Gartner: This GenAI App Development Strategy Could Cut Delivery Time by 50%

Eighty percent of generative AI business apps will be developed on existing data management platforms by 2028, reducing complexity and cutting delivery time by 50%, according to Gartner.

Currently, GenAI business applications are developed by integrating large language models (LLMs) with an organization’s internal data, as well as rapidly evolving technologies such as vector search, metadata management, prompt design, and embedding. However, organizations risk adopting “scattered technologies” with longer delivery times and higher costs without a unified management approach, the firm announced during the Gartner Data & Analytics Summit, held in Mumbai last week.

The role of RAG in building more accurate GenAI apps

Retrieval-augmented generation (RAG) — a framework for enhancing the accuracy and reliability of generative AI models — will play a pivotal role in mitigating these issues.

RAG is becoming foundational for deploying GenAI applications, because it offers “implementation flexibility, enhanced explainability and composability with LLMs,’’ Gartner said.

“One of the important use cases of RAG is process improvement and automation of tasks in many business functions such as sales, HR, IT, and data management,” Prasad Pore, senior director analyst at Gartner, told TechRepublic. “Currently, data engineers or data professionals face many challenges while developing, testing, deploying, and most importantly, maintaining complex data pipelines and applications.”

This is because current processes around data management take considerable time and human effort, which Pore said can be reduced using RAG, while also improving productivity. “Also, data governance is complex in nature,” and can benefit from RAG in areas including data discovery, business context generation, and security anomaly detection with log analysis, he added.

Additionally, generative models such as LLMs are static and unaware of the latest information, apart from the data on which they are trained, Pore noted. These models are mostly trained using publicly available data. They can be used for general tasks but are not useful for business/organization-specific tasks because they lack context, he said.

RAG integrates the latest business or organization-specific/proprietary data “and even the latest public data, as context, to the LLM model so that it can achieve the goals such as answering questions, analyzing logs, [and] decid[ing] which action to perform based on the question/input,’’ Pore said.

Types of GenAI business apps

Regarding the types of business apps Gartner is referencing, Pore said there are many use cases and applications of GenAI for various industries and sectors. At a high level, it can be categorized in these three broad categories.

1. Process improvements and automation: For example, enterprise knowledge management, document processing automation, research, software developments and operations, and internal help desk.
2. User experience: For example, customer support automation, chatbots for product related queries, personalized shopping experience, travel assistants, and natural language interface for many IT tools.
3. Insights and predictions: For example, conversational BI and analytics tools, data discovery, augmented data management and business intelligence, automation of traditional BI/analytics, and natural language processing.

3 tips about creating and deploying GenAI apps

When building and deploying GenAI apps, Gartner recommends enterprises consider:

• Evaluating whether data management platforms currently in use can be transformed into a RAG-as-a-service platform, replacing stand-alone document/data stores as the knowledge source for business GenAI applications.
• Making RAG a priority and integrating technologies such as vector search, graph, and chunking, from existing data management systems or their ecosystem partners, when building GenAI applications. Technical disruptions are less likely to occur with RAG technologies, and they are also compatible with organizational data.
• Leveraging metadata and operational data at runtime in data management platforms. This will protect against malicious use, address privacy concerns, and prevent intellectual property leaks.

Read TechRepublic’s recent coverage about generative AI entering the Trough of Disillusionment in Gartner’s Hype Cycle.