As an entrepreneur and technologist working in the Enterprise Software field for more than 30 years, I've been watching the growing AI landscape for years now, and we are still at the early times of enterprise AI. This is especially true for the smaller guys. We all have a sense of positive impact of AI now, but CEOs, CAIOs and other CxOs are all facing roadblocks while trying to get AI in place and working right for them.

Let's have a look at some of those hurdles together.

People

  • Employees fear that AI will soon replace them. That makes them stick to what they know.
  • Employees don't have the knowledge and skills to use these tools in an effective way. Training costs are hard to justify for small businesses.
  • AI-savvy talent sourcing is hard; keeping them is harder. Small businesses generally can't afford it.

Strategy

  • Pinpointing the right use case to implement and the right process to automate requires knowledge and practice. Not all use cases have a meaningful impact in business.
  • Many companies don't have a clean AI roadmap outlining what they're doing and what they need to do.
  • Most big bang projects fail because internal processes are not well-documented, or documentation is not up to date. AI projects are no exception.
  • Companies are in general resistant to change, and AI does involve some dramatic changes.

Data, Rules, Safety, Alignment, Regulation

  • Company data are spread in many systems, unstructured and noisy. Worse, there isn't good enough quality data to be used effectively. This is especially true for smaller or younger companies.
  • Poor data quality, and all the issues regarding privacy (e.g., GDPR) and other regulatory issues highly slow down the process.
  • AI adds a new attack surface. Businesses need to protect the data used for training, the AI models and systems they build, and protect the whole system.
  • Role-based access to data in large enterprises is crucial, but preventing AI from leaking data is hard and unproven. There is still a lot to do in this field.

Money and Tech Headaches

  • It's still unclear whether AI truly saves or spends money. The technology, manpower, security, and maintenance of AI costs money, and they're not always easy to predict or provision.
  • For most small businesses, building their own AI from scratch is too costly and labor-intensive with uncertain results. Perhaps they need to rely on packaged AI.
  • Getting new AI to work with existing computer infrastructure without breaking things or opening security holes is a real challenge.
  • Moving from pilot project to real-world scale AI is a giant leap — better to start with that in mind.
  • The AI algorithms themselves are susceptible to being tricked or hacked (jailbreaking), messing up their results or creating security flaws.

Ethics, Legislation, and All That Tape

  • Everyone worries that AI will be biased — in fact all the known LLM models have some bias. It makes results unreliable in some contexts.
  • There is no proven way to explain how the AI results are related to the problem and the input (XAI is a very hot topic these days). That makes AI inadequate for precision work or autonomous decision-making in many critical scenarios.
  • The regulatory landscape on AI is still taking shape and is a work in progress. This cloudy sky holds many firms in health, finance, energy, and telecom sectors on edge to invest big.

Final Words

While the AI landscape is still changing and taking shape, there are a lot of opportunities. The barrier to entry for newcomers is low and everybody can participate. I urge every company to rapidly draft an internal roadmap and start implementing AI — even on small scales and directed to a fragment of their workforce. Accumulating experience and practice are the most valuable assets you may keep for the future of your business.