm AI, while at the same time meeting the strenuous quality bar. Our conversational AI relies on AI agents that can play various roles and use different tools, including, for instance, a Sales Representative who collects user information, a Product Expert who recommends and compares products, or an Industry Specialist who acts as a trusted advisor for the concerned products domain.
We are now actively deploying our product to our design partners’ websites to boost their sales conversion and growth with conversational AI. This time, the results speak for themselves. Our conversational AI yields 170–240% more leads and drives a 35–55% increase in conversion rates.
As we collaborate with them to continue configuring the product to their specific engagement and conversion objectives, we’re already uncovering a treasure trove of insights within the conversations; insights which will be used to improve their deployed AI agents. This highlights the beauty of conversational AI for our customers — the sooner they embrace it, the greater the flywheel effect on revenues.
This is only the beginning of the journey for Autodm AI; as we develop hyper-personalized immersive experiences, we are excited to offer buyers the purchasing experience they have longed for, and help companies better nurture their leads and promptly advance them through the sales funnel. If you share our vision of a superior and more effective customer buying experience through conversational AI, or if you have questions or comments, please get in touch!
