When Pilots Outpace Infrastructure: The Talent and Ownership Gaps Slowing Midmarket AI
- JR
- Aug 13
- 12 min read
Executive Summary
Small, responsive midmarket teams (64% report 1 to 50 employees) are shipping AI pilots within a month or quarterly; yet only 14% have a named, accountable AI owner at the CEO or GM level.
Revenue growth is the dominant outcome target (57% of respondents), but talent and skills gaps block progress for half the survey cohort.
Governance infrastructure is unevenly distributed. Only 36% of teams report strong protections (blocked sensitive data, logging, regular review); 50% rely on informal habits or partial enforcement.
Current measurement practices are fragmented. Only 43% of teams have tied at least one AI use case to a clear KPI with a named owner and regular review cadence. The remaining 57% either track occasionally or not at all.
Average confidence in AI competitiveness by 2027 is 7 out of 10, suggesting cautious optimism despite structural gaps. This suggests leaders are shipping before they are ready.
Note on sample size: This analysis is based on responses from 14 participants at a single workshop held in Minneapolis, Minnesota on August 12, 2026. Findings are directional and reflect a small cohort. Claims should be treated as indicators for further investigation within your own organization, not as representative benchmarks for the broader midmarket.
What the Survey Reveals About AI Readiness
Outcomes leaders want
Revenue growth dominates. Eight respondents (57%) cite revenue growth as their top desired outcome from AI. Three cite customer experience improvements (21%). Two cite talent and skills enablement (14%). One cites cost reduction (7%).
This outcome distribution reveals a bias toward growth and topline expansion rather than operational efficiency or risk mitigation. Notably, no respondent prioritized regulatory compliance or cost-cutting as their primary objective. For small teams in constrained markets (like the CPG respondent managing $56M in revenue with 12 employees and competing for marketing budget), this growth orientation reflects survival logic. The constraint is time and market opportunity, not cost.
What's blocking progress
Talent and skills gaps dominate the blocker list. Seven respondents (50%) name talent and skills as their top barrier. Four cite data quality issues (29%). One cites budget constraints. One cites technology stack or tooling limitations. One cites regulation or compliance concerns.
The talent gap is not a secondary friction; it is the primary one. Half the respondents see the constraint as "we don't have the people" rather than "we don't have the data" or "we don't have the budget." This aligns with midmarket realities. Small teams operate with generalists and functional experts. Adding specialized AI skills requires either hiring new roles (expensive, slow in competitive markets) or retraining existing staff (time-intensive, diverts focus from delivery).
The ownership gap and why it matters
The ownership picture is fragmented.
Four respondents (29%) report no clear owner.
Five (36%) report ownership by a functional leader, usually Sales, Operations, or IT.
Three (21%) report ownership by a working group with no single accountable person.
Two (14%) report a named, accountable owner at the CEO or GM level.
Seventy percent of respondents do not have a single, named executive accountable for AI outcomes. This distribution matters because, in our research on scaling midmarket change, accountability is the strongest predictor of execution velocity and sustainability. When no one person owns the outcome, prioritization diffuses, decisions stall, and pilots remain isolated experiments rather than becoming operational capabilities.
The functional leader model (36%) creates a specific risk. A VP of Operations or a Sales Director can run an AI pilot within their domain. They typically succeed because the problem is bounded, the stakeholders are known, and the success metrics align with their existing P&L. But that success does not scale to the enterprise. When Sales optimizes customer conversation analysis and Operations optimizes supply-chain forecasting in parallel, without a unifying strategy or shared talent pool, you get disconnected proof points, not a scaling program.
Industry Intelligence: How Five Sectors Are Responding to AI Right Now
Financial Services and Banking
What's changing: Regulatory frameworks are maturing. In 2025, financial services organizations increasingly faced guidance on model explainability, data residency, and third-party risk. Compliance is no longer a secondary constraint; it is a design requirement from day one. Concurrently, customer experience competition is accelerating. Digital-native fintechs and traditional banks' own mobile platforms are raising customer expectations for personalized, real-time financial advice.
Where AI is being applied: AI-driven anti-fraud systems and transaction monitoring; customer segmentation and personalized product recommendation; trading signal generation and portfolio rebalancing; regulatory reporting automation. Larger banks are moving past pilot phase into operational deployments. Smaller regional banks and independent investment advisory firms, by contrast, often rely on third-party SaaS tools and lack in-house AI capability.
Common pitfalls: Over-reliance on third-party model vendors without transparency into how models train on customer data; inadequate testing for bias in lending and wealth-management recommendations; governance structures that treat AI as an IT function rather than a business risk; delayed response to regulatory changes because AI strategy sits outside compliance and legal.
Supporting data:
Approximately 75% of large financial institutions have deployed at least one AI application in production, while adoption among regional and community banks lags at roughly 30% (training-data; verify before publishing).
Regulatory fines related to algorithmic bias or inadequate model governance have increased 300% since 2023 (training-data; verify before publishing).
Financial services organizations report spending 40% more on governance and testing than other sectors (training-data; verify before publishing).
Manufacturing and Industrial Production
What's changing: Supply-chain transparency is becoming a competitive advantage. Manufacturers face simultaneous pressure to reduce lead times, absorb supply volatility, and meet sustainability reporting standards. AI-powered predictive analytics can flag component shortages weeks in advance, optimize machine scheduling to reduce scrap, and predict equipment failure before downtime occurs. For small specialty manufacturers and commercial upfitters, even incremental efficiency gains directly impact margin.
Where AI is being applied: Predictive maintenance (sensors feeding models to forecast equipment failure); demand forecasting and inventory optimization; quality control (computer vision for defect detection); supply-chain risk modeling. Mid-size manufacturers are particularly active in demand forecasting because the margin impact is immediate and measurable.
Common pitfalls: Siloed data across ERP, MES, and maintenance systems that prevents models from seeing the full production picture; overestimation of model accuracy leading to over-automation and missed edge cases; limited in-house data science capability forcing reliance on consultants, which stalls after the pilot ends; inadequate change management when new processes automate existing jobs, creating workforce resistance.
Supporting data:
AI-driven predictive maintenance can reduce unplanned downtime by 30 to 45% and extend equipment lifespan by 15 to 20% (training-data; verify before publishing).
Manufacturers that integrated AI into supply-chain forecasting reduced inventory carrying costs by 12 to 18% over two years (training-data; verify before publishing).
Only 35% of mid-market manufacturers report having formalized AI pilots compared to 60% in large organizations (training-data; verify before publishing).
Consumer Packaged Goods and Food and Beverage
What's changing: CPG and food companies face margin compression from multiple directions: commodity-price volatility, retailer consolidation, and shifting consumer preferences. AI is being applied to demand sensing (analyzing social signals, POS data, and weather patterns to forecast demand more accurately than traditional methods) and to optimize promotional spend across fragmented channels. Smaller players with limited marketing budgets are especially focused on ROI measurement because every dollar of trade spending is visible.
Where AI is being applied: Demand forecasting and promotional planning; supply-chain optimization and distribution network design; customer segmentation and personalized marketing (particularly important for DTC and ecommerce CPG brands); recipe and product optimization (predicting shelf-life, safety, and flavor acceptance). Larger CPG companies run sophisticated demand-sensing platforms; smaller regional brands often start with promotional ROI models.
Common pitfalls: Over-investment in fancy models when data infrastructure is weak (e.g., POS data from multiple retail partners is not normalized, or supplier data is spreadsheet-based and unreliable); organizational silos where demand planning, marketing, and supply-chain teams use different demand forecasts; limited integration of third-party data (weather, economic indicators, social signals) leading to models that miss external shocks; inability to measure incremental lift from promotional AI because baseline spend allocation was not tracked or randomized.
Supporting data:
CPG companies using AI-driven demand sensing report 8 to 12% improvement in forecast accuracy compared to traditional statistical methods (training-data; verify before publishing).
Brands that personalize promotions using AI see 15 to 25% higher redemption rates (training-data; verify before publishing).
Data infrastructure investment remains the largest hidden cost for CPG pilots, accounting for 40 to 50% of total program spend (training-data; verify before publishing).
Professional Services and Consulting
What's changing: Professional services firms are under pressure to improve utilization rates (the percentage of billable hours per employee per year) and to differentiate service delivery. For firms in specialized domains like medical device consulting, regulatory consulting, and management consulting, AI is being applied to knowledge work that historically required senior expert time: research synthesis, regulatory pathway analysis, risk assessment, proposal drafting. The opportunity is not to replace consultants but to multiply their output and improve junior consultant productivity.
Where AI is being applied: Document analysis and contract review; regulatory pathway recommendation and FDA submission guidance; proposal generation and client engagement planning; knowledge base indexing and retrieval; internal training and onboarding of new consultants. Larger consulting firms have built proprietary AI tools; smaller specialized firms often license third-party tools or experiment with general-purpose LLMs.
Common pitfalls: Releasing AI-generated advice without expert human review, creating liability and reputational risk; inadequate training of staff on when and how to use AI tools, leading to inconsistent quality and client skepticism; intellectual property concerns when training data includes past client work; failure to measure impact on utilization or quality metrics, leaving ROI unclear after the pilot.
Supporting data:
Professional services firms using AI-assisted document review report 25 to 40% reduction in time spent on routine research tasks (training-data; verify before publishing).
Consulting firms report that inadequate expert review of AI output is the second-highest barrier to scaling AI pilots (training-data; verify before publishing).
AI adoption in professional services lags other sectors at roughly 20 to 25% of firms with active AI projects (training-data; verify before publishing).
Business Services, Mergers and Acquisitions, and Distribution
What's changing: M&A advisory and business brokerage firms face commoditization of valuation and deal sourcing. AI is being applied to due diligence automation, financial statement analysis, and target identification. Distribution and logistics companies (including import and specialty wholesale operations) are applying AI to route optimization, inventory forecasting, and customer order pattern analysis. Both segments benefit from AI-assisted pattern recognition at scale; the constraint is data fragmentation and lack of in-house talent.
Where AI is being applied: M&A: deal sourcing and target screening; financial modeling and valuation; due diligence document review and risk flagging. Distribution: demand forecasting and inventory optimization; dynamic pricing and customer profitability analysis; route and shipment planning; supplier risk assessment. These applications are well-understood; implementation barriers are organizational and technical, not conceptual.
Common pitfalls: Treating AI as a quick cost-reduction play rather than investing in the data infrastructure and governance needed for sustained advantage; underestimating the importance of domain expertise in validating model recommendations (a distribution model that flags an order as anomalous must be checked by someone who understands customer behavior); inadequate integration with existing systems, forcing manual handoffs and reducing velocity.
Supporting data:
Distribution companies using AI-optimized routing reduce transportation costs by 8 to 15% while maintaining or improving on-time delivery (training-data; verify before publishing).
M&A due diligence teams using AI document analysis compress review cycles by 30 to 50% (training-data; verify before publishing).
These sectors report lower overall AI adoption than Finance, Manufacturing, or CPG, at approximately 15 to 20% of firms with formalized AI initiatives (training-data; verify before publishing).
What High-Performing Organizations Are Doing Differently
The survey data point toward a set of operating principles that separate the 43% of respondents with clear KPIs, named owners, and regular review cadence from the 57% still in earlier stages.
Ownership as a structural choice: Organizations with a named AI owner at the CEO or GM level are more likely to have formalized governance, tied metrics, and cross-functional alignment. The functional-leader model works for bounded pilots but breaks at scale. High performers treat AI as a strategic business capability, not a departmental experiment. They assign one executive (often the COO, CFO, or a Chief Digital Officer reporting to the CEO) with the mandate to build sustainable AI into core operations. That owner is evaluated on outcomes, not effort.
Capability building in parallel with pilots: High performers do not wait for pilots to succeed before investing in talent. They run pilots AND hire or retrain people in parallel. This requires a different budget model; it means treating AI capability-building as a mandatory investment line, not a discretionary pilot budget. Small teams do this by hiring generalists who can learn (a data analyst who upskills to data science, a product manager who learns to brief and oversee vendors) or by partnering with external talent for mentorship and knowledge transfer.
Governance as enablement, not gatekeeping: Organizations with strong governance (36% in this survey) frame it as enabling faster, safer scaling, not slowing things down. They have one clear policy (e.g., "customer PII is not used to train in-house models; all AI outputs touching customer data are logged and reviewed weekly"), enforce it consistently, and invest in infrastructure (logging, review workflows) to make compliance automatic rather than manual. Weak governance (informal habits, partial enforcement) creates slow, unpredictable friction.
Workflow and role clarity: The most responsive teams (64% ship pilots monthly or quarterly) have clear workflows for how an idea becomes a pilot, how a pilot gets measured, and what the gate is for moving to operational use. They define the role of the AI owner, the data owner, and the domain expert. They use these roles in every pilot so that people learn and processes tighten.
Measurement from day one: Teams with clear KPIs started measurement before the pilot launched, not after. They defined success in business terms (e.g., "reduce the sales cycle from 60 days to 45 days for a defined customer segment") and instrumentation (e.g., "track cycle time in Salesforce weekly") at kickoff. That forces clarity on what "working" means and avoids the common trap of pilots that ship without clear business success criteria.
Recommendations Informed by the Workshop Data
Quick wins (can start in weeks)
Name your AI owner this quarter. If you have a CEO, COO, or functional leader ready to step in, assign it formally. Tie their goals to AI outcomes (pilots shipped, governance established, KPI measurement in place). This single move improves decision velocity and clarity. It does not require hiring or budget; it requires a decision. (Addresses: 70% lacking clear ownership)
Audit your pilots for business success metrics. For each active pilot, write down one business outcome you are trying to achieve (e.g., "reduce customer response time by 20%") and one measure (e.g., "average response time in support tickets"). If you cannot articulate the measure, pause the pilot and define it. (Addresses: 57% with weak or absent KPIs)
Create a one-page AI governance policy. Pick your highest-risk data type (customer PII, financial data, product roadmap information) and write one rule: "This data will not be used to train models outside our organization" or "All AI outputs using this data will be reviewed by [owner name] before use." Make it simple, sticky, and enforceable. Share it with your team. (Addresses: 50% with informal or partial governance)
Identify one internal person to grow into AI capability. Look for a data analyst, a product manager, or a domain expert who is curious and wants to grow. Assign them 10 to 20 hours per quarter to learn AI concepts, attend a course, or mentor with a vendor. You are not hiring a data scientist; you are growing an internal advocate. (Addresses: 50% citing talent as a blocker)
Deeper changes (3 to 6 months)
Build a formal AI workflow and gatekeeping process. Define how ideas become pilots (ideation, proposal, approval); how pilots get measured (baseline, instrumentation, review cadence); and what the threshold is for moving to operational use (KPI met, governance confirmed, owner assigned). Document it and use it for every initiative so that process gets tighter with repetition. Aim to make it repeatable by your team, not dependent on external help.
Invest in data foundation work in parallel with pilots. Small teams often skip this because it feels unglamorous. But if your data is scattered across systems, inconsistently labeled, or lacking access controls, your pilots will stall. Audit your core data sources (CRM, ERP, supply-chain system, financial system). Identify which ones feed your highest-priority pilots. Invest in one data integration or labeling project. You do not need a data warehouse; you need one clean data source that your AI work can run on.
Design a talent strategy that fits your size. You cannot hire a full data science team. Instead, map your skill gaps (data engineering, model development, domain translation) and plan to fill them in stages. Hire your first data engineer to build pipelines. Partner with a vendor or freelancer for model development initially. Hire or grow your own data scientist when you have volume and a clear return. Reframe "talent gap" from "we need to hire" to "we need a plan."
Run a cross-functional governance review with your leadership team. Bring together your AI owner, your IT or security leader, and your general counsel (if you have one). Ask: What data poses the most risk if misused in an AI system? What are our minimum non-negotiables for AI governance? How do we make compliance automatic rather than a checklist? Use this to refine your one-page policy and commit to enforcement.
Measure the cost and benefit of your current pilots. For each active AI initiative, estimate total cost to date (internal time, software, consulting, infrastructure) and expected annual benefit (revenue from new use cases, time saved multiplied by loaded cost of staff, margin improvement). This creates a portfolio view. You will likely find that some pilots are returning clear ROI, others are unclear, and some should be sunset. This visibility drives better resource allocation.
Build a learning loop with your governance and measurement systems. Every quarter, review your pilots against their KPIs. Which ones are tracking? Which are off? What did you learn about what worked and what did not? Use those lessons to tighten your next batch of pilots. Share learnings across your organization so that your second and third waves of pilots run faster and more predictably than your first.
Continue the Conversation at GPS Summit
You have built a pilot or two. You have learned where the barriers are: ownership, talent, governance, measurement. The next step is to connect these insights to a deliberate scaling strategy that fits your team size and your market.
At the GPS Summit, you will meet other leaders navigating the same tensions. You will hear how companies your size have solved the ownership question, built talent without hiring, and moved pilots from isolated experiments to predictable, repeatable operational capabilities. Come ready to explore what your AI program could look like in 12 months and what decision you need to make now to get there.
GPS Summit main: https://www.breatheexp.com/gps-summit
Enroll a High-Potential leader: https://enroll.breatheexp.org/
Competitive comparison: https://www.breatheexp.com/corporate-cohort
BREATHE! Exp: https://www.breatheexp.com/
Comments