Reviewing gen-AI tools over the last two years I’ve come across some issues larger organisations will come up against when trying to integrate them into workflows. Here are my top 10 (with assistance from some of my network on LinkedIn – particularly John Levitt at elvex.
1. The market is VERY fragmented, with most AI tools doing very specific things. ChatGPT is general purpose, but something like Midjourney really isn’t.
2. Each AI tool thus comes with its own Terms and Conditions, which doesn’t really sit well for processing through legal departments in larger organisations. Do you just take a risk on them all? Paradoxically, waiting to assess everything is unfeasible, but so is allowing everything because apps are being released in multiple jurisdictions.
3. Payment problems. Possibly worse still, payment is almost always via credit card, and managing this within a large org creates a headache. Who needs what, and who pays when there’s department overlap?
4. The worker manager divide. There’s a disconnect between the people who could use AI most effectively, and the managers who hold the purse strings. Partially this is a knowledge gap, but there’s another problem…
5. …AI anxiety. The people who may benefit from AI (it could increase productivity or make their job more interesting) understandably may have a reluctance to request it – because of anxiety around it eventually replacing them. This is an empowerment problem.
6. Data security. With security governance difficult to deploy, companies often leak data into training models.
7. Risk in enablement. For instance, when you enable unfettered access to a Large Language Model, will those using it pass off AI work as their own? This in itself demands an internal AI compliance training programme. Gen-AI inevitably raises copyright issues too.
8. Usage reporting. Platforms don’t share basic usage, making Return on Investment and enablement difficult to quantify. Why enable ChatGPT enterprise if you have no concrete data of its internal usage?
9. Access control. Because of fragmentation (point 1) and payment (point 3), managing access on a per user is difficult, principally because most of the SaaS based AI systems have cost effective individual creator tiers.
10. Access to internal enterprise knowledge is a challenge when building a useful internal system like a Custom GPT or Retrieval-Augmented Generation (RAG). Stakeholders may not see it as a priority to grant unfettered access.
It’s not even easy for smaller orgs (under 50 people) to effectively implement gen-AI for similar reasons, but largely because there’s a mind boggling number of solutions on the market – related to point 1.
