A global poll of 1,600 small and midsize business leaders points to a widening gap between AI ambition and execution, according to SAS. The SAS SMB AI survey, highlighted on the company’s site, frames the problem as a “readiness–reality” gap and urges firms to benchmark on an AI maturity scale before pressing ahead.
That diagnosis tracks with what many teams feel day to day. Enthusiasm is high. Production wins are rarer. Budgets tighten when data quality, security reviews, and change management land on the same project plan.
Product velocity among foundation model vendors isn’t helping. Google DeepMind now lists multiple Gemini variants tuned for cost, speed, or multimodal reasoning. Choice can be healthy, but in lean environments it drives decision paralysis and stalls procurement cycles.
What the SAS SMB AI survey actually measured
SAS describes a global survey of 1,600 small and midsize business leaders that compares AI aspirations to concrete action. The write-up points leaders to an AI maturity scale and a self-serve readiness assessment designed to map next steps (SAS). The headline takeaway is straightforward: intent is widespread; delivery lags.
That framing matters because it highlights where to focus scarce resources. Not on generic proofs of concept, but on specific decisions where better predictions, scoring, or routing will cut time or cost. In other words, start where the data is already good enough and the impact shows up on a dashboard the CFO reads.
It also hints at a cultural hurdle. Many SMB leaders still see AI as a strategic bet rather than the next phase of analytics. Treating it as a series of small, measured process changes makes it easier to fund, govern, and sustain.
Why SMBs stall: data debt, unclear ROI, and risk
Most midmarket roadblocks sit upstream of any model. Fragmented customer records, event data without timestamps, and spreadsheets that don’t match make training brittle and monitoring messy. Fixing that isn’t glamorous, but it’s the bottleneck.
Governance adds weight. The NIST AI Risk Management Framework lays out controls that buyers increasingly expect: clear purpose, dataset lineage, bias checks, testing, and ongoing oversight. Even light-touch versions take time and people. Skipping them makes pilots fast and deployments fragile.
Then comes the market itself. With new model families, features, and pricing shifts landing often, teams worry about lock-in and regret. DeepMind’s Gemini catalog shows how quickly options change, from lighter variants for throughput to heavier ones for reasoning. That is useful, yet it pushes smaller buyers to extend evaluations and postpone big bets.
Finally, ROI stories are still uneven. Leaders want payback windows and measurable lifts. Without baselines and clean A/B tests, many projects overpromise and then bog down in exception handling.
A practical playbook to close the readiness gap
The SAS SMB AI survey underscores an obvious but underused move: scope around a single decision. Pick one that repeats, touches revenue or risk, and has available data. Then automate the prediction and keep the human-in-the-loop for final calls until the metrics prove out.
Make measurement boring and precise. Define the one or two outcome metrics that matter before any model is trained. Instrument them in production. If the lift doesn’t show up in a month, fix the data or stop the work.
Cut data debt early. Stand up a basic catalog, write down feature definitions, and agree on retention and access rules. These steps speed audits and reduce rework later.
Right-size the stack. Aim for simpler models when they meet the bar, and weigh latency and token costs against margins. Keep pilots small, then scale only when the value clears a hard hurdle rate. A short vendor shortlist also helps; too many demos add noise and delay.
Use outside frameworks, not just sales decks. The OECD’s guidance on SME digitalization captures common constraints and can help boards calibrate pace and investment. Combine that with a lightweight version of NIST’s controls to keep risk reviews sane.
What sports can teach: SAS, Liverpool FC, and decision-first AI
SAS spotlights a partnership with Liverpool FC that aims to bring advanced data and AI into football operations and business strategy (SAS). The appeal isn’t the tech; it’s the decisions. On the pitch that might mean translating performance signals into selection or recovery choices. Off it, smarter fan engagement and ticketing tactics can move real revenue. The lesson for smaller firms is the same: pick a decision, wire the data, measure the result, then repeat.
High-profile cases can feel distant from small shops. Still, they show what “production” looks like when AI is tied to outcomes, not demos. They also show why governance matters. Sports teams face scrutiny from fans and regulators alike; they need clear methods and records when choices are questioned.
What this means for midmarket buyers in 2026
Expect more models, more claims, and more pressure to “do something.” The SAS SMB AI survey is a useful counterweight. It reminds leaders to sort wants from wins, to fund data fixes early, and to measure value in weeks, not quarters.
One final point on choice: rapid model iteration will continue. DeepMind’s growing Gemini family is proof enough. That’s fine. The durable advantage won’t be the model of the month; it will be a repeatable way to turn decisions into measurable gains, with clean data and clear guardrails. For more on this, see bloomberg.com and nytimes.com.
Related reading: Copilot • OpenAI • Productivity & AI