Dabbling in AI projects that go deeper than casual ChatGPT/Copilot interactions? Here’s why projects stall — and how to avoid it.
The use of LLM-based generative AI tools is seemingly everywhere these days. Even if you haven't used it personally, chances are someone on your team is using it whether you approved it or not.
Many small businesses are using AI in basic ways (things like employees chatting with ChatGPT to answer questions or asking Copilot to retrieve a file). But going beyond the basics? That’s a little less common at the small business level.
And the bigger problem is what happens with the companies that do try to launch something more significant with AI: most projects stall out.
Most Projects Stuck in Concept
One recent report showed that around 50% of AI initiatives have yet to exit proof-of-concept mode. Businesses expect they will see gains from AI; they’re even willing to spend more on it — even without seeing much return as of yet.
There’s a vagueness and uncertainty around AI: what exactly does a business want to get out of it? Experimentation and testing is great, but AI can become a true force multiplier only once an organization decides exactly what they want it to do, how they’ll measure progress, and what performance level they need to achieve before exiting concept and testing stages.
Going Deeper Is Technically Demanding
It’s not that AI can’t deliver on its promises exactly. It’s more that implementation and execution are a lot more technically demanding than sometimes advertised.
Businesses get sold something along the lines of:
Broken system → AI magic → unparalleled success
…And that’s just not how AI works.
What’s needed is more like:
Well-functioning system → clean data → AI tooling → force-multiplied success
To make AI work for you beyond the basics, you need a system, process, and workflow that’s already working. Then you need good data you can feed to an AI tool. Then you need the right AI tool.
Another way to look at this: advanced AI solutions don’t fix what’s broken; they simply accelerate what’s already going on. If you 10x a broken process, you just get to those broken results faster and automatically.
Businesses Still Unsure About AI Governance
Governance — decisions about who can use what tools in which ways and how it all gets tracked and enforced — remains a bit challenging for many small businesses. Add in the nuances and complexities of AI and the challenge only grows.
Business leaders appropriately worry about AI projects remaining secure, keeping data secure, and avoiding any compliance violations where applicable. Since AI requires data and now generates unique outputs from that data, governance is extra important yet harder to work through.
The AI Skills Gap
Many AI products are sold as plug-and-play solutions that sound an awful lot like that “AI magic” workflow we shared above.
But the day-to-day reality is different. Organizations still need professionals and/or partners who understand how to work with these AI systems, which still need monitoring, management, troubleshooting, and at times human intervention.
Three Tips to Avoid the Stall
If you’re looking to accomplish more with AI and want to avoid the stall, try these three quick tips. Businesses already doing well with AI tend to execute all three.
1. Tie to Specific Outcomes
First, before you start an AI pilot, determine what exactly you want it to do. This should be something narrow, specific, measurable, and probably a bit boring. Improving accuracy here, saving time there.
Once you know what you’re watching for AI to do, you’re equipped to evaluate whether a pilot is working or needs to be scrapped or retooled.
2. Set Boundaries and Establish Governance
AI tools need to be told what they can and can’t touch, and humans need to know what outputs still need human oversight. Establish these guardrails, and AI can do more with lower risk.
3. Scale Slowly
Only once #1 and #2 are in place are you ready to scale up AI from a pilot to a live system. Do this slowly, not all at once. Prove value somewhere small, then expand from there.
Getting more out of AI often starts with aiming for something smaller but very specific. Need help figuring out what that looks like? Our team is here for you.