AI Stocks: Why Feeding Chatbots Proprietary Company Data Is Key

Generative artificial intelligence has yet to produce material revenue growth for most software companies, which has been disappointing investors in AI stocks. One Wall Street analyst calls it the “Gen AI waiting game in software.”

X

The good news is that cloud computing giants and software companies are providing enterprise customers with more tools to master generative AI technologies. To build gen AI software applications, large companies need to harness proprietary data to train and deploy AI models.

The rub is that there are potential risks for companies in commercializing gen AI apps. Gen AI chatbots are known to “hallucinate,” or make up incorrect answers when they can’t find an accurate response. And no company wants a public-relations nightmare from a chatbot running amok.

In March, Alphabet‘s (GOOGL) Google became ensnared in a controversy over its Gemini-powered gen AI chatbot. The challenge for large companies is developing gen AI apps that leverage clean proprietary data and produce reliable answers.

AI Stocks: Proprietary Company Data Key

That’s where “vector search” comes in. It’s emerging as a key way to unlock the value of proprietary company data. Many companies aim to integrate “co-pilots,” or conversation chatbots, into enterprise software applications.

“Today you have public chatbots — ChatGPT, Gemini — that tell you what’s going on with the public internet data they’re trained with,” Ganapathy “G2” Krishnamoorthy, VP of analytics at Amazon Web Services, told IBD in an interview. “But they don’t do a great job with company data and applications because they don’t have that information.”

“The big change that is happening are chat-based interactions. It’s a very compelling, interactive user experience. It’s a capability that’s relevant for every application and now companies want to bring it to every application. But that means you have to make these AI models work with your data to answer questions.”

For example, AI vector search should play a big role in e-commerce.

Cloud Titans Target AI Vector Search

Krishnamoorthy said outdoor gear maker REI could use gen AI to leverage its product catalog. An REI customer could tell a website chatbot that he or she is planning a summer hiking and camping trip to Mount Ranier in Washington. The chatbot would respond with a list of recommended equipment, options to buy, and top-rated items.

AWS, which is part of Amazon.com (AMZN), is by far the biggest provider of cloud computing services.

Microsoft‘s (MSFT) Azure cloud computing unit and Google Cloud Platform also offer vector search services to customers. Also, Microsoft is the biggest investor in gen AI startup Open AI, the leader in large language models.

“The AI vector search market is nascent,” said JPMorgan analyst Pinjalim Bora in an April 17 report. “The main ask from enterprise (customers) is how to empower LLMs (large language models) with proprietary enterprise data, while at the same time preserving the privacy and governance of the data.”

Generative AI technologies create text, images, video and computer programming code on their own.

Large language models understand the way that humans write and speak. They allow users to interact with AI systems without the need to understand or write algorithms.

AI Stocks: How Vector Search Works

The models process “prompts,” such as internet search queries, that describe what a user wants to get. LLMs require training data for specific tasks. They’re made of neural networks — or mathematical models that imitate the human brain — that generate outputs from the training data.

In enterprise use cases, the AI model training process creates vector “embeddings” in databases that provide the relevant search context needed to produce reliable answers, not hallucinations. Vector embeddings are a way to convert words and sentences and other data into numerical representations that capture their meaning and relationships.

“Large language models understand the world in numbers,” Krishnamoorthy said. “So you’re taking company information — customer records or a product catalog — and representing it as a string of numbers called vectors.”

He added: “While a generative AI application relying solely on a foundation model will have access to broad real-world knowledge, FMs on their own can be prone to hallucinations. Vector databases can complement FMs to provide an external knowledge base for generative AI applications, helping ensure they provide trustworthy information by grounding the model on facts. This is done via a process called Retrieval Augmented Generation.”

“RAG optimizes the output of a foundation model, so it references an authoritative knowledge base — often a vector database — outside of its training data sources before generating a response. Organizations can connect FMs to their own internal data stored as embeddings in a vector database for more relevant, context-specific, and accurate responses.”

Amazon offers vector services through Bedrock, its AI development platform.

Public Companies In Vector AI Space

A wide array of tech companies aim to make it easier for enterprises to harness proprietary data for gen AI apps.

Data management software firms such as Snowflake (SNOW) and privately held rival Databricks provide vector search services. Salesforce (CRM) offers DataCloud services.

Also, other public companies to watch include Oracle (ORCL), MongoDB (MDB) and Elastic (ESTC).

“Vector search is a requirement for AI workloads to scale over time, and we believe the market is at an inflection point as demand for AI applications continues to grow,” said Jefferies analyst Brent Thill in a report.

Thill added: “We believe that some platform companies (Microsoft, Oracle, Snowflake, Elastic) are well-positioned should the industry trend toward hybrid search — combining text-search and vector search, traditional databases and vector databases — in the future.”

Wave Of Vector Database Startups

In addition, a new wave of vector database startups also aim to bring about the gen AI enterprise revolution. They include Pinecone, DataStax, Chroma, Weaviate, and Qdrant. Datastax is among Google’s partners.

IBM (IBM), chipmaker Nvidia (NVDA) and Databricks are investors in startup Unstructured. It helps companies clean up enterprise data for gen AI applications and use by foundation models.

So far, most companies have been using gen AI for chatbot support in customer service centers, transcription and summarization, and document drafting and code generation. But many industry-specific apps, such as fraud detection in financial services, are still in early stages.

“Many companies are still searching for the killer app with gen AI,” Greg Kogan, Pinecone’s VP of marketing, told IBD. “Some have it easier with a customer service bot but most companies have to work a little harder. When you decide what data the AI models have access to on demand, it opens up the opportunities significantly. But economics also matter. Vector databases scale up and support big AI workloads.”

Research firm Gartner says that 2% of enterprises had adopted vector databases in some capacity as of late 2023. It expects that to grow to 30% by 2026.

Further, cloud computing giants are adding more vector search capabilities to their database offerings.

At AWS, Krishnamoorthy says companies may not actually need a separate, special-purpose vector database. Instead, they can add vector indexing capabilities to existing databases.

Vectors Target ‘Unstructured’ Data

In the long-run, vector technology will play a role in search across massive data sets of semi-structured and unstructured data — images, social media posts, emails, audio files and sensor data.

“Traditional databases are optimized for storing data such as tables, documents, and key-value pairs,” added JPMorgan’s Bora. “However, with advancements in AI and natural language processing, increasing quantities of semantic vector data have required new repositories. Vectors allow for storing the intrinsic meaning of unstructured content, such as images, videos and natural language in a machine-readable format.”

Meanwhile, many companies are scrambling to launch generative AI pilot programs. For investors in AI stocks, the bottom line is that deploying production-ready gen AI apps commercially will take time.

“While gen AI’s impact on IT is going to be material, any material revenue reacceleration due to gen AI spending in the enterprise market is more likely a 2025 event,” said Evercore ISI analyst Kirk Materne in a report.

Follow Reinhardt Krause on Twitter @reinhardtk_tech for updates on artificial intelligence, cybersecurity and cloud computing.

YOU MAY ALSO LIKE:

Want To Trade Options? Here Are The Basics To Get You Started

IBD Digital: Unlock IBD’s Premium Stock Lists, Tools And Analysis Today

Learn How To Time The Market With IBD’s ETF Market Strategy

Monitor IBD’s “Breaking Out Today” List For Companies Hitting New Buy Points

Market Breaks Expectations On Iran Fears; Here’s What To Do Now