The middle market has proven AI can deliver.
You have AI in your business. You are not alone. According to the U.S. Chamber of Commerce, 89% of small businesses now use AI in some capacity - up from 36% in 2023. The RSM Middle Market AI Survey, based on 1,030 senior leaders at companies between $30 million and $10 billion in revenue, found that 86% have integrated AI into operations. And here is the number that makes everyone feel good: 97% are satisfied with their AI investments. Over half - 54% - say AI has exceeded their ROI expectations.
Feels like a victory lap, right?
Here is the number that should keep you up at night: only 17% of those same organizations are pursuing enterprise-wide AI transformation. Only 36% have AI fully embedded across their core processes. The overwhelming majority has proven that AI can work in a pilot, in a single department, on a narrow use case - and then stopped.
The middle market has collectively solved the wrong problem. We spent two years asking “does AI work?” The answer is clearly yes. The question nobody asked is “how do we make it compound?”
The Satisfaction Trap
The RSM survey exposes something uncomfortable. When asked why they are satisfied with AI, the top three reasons are having a clear strategy (41%), choosing the right technology (40%), and having sufficient infrastructure (38%). Notice what is missing: measurable business outcomes at scale.
Respondents are measuring what is easy: process efficiency (40%), productivity and time savings (38%), decision quality (36%). They are not measuring revenue growth, competitive differentiation, or innovation - the things that actually compound. As the RSM report puts it, organizations are “optimizing for what is easy to prove, not what is required to scale impact.”
This matters because the fundamental economics of AI have gone in the wrong direction for transformation and the right direction for experimentation. The price of GPT-4-level capability has dropped from roughly $30 per million tokens in early 2023 to under $1 today - a 30x reduction in three years. Mistral released its 1-trillion-parameter Large 4 model today at $1.36 per million input tokens, roughly one-third the price of Claude Opus 5.5 and one-seventh the price of GPT-6 Astra. Open-weight models now account for 45% of all traffic on model aggregators like OpenRouter.
Cheaper AI means more pilots. But cheaper AI does not automatically mean more transformation. It means you can experiment faster, fail cheaper, and - unless you change your approach - stay stuck in the same pilot loop at a lower cost.
The Real Barrier Isn’t Technology
BCG’s 2026 Applied AI Index, surveying 1,300 executives across 20-plus sectors, found that nearly 50% of companies now generate measurable value from AI. That is up from almost none two years ago. But BCG also found that AI spending has more than tripled - from 1% of revenue to 3.3% - and 80% of that spending sits outside the enterprise IT budget.
The gap between investment and impact is an organizational gap, not a technology gap. BCG’s research led to their 10-20-70 model: 10% of the effort goes to the algorithms, 20% to technology and data, and 70% to changes in people, organization, and processes. The companies that perform well on both strategic clarity and applied AI generate five times as much value from AI as companies that don’t.
The RSM data confirms the same pattern. The top barriers to scaling AI are not model quality or access to GPUs. They are data quality (34%), security and privacy concerns (30%), legacy system integration (28%), and talent gaps (28%). These are operating-model problems, not technology-selection problems.
And the workforce dimension is stark: 85% of leaders acknowledge that executive enthusiasm for AI outpaces employee buy-in. Eighty-eight percent expect their workforce composition to change fundamentally within two to three years because of AI. Yet the RSM data also shows that only 4% of organizations point to employee upskilling as a priority investment area.
The technology is ready. The organization is not.
The ROI Data That Actually Matters
This is not an argument against AI. It is an argument for being more specific about where you deploy it. The most recent deployment data from dozens of small and mid-size businesses shows exactly which workflows generate outsized returns.
A service business in Pennsylvania deployed a lead follow-up agent and saw response time drop from 4.5 hours to 2 minutes, lead-to-appointment rate jump from 12% to 34%, and monthly revenue from leads go from $14,400 to $39,600. Monthly cost: $350. ROI: 72x.
A 30-person e-commerce company replaced two part-time support reps ($52,000/year) with a Claude-powered agent handling 73% of tickets autonomously. First-year cost: $18,400. Net savings: $33,600. ROI: 280%.
A dental practice deployed a scheduling and reminder agent. No-show rates dropped from 15% to 6%. Recovered revenue from filled appointments: $90,000 per year. Agent cost: roughly $22,000 per year. ROI: 220%.
Across all tracked deployments, the average return is $3.70 for every $1 invested. Average time to positive ROI: 60 to 90 days.
The pattern in every case is the same: one high-volume, rules-based workflow. Clear inputs and outputs. A human-in-the-loop escalation path. Measurement from day one. Not a grand strategy. Not an enterprise-wide platform. One workflow, proven, then the next.
The Path From 17% to Enterprise Transformation
The organizations that break out of pilot purgatory do three things differently.
First, they stop asking whether AI works and start asking which specific workflows produce the highest return. They use ROI data - real data from their own operations, not vendor benchmarks - to decide where to invest.
Second, they treat AI adoption as a change management problem, not a technology installation. The 10-20-70 model is not a metaphor. Seventy percent of the effort goes to people, process, and organization. If you are not spending most of your time on those three things, you are building a portfolio of pilots, not a transformation.
Third, they build governance early. BCG found that companies with all six AI controls in place generate three times as much agentic AI value as companies with just one. Governance is not a brake on speed - it is the thing that lets you move faster without breaking something.
The middle market has done the hard work of proving AI works. The next phase - making that success compound into enterprise-wide value - is a different kind of challenge. It requires strategy, not shopping. Process redesign, not API integration. Workforce readiness, not another pilot.
You have the tools. You have the data. You have the proof that AI delivers. The question now is whether you are ready to do the organizational work required to turn isolated wins into sustained advantage.
That is exactly the gap Black&Tan Labs exists to close. Our Discovery engagement helps mid-market companies identify the workflows where AI actually compounds, build the governance and measurement frameworks to prove it, and create a roadmap from pilot to production. Not another vendor pitch. A strategy. If you are tired of running pilots that never scale, that is where we start.
Enjoyed this? New essays on AI, operations, and building in contrast - delivered to your inbox.
Subscribe on Substack →