What AI Realistically Looks Like at Your Company in 3 Years: The Honest Planning Horizon for Mid-Market Executives

Executive Summary

• The 36-month view is knowable, but not from vendor decks. The gap between what software companies sell and what actually shows up inside 300-person companies is the single most misleading input to mid-market AI planning. The corpus of independent evidence — MIT CISR (n=721), BCG (n=1,250 executives + 10,635 workers), McKinsey (n=1,993), Deloitte (n=3,235), Stanford HAI, NBER, HBS — describes a curve that is slower than the roadmaps and faster than the skeptics.
• Most of the productivity will come from existing tools your company already pays for. Microsoft 365 Copilot, Google Workspace AI, Salesforce Einstein, and the AI features embedded in existing SaaS will do more of the work than any greenfield deployment. The 2027-2028 story is integration and workflow redesign around tools already on the invoice, not net-new vendor selection.
• The 5% / 95% gap compounds through workflow, not spend. BCG’s September 2025 data shows future-built companies generate 1.7x revenue growth and 3.6x TSR vs. laggards — while spending only 2x more. The differentiator is not budget; it is the discipline to redesign work around the tool rather than add the tool to existing work. That discipline takes 18-24 months to install and another 12-18 months to compound.
• What will not change at 300 people: headcount ratios, core workflows, or decision velocity. AI does not flatten the org chart of a services firm or a regional manufacturer within 3 years. It changes the shape of individual jobs faster than it changes the shape of the company. Plan around that asymmetry.
• The binary question “will we be disrupted” is the wrong frame. The right question is: “where does a 2-3% operating margin lift from AI show up in our P&L, and what does the work look like to capture it?” That is a planning exercise, not a strategy exercise.

The 12-Month Horizon (April 2026 → April 2027): Pilots, Plumbing, and Payroll

What actually happens. Most of year one is below the waterline. The work is boring and necessary: identity and access cleanup so AI tools can see the right data without seeing the wrong data, a data classification pass that nobody has done since 2019, contract reviews for the eight SaaS tools that quietly added AI features and new data-processing clauses. This is the phase where IT and Legal earn their seats and where the CFO signs off on recurring costs that will not produce a headline ROI number.

What the evidence supports. BCG AI at Work 2025 (n=10,635) found workers who received 5+ hours of training had 2.6x higher AI usage consistency. McKinsey’s November 2025 State of AI shows 88% of companies use AI in at least one function, but only 6% are high performers. The gap between “using AI” and “getting EBIT impact from AI” is the entire first year for most companies. Stanford Digital Economy Lab’s Enterprise AI Playbook (March 2026, 51 enterprise deployments) documents that agentic deployments produce 71% median productivity gains — but the same dataset shows median deployment time from pilot to production is 8-14 months.

What will not happen. Headcount reductions of the kind vendors imply. NBER’s Brynjolfsson-Li-Raymond RCT (n=5,172 customer support agents) shows a 15% average productivity gain — but the gain is heterogeneous: junior workers see 35% gains, senior workers see small declines, and the company did not reduce headcount; it absorbed the capacity into growth. The MIT CISR maturity model (n=721) shows Stage 1 firms have worse financial performance than non-AI peers by 12.6 percentage points. Year one of a poorly planned AI program destroys value before creating it.

The mid-market advantage. A 300-person company can complete this phase in 6-9 months. A 30,000-person enterprise takes 18-24. Use the speed gap.

The 24-Month Horizon (April 2027 → April 2028): Workflow Redesign and the First Real Gains

What actually happens. The companies that installed the plumbing in year one spend year two redesigning two to four workflows per function. Not “adding AI to” a workflow — redesigning it. This is where the Atlan analysis of 200 deployments (median +159.8% ROI) comes from: teams who treated AI as a reason to rebuild the process, not as a button added to the existing process. Three to five use cases per function is the realistic ceiling, not the twenty the vendor proposed.

What the evidence supports. The HBS “Cybernetic Teammate” RCT (n=776 P&G professionals, 2025) found AI-enhanced individuals matched full human team performance on complex tasks — but only when the workflow was redesigned around human-AI collaboration. BCG’s “Widening AI Value Gap” (September 2025) documents that companies moving beyond tool deployment to workflow redesign capture the value; others don’t. Deloitte’s State of AI Enterprise 2026 (n=3,235) finds 66% report productivity gains but only 34% are transforming core processes — the 32-point gap between those two numbers is the entire value opportunity in year two.

What will not happen. Full autonomy for anything touching customers, money, or regulated activity. The mid-market compliance posture in 2028 will still require human-in-the-loop review for contract review, financial reporting, medical decisioning, and any use case where an error creates regulatory or customer-trust exposure. This is a design decision that locks in for multiple years; the companies that build review into the workflow rather than bolting it on later avoid a painful rework cycle.

The quiet cost. Year two is where middle managers feel the change. The HBR finding that 76% of executives think employees are enthusiastic while only 31% of individual contributors actually are (November 2025) reflects a gap that widens in the redesign phase, not the pilot phase. Plan communication and change management now, before the redesigned workflows land on the people doing the work.

The 36-Month Horizon (April 2028 → April 2029): Where the Operating Margin Shows Up

What actually happens. By month 36, the companies that stayed disciplined see AI-attributable operating margin improvement in the 2-4% range. This is below what vendors promise and above what skeptics predict. It is also where agentic systems — now maturing past the 2026 hype cycle — begin to contribute meaningfully for specific bounded workflows (IT ticketing, procurement intake, vendor onboarding, some customer service tiers). The companies that redesigned workflows in year two now have the data infrastructure to deploy agents on top of those workflows; the companies that did not are still trying to pilot them.

What the evidence supports. MIT CISR’s maturity model shows Stage 3 firms outperform the baseline by 11.3 percentage points and Stage 4 firms by 17.1 percentage points. Reaching Stage 3 from a standing start takes most mid-market companies 30-36 months of consistent investment. McKinsey’s 6% high-performer cohort (November 2025) is disproportionately firms that started before 2024 — by 2029, the high-performer cohort expands meaningfully as mid-market companies who started in 2026 reach the threshold.

What will not happen. A flattened org structure. A fundamentally different product. A 50% headcount reduction. For a 300-person professional services firm or a 500-person regional manufacturer, the 2029 business looks recognizably like the 2026 business — same clients, similar headcount, similar product mix — but with materially better unit economics, faster turnaround on knowledge work, and a smaller portion of time spent on the kind of first-draft synthesis and data-shuffling that AI does well. The transformation is real. It is just not cosmetic.

What the honest CEO tells the board. “We will spend 3-4% of IT budget on AI for three years. We will not see meaningful EBIT impact in year one. We will see operational productivity gains in year two that do not yet flow to the margin line. We will see 2-3% operating margin improvement by year three if we execute the workflow redesign discipline. The alternative — waiting for the market to settle — means starting this conversation in 2029 and finishing it in 2032, which is the real competitive risk.”

Key Data Points

What This Means for Your Organization

The 3-year horizon is not a forecast. It is a planning artifact. The question the CEO should ask the leadership team in April 2026 is not “how fast can we go” but “what does the 36-month commitment actually cost, in what order, and who owns each phase.” The companies that answer that question clearly are the ones that reach MIT CISR Stage 3 by 2029. The companies that keep buying tools without the underlying discipline land in the 60% laggard cohort BCG documents — same spend, a fraction of the outcome.

The honest mid-market pattern across the corpus: year one buys you nothing visible; year two produces workflow-level gains that do not yet hit the margin line; year three produces the 2-4% operating margin improvement that justifies the cumulative investment. Any vendor or consultant promising a different curve is selling a roadmap that does not match how the change actually lands at 300 people.

If the 12/24/36 view above does not match the story currently being told inside your company, the gap is worth talking through. I am reachable at brandon@brandonsneider.com.

Sources

1. MIT CISR, “Enterprise AI Maturity Update” (2025) — Survey of 721 companies; 4-stage maturity model with measured financial performance deltas (Stage 1 −12.6pp, Stage 2 −3.5pp, Stage 3 +11.3pp, Stage 4 +17.1pp). Independent academic research. High credibility. https://cisr.mit.edu/publication/2025_0801_EnterpriseAIMaturityUpdate_WoernerSebastianWeillKaganer

2. BCG, “AI at Work 2025” — Survey of 10,635 workers across 11 countries. Key finding: 5+ hours training produces 2.6x usage consistency; 72% use AI regularly but only 5% of organizations get substantial financial gains. Consulting firm research with service interest; moderate credibility but large sample. https://www.bcg.com/publications/2025/ai-at-work-momentum-builds-but-gaps-remain

3. BCG, “Widening AI Value Gap” (September 2025) — Analysis of 1,250 senior executives. Future-built companies (5%) generate 1.7x revenue growth and 3.6x TSR vs. laggards (60%). Workflow redesign is the differentiator, not spend level. Moderate credibility; self-assessed classification. https://media-publications.bcg.com/The-Widening-AI-Value-Gap-Sept-2025.pdf

4. McKinsey, “State of AI” (November 2025) — Survey of 1,993 respondents. 88% use AI in at least one function; only 6% are high performers with >5% EBIT impact; one-third scaling enterprise-wide. Longest-running annual AI survey. Consulting firm research; moderate credibility with large sample. https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai

5. Deloitte, “State of AI in the Enterprise 2026” — Survey of 3,235 leaders across 24 countries. 60% employee AI access; governance readiness 30%; 66% report productivity gains but only 34% transforming core processes. Consulting firm research; moderate credibility. https://www.deloitte.com/global/en/issues/generative-ai/state-of-ai-in-enterprise.html

6. NBER, “Generative AI at Work” (Brynjolfsson, Li, Raymond, 2023) — Randomized controlled trial of 5,172 customer support agents. 15% average productivity gain with heterogeneous effects (juniors 35%, seniors small declines). Most-cited enterprise AI RCT. Peer-reviewed; high credibility. Note: predates current models by 2-3 generations; directional findings remain cited across 2025-2026 research. https://www.nber.org/papers/w31161

7. HBS, “Cybernetic Teammate” RCT (Sadun et al., 2025) — Randomized controlled trial of 776 P&G professionals. AI-enhanced individuals matched full human team performance when workflow was redesigned around human-AI collaboration. Only large-scale RCT on AI and teamwork. Peer-reviewed; high credibility. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5188231

8. Stanford Digital Economy Lab, “Enterprise AI Playbook” (Brynjolfsson et al., March 2026) — Analysis of 51 enterprise deployments. Agentic AI produces 71% median productivity gains vs. 40% for high-automation deployments. Academic research; high credibility; small sample. https://digitaleconomy.stanford.edu/publication/enterprise-ai-playbook/

9. HBR, “Leaders Assume Employees Are Excited About AI. They’re Wrong.” (November 2025) — 76% of executives think employees are enthusiastic; only 31% of individual contributors actually are. Independent academic publication. High credibility. https://hbr.org/2025/11/leaders-assume-employees-are-excited-about-ai-theyre-wrong

10. Atlan, “200-Deployment Analysis of AI ROI” (2025) — Analysis of 200 enterprise AI deployments showing median +159.8% ROI where workflow was redesigned, concentrated in deployments that treated AI as a reason to rebuild the process. Vendor-published (data governance company); moderate credibility, but distinctive dataset for mid-market ROI benchmarking.

Brandon Sneider | brandon@brandonsneider.com April 2026