|
Nearly all companies are planning to deploy Agentic AI, but only a small share has managed to move these systems into production, highlighting a wide gap between artificial intelligence experiments and actual business deployment, according to a report by Ness Digital Engineering.
The report describes the gap between proof-of-concept (POC) and production as Death Valley, where many Agentic AI projects struggle to progress beyond initial trials. It stated, "Approximately 99 per cent of companies plan to put AI agents into production, but only about 9-14 per cent have fully done so, indicating a significant implementation gap. This can be called the "Death Valley" between POC and production". The report said the implementation gap is linked to concerns around trust in probabilistic solutions based on generative AI. It also added, "Some reasons for this are a loss of trust in probabilistic solutions using Generative AI and alack of visible difference in user experience in their daily operations". It said traditional financial institutions tend to focus on using Agentic AI for back-office efficiency and IT development, while digital-native fintech companies are more likely to build the technology into customer-facing experiences and product design. At the same time, the report said the Agentic AI market in financial services is projected to reach USD 33.26 billion by 2030. This is increasing pressure on company boards and investors to demonstrate that AI investments are producing results. The report argues that companies need to take a more structured approach before scaling Agentic AI. This includes assessing the business domain, existing technology infrastructure, processes, data readiness, API readiness and the willingness to explore new ways of improving user experience. It said success in building Agentic AI applications will be an important priority over the next 4-5 years. It also highlights the problem of employees having to switch between multiple systems and windows to complete their work. The report recommends using an existing messaging or interaction window as a single interface, with the AI system gathering relevant information from different systems and using APIs based on business rules and user permissions. The report also argues that companies should judge AI projects based on whether they improve the experience of different groups of users, rather than simply measuring whether an individual task has been automated. Cost is another issue for companies looking to scale Agentic AI. The report recommends tracking Large Language Model (LLM) token consumption, AI subscriptions, cloud costs and other AI-related expenses at the application level so companies can compare spending with the expected return on investment. Overall, the report argues that companies need to move beyond simply experimenting with AI agents and focus on building systems that can work reliably across complete business processes. (ANI)
|
