If 2023 was the year of Generative AI hype, 2024 (and probably 2025, 2026 and maybe 2027, if we’re being honest) will be the year(s) where we have to figure out if this shiny, incredibly expensive (and morally dubious) new technology actually has real world applications that are worth the price of admission.
Where procurement is concerned, given the mixture of factors creating new and interesting pain points for the sector, the benefits of generative AI can’t arrive soon enough. These factors range from the skills shortage to an overall strategic shift in the nature of sourcing and ongoing political (and economic) tensions, all of which have the potential to create profound pain points for industry leaders.
The risks of generative AI investment are not to be taken lightly, despite the allure of greater efficiency and lower costs. Specifically, risks range from AI “hallucinations” and unreliable unreliable outputs, to issues of IP ownership, inherent bias, and cybersecurity threats. However, the benefits are, some would argue, well worth the risk. This is especially true if the nature of generative AI deployment is well suited to the task at hand.
Here are three levels to which Generative AI can be used throughout the procurement process:
1. Shallow deployment
This is probably the least intrusive and easiest to execute in terms of Generative AI deployments. A shallow deployment might simply provide a superficial gateway to ChatGPT or other external GenAI tools.
Results mirror those obtained by directly interfacing with ChatGPT outside of the procurement application. Solutions could be used for rephrasing communications, generating early-phase ideas, and other simple, language-related tasks.
2. Contextualised and optimised deployment
The next level optimises the large language model underpinning a generative AI deployment. It does this by training it using specific data relevant to the procurement function or company as a whole.
Essentially, a contextualised Generative AI deployment involves furnishing the GenAI model with information derived from data within the solution. This can include sources of information like raw sales data and compliance documents. The solution employs prompt engineering to enhance output quality. User prompts undergo repackaging or reformatting before being forwarded to ChatGPT or another external GenAI tool. A single user request may trigger multiple GenAI prompts managed concurrently.
3. Tailored deployments
Lastly, a much more in depth deployment involves a solution that utilises its own LLM (a pre-trained language model that has been further refined) to enhance outputs and dramatically reduce the risk of hallucinations. Using AI in this way can also allow for a hybrid approach, in which the solution employs either the internal LLM or an external one depending on the specific use case.
Overall, tailored generative AI solutions offer a range of benefits, including improved performance, reduced risks, increased efficiency, enhanced contextual understanding, and greater flexibility, making them valuable tools.
