AI STRATEGY
and true AI agents. The more deterministic the task – the more transactional and non-strategic – the more you find deterministic systems work well. When it’ s more strategic, agents can help you be more creative in solving complex tasks.
I’ ve seen overpromise and underdeliver on actual AI products. I call this the AI capability overhang – AI has gotten so much better, but products built using AI are lagging behind the technology’ s progress. Companies have AI strategies they want to pursue, but the application space is quite limited. I think as technology providers, if we are really ahead of the curve, we have to do a better job of bringing the level of our products up to the true capability of the AI progress that we’ ve seen today. And I don’ t think it’ s really happened yet.
Q. WHAT ARE THE BIGGEST CHALLENGES ORGANISATIONS FACE WHEN ADOPTING AI-DRIVEN AUTOMATION IN PROCUREMENT AND HOW DOES LIGHTSOURCE ADDRESS THESE?
» You have to start with personal productivity. If people in an organisation can’ t use AI in their individual tasks, it’ s very hard to imagine organisation-wide AI adoption. There’ s a mandate to find AI technologies and make them useful, but IT teams don’ t always have incentive structures built in to be adventurous. If something goes wrong with AI and you have a data leak, you lose your job. But if you’ re six months late to the party, you were just being appropriately cautious.
“ What this means for our product is that we have to meet the moment. We’ re lucky we’ ve built a flexible system that can include metadata like HTS codes and tariffs as part of total cost of ownership analysis”
SPENCER PENN, CO-FOUNDER AND CEO, LIGHTSOURCE
Our responsibility at LightSource is making AI useful and applicationoriented. If you tell someone to use ChatGPT to make their job easier, you’ re asking procurement people to suddenly have this totally new skill: being creative about technology use. That’ s very different from the actual job itself.
Our job as tool providers is to understand both the technology space and the domain, and marry those together. We don’ t boil the ocean on all AI capabilities, but focus on real pain points faced by direct materials teams and work backwards to figure out the right AI technologies. If you get ChatGPT or Claude, you need to be
120 July 2025