Four Custom AI Chatbot Solutions to Boost Company Value

Reid McCrabb

6 min read

April 25, 2024

Four Custom AI Chatbot Solutions to Boost Company Value

While Language Models are the headliners for today’s AI innovation, other AI technologies are breaking through at roughly the same time. Notably, image generation, computer vision, and recommendation systems.

2023 provided some incredible demos for these emerging technologies. But, it was just the first inning for AI in the workforce. 2024 is poised to be a big step forward in production-ready applications, efficiency-unlocking integrations for businesses, and a new paradigm for how humans work. This article will explore how innovative companies use custom AI chatbots to increase efficiency.

Automation

Replacing the work done by humans with Artificial Intelligence has long been a theme in sci-fi. Today, it’s the core focus of every Fortune 500 R&D team.

The key differentiator between the sci-fi predictions and today is AI was thought to replace redundant blue-collar work first, and knowledge work last. Instead, the cost of knowledge is being driven to zero, with physical labor jobs in a shortage.

With the breakthrough of language and diffusion models, text and image generation has become basically automated. While image creation is more nuanced, text is required in every single business on the planet. If you or someone who works for you is writing redundant text, you are not operating at optimal efficiency, speed, and margins thanks to advancements by leading AI solution companies.

Automation

Many companies are fine-tuning, prompt engineering, and experimenting with LLMs. The potential gains from properly implementing these models is massive. But these gains do not all require custom code. You can start by simply combining tools and defining workflows.

One hacky automation Linkt is using internally is Fireflies AI x a template x Claude. Here is how it works.

  1. Fireflies AI joins all of our Google Meets and records a transcription of everything said and who said it, this is the main way our company communicates, and we use it for hiring, prospecting, contract negotiation, and our project update meetings.
  2. We download the transcript from Fireflies AI and drag and drop it into the Claude chatbot.
  3. Go to Google Drive and select a template we have created, this includes a weekly update email template, proposal template, contract template, and more. The items in these documents that change are displayed in brackets, like [date] or [estimated length of project]. We then copy and Paste the template into Claude
  4. With both a transcript and a template in the context window for Claude, Prompt it the direction, i.e “fill in the proposal template based on the conversation we held with the client.”
  5. As long as the key points were mentioned in the transcript, 9-10 bracketed variables are filled in, with nuanced, detailed variables pulled from the call.

This process saves our company a few hours a week, and also improves our operating margins as we don’t need to hire someone to do it – we just rely on two robots!

We are currently building out the software to automate this process, and are imagining many other automations that could help our agency run on autopilot while improving the quality of our work!

Chatbots

With LLMs, the most popular use case has been using a chatbot, similar to the likes of chatGPT, but with a company’s internal knowledge base to query answers from, leveraging an advanced AI system..

This technique is called Retrieval Augmented Generation or RAG for short. The technique uses the best of both worlds from AI and searches, using data to ensure accuracy while using the LLM to generate the most relevant responses in real-time.

RAG

The clearest example of this done successfully is Klarna, Linkt wrote about Klarna’s $40 million value add here. In summary, Klarna’s mundane task of answering user questions is now automated with chatbots, and the chatbots are performing even better than humans were – reducing repeat inquiries by 25% and slashing the average query resolution time from 11 minutes to just 2.

Other repetitive tasks are being automated by LLMs, including content generation, email management, translation, summarization, code generation, and more.

Much of the value is finding itself accruing to the big technology companies. Subtle changes are appearing in Google Suite, Microsoft Teams, and Adobe Acrobat (i.e. Adobe Firefly’s image generation.) Linkt aims to help SMBs realize these same unlocks, and ride the waves of this new paradigm.

Recommendation Systems

A recommendation system has become a requirement for any company that has a consumer-facing platform. The clearest example of how valuable a recommendation system can be is looking at TikTok. The entire basis of the company is figuring out a custom AI algorithm to hook in users. While this strong algorithm, especially for goofy dance videos, is questionable ethically, it is without a doubt the way the world is heading. The recommendation system has become the staple of value for social media companies Instagram and X (formerly Twitter) as well.

Tik Tok’s Recommendation Algorithm

But it’s not just social media platforms taking advantage of this. Any company that wishes to provide a personalized user experience is optimizing self-improving algorithms. From scheduling a trip to matching with a co-founder, taking user data, and interpreting it with AI to provide an improved experience and faster time to the desired outcome is an incredible business advantage.

Recommendation systems can also be used internally. From job matching to route matching, AI use cases like these make huge differences in company efficiency.

Computer Vision

Companies interacting with the real world often rely on human vision, recording, and analysis. We are now living in a dystopian world where that is no longer required. Computers can reason based on the pixels they take in, understand what is happening, predict what will happen next, and take action to best operate under the circumstances.

Take self-driving cars as an example. Other use cases like medical image analysis, traffic monitoring, sports analysis, or analyzing visual data on a computer screen i.e. stock charts, user interfaces, etc. can be impacted.

Computer Vision

How to Proceed

The general advice for how companies should think about integrating AI, is to start with a hacky system around pre-built tools. Much like how Linkt is downloading transcripts from Fireflies AI and plugging it into the context window of Claude with a template, you can find creative systems to get more efficient.

Once you’ve found a way to be more efficient, automate the process. Use API’s, Retool, prompt engineering, and other simple tactics to take it one step past your hacky setup.

If you are building customer facing i.e a chatbot this can work too. As you build out proof of concepts, aim to prompt engineer into product market fit, and later fine-tune for scale.