Generative AI for enterprises: ADP, IKEA launch tools

ADP, IKEA and time to market

Businesses find generative AI lends itself to rapid prototyping, with pilots emerging in a matter of weeks. Other deployment drivers include widespread cloud adoption, which provides a ready-made onramp for cloud-based generative AI services, and the efforts of large language model (LLM) vendors to offer enterprise-oriented features such as specialized plugins.

As a result, enterprises are going to market with production tools much faster than with traditional AI.

"I think with the generative technology that's really come into play in the last 12 months, we're being able to iterate on ideas a lot quicker than we could before," said Naomi Lariviere, chief product owner and vice president of product management and shared services at ADP. "Five years ago, it took us a while to get [an AI idea] underway."

With generative AI, ADP was able to gain an understanding of vendors' LLM technologies, create pilots and bring a product to market within 12 months -- a pace Lariviere called "unheard of in this space."

IKEA followed a similar course, moving from tracking technology developments to releasing a product in a matter of months.

I think with the generative technology that's really come into play in the last 12 months, we're being able to iterate on ideas a lot quicker than we could before.

Francesco Marzoni, chief data and analytics officer at Ingka Group, IKEA's largest franchisee, said the path to IKEA GPT began in May 2023, when the group began exploring OpenAI plugins that arrived with the launch of ChatGPT 4. OpenAI's plugins, which let ChatGPT use third-party services, support a broader range of use cases than was possible before.

The next milestone was OpenAI's November 2023 announcement of GPTs, tools that let developers customize the chatbot for their enterprise-specific needs. At that point, "we directed our focus towards the development of IKEA GPT," Marzoni said.

Development continued apace from there. OpenAI's online GPT Store launched January 10 and IKEA GPT AI Assistant became available less than a week later, he noted. The custom version of ChatGPT acts as an intermediary, connecting with IKEA's backend APIs to retrieve relevant information, execute tasks or respond to inquiries as customers shop, Marzoni said.

Deploying generative AI for enterprises

Once a technology is ready for production deployment, the next task is spreading it across the user base. IKEA uses the GPT Store, where developers list custom versions of ChatGPT, to put its tool in the hands of consumers.

As the number of users grows, IKEA will consult partners to make sure its generative AI application can scale accordingly.

"We are leveraging our Google Cloud Application Load Balancer to optimize increased demand for our application," Marzoni said. The OpenAI platform, however, is responsible for optimizing the ChatGPT model and addressing fluctuations in demand to support IKEA GPT, he added.

IKEA GPT's distribution is currently limited to ChatGPT Plus users in the U.S. market. But the organization plans to release the tool in additional markets during 2024. ChatGPT Plus is a paid subscription plan for ChatGPT.

ADP's deployment approach, in contrast, involves integrating generative AI into products such as Benefits and the Wisely digital pay offering plus platforms such as ADP Workforce Now and ADP Next Generation HCM. ADP Assist is currently integrated into "select ADP products" and activated for early release clients, according to the company. ADP plans to eventually roll out the tool to all clients.

The idea is to place ADP Assist in an HR employee's day-to-day workflow. For example, the tool can help practitioners flag missing tax identification numbers when they are processing payroll taxes on the ADP payroll dashboard, Lariviere said.

"What we really wanted to do is bring the data and insights into where they are working so that they don't have to … hunt throughout the system to get that information," she said.

Making it easier to identify payroll anomalies improves productivity and, in the case of missing tax ID numbers, helps businesses avoid state and federal penalties, Lariviere said.

Working with AI models

ADP Assist employs both external and proprietary LLMs. ADP generally uses LLMs that are not pre-trained or fine-tuned with ADP's data, according to the company. Those models are hosted in the public cloud, but in ADP's private instance within an AWS and/or Microsoft Azure infrastructure.

ADP, however, does fine-tune and pre-train some models, hosting them in-house for highly specific types of responses, the company said. For instance, it uses internally trained and tuned embedding models to build a retrieval-augmented generation (RAG) framework over its compliance Q&A. RAGs are used to retrieve data from external knowledge sources to improve the quality of GenAI responses.

ADP's models use the company's internal knowledge management materials as a primary source when responding to client inquiries about key compliance issues, the company noted.

John Moore is a writer for TechTarget Editorial covering the CIO role, economic trends and the IT services industry.