Seven Factors That Actually Drive AI Returns: What 1,006 Executives Got Wrong About Their Own Programs

Executive Summary

• 90% of organizations in this survey report getting value from AI — yet the drivers of that value are widely misunderstood. A survey of 1,006 senior executives across 11 countries found most leaders correctly deploy AI but attribute success to the wrong causes. The real differentiators are measurement discipline and change management architecture, not tool selection or model capability.
• Data quality doubles value outcomes. Organizations with clean, complete, and current data are 2x as likely to achieve strong AI returns compared to those working from fragmented sources. Data governance is the upstream investment most mid-market companies skip.
• The change management gap is measurable. Organizations that combine technology rollout with structured change management — training, stakeholder engagement, deliberate communication — achieve adoption rates above 80%. Organizations that skip this (“set-and-forget” implementations) achieve below 50%. The gap is 30+ percentage points, achieved by process design not technology selection.
• Static AI deployments decay. Companies without real-time monitoring and automated retraining see a 15% effectiveness decline within 6 months due to data drift. Most mid-market companies treat AI deployment as a project with an end date. It is an ongoing operational discipline.
• The key differentiator is not which tools you deploy — it is how deliberately you measure, manage, and report the value those tools create. This is the core finding. Most AI programs compete on tooling; the programs that win compete on measurement and management.

Context: What This Study Adds to the Existing Evidence

The corpus already has robust gap-quantification: McKinsey finds only 6% of companies achieve >5% EBIT impact; PwC finds 74% of AI economic value concentrated in 20% of organizations; Grant Thornton finds fully-integrated AI companies 4x more likely to report revenue growth (58% vs. 15%). These studies tell you how big the gap is.

The Davenport/Srinivasan Return on AI Institute study (March 2026, n=1,006, HBR) is different. It asks: what do the leaders do differently? The seven-factor framework is the corpus’s first evidence-based answer to that question from a primary survey of more than 1,000 C-suite executives.

Important source caveat: The full HBR article is paywalled. The seven factors and supporting statistics in this file were synthesized from the companion webinar description, author interviews, and secondary summaries of the published article. The survey methodology (n=1,006, 11 countries, 32 industries) and primary statistics (90% value, 71% CIO freeze risk, 2x data quality multiplier) are author-confirmed in public sources. The factor descriptions draw from publicly available summaries and a Freshworks interview with Laks Srinivasan. Cross-reference with the original HBR article before citing specific language.

Temporal tier: TIER 1 — Published March 17, 2026. Survey conducted by the Return on AI Institute with respondents who had close familiarity with their organizations’ AI initiatives.

The Seven Factors: What the Data Actually Shows

Factor 1: Data Quality and Availability

Organizations rating their data as “clean, complete, and current” are 2x as likely to achieve strong AI value compared to those working with fragmented sources. Early investment in data governance — standardizing formats, enforcing metadata policies — is the single highest-leverage upstream action.

This corroborates the IBM IBV Tech Debt Reckoning (n=1,300): organizations that fully account for tech-debt remediation in their AI business cases project 29% higher ROI. The mechanism is the same — data quality is not a prerequisite to deploy, but it determines whether deployed AI sustains value.

Implication for mid-market: Most 200-500 person companies can identify one or two domains where their data is already structured and trustworthy (payroll, accounts payable, CRM deal stages). Start there. The mistake is deploying across 15 use cases simultaneously with uneven data quality.

Factor 2: AI Talent and Expertise

Organizations with dedicated AI centers of excellence or academic partnerships report a 30% increase in project delivery confidence compared to those without. The bottleneck is not model access — it is practitioners who can translate technical AI outputs into specific business decisions.

This is the practitioner gap. Access to GPT-4o, Claude 3.7, and Gemini is commoditized. The scarce resource is the person who understands both the business process deeply enough to know what “good output” looks like and the AI system well enough to identify when it is drifting.

For mid-market companies: the AI center model doesn’t require hiring a team of data scientists. It requires designating 2-3 people with protected time, clear ownership, and a mandate to build replicable deployment patterns. The KPMG finding that talent investment produces a 4x return (77% value vs. 20% without it) is the financial case.

Factor 3: Clear Business Objectives

Projects beginning with a specific, defined problem — churn reduction target, supply chain cycle time improvement, accounts payable processing cost benchmark — outperform technology-first approaches. Alignment with measurable KPIs enables mid-course correction.

This is the hardest behavioral change because most AI programs start with the tool (“we’re implementing Copilot”) not the outcome (“we want to cut proposal turnaround from 5 days to 2”). The tool selection is a second-order decision.

The Futurum 2026 finding (n=830 IT decision-makers) that companies are shifting primary AI ROI metrics from productivity to direct revenue impact reflects this same maturation: organizations that started with outcome clarity are now measuring revenue; organizations that started with tooling are still measuring adoption rates.

Factor 4: Effective Change Management

Structured change management — deliberate training, stakeholder engagement, and communication — produces adoption rates above 80%. Organizations that deploy without this (“set-and-forget”) achieve below 50%.

The 30+ percentage point gap is not explained by technology differences. It is entirely organizational design. BCG’s 10-20-70 rule (10% technology, 20% data, 70% people and process) is the framework; the Davenport/Srinivasan survey is the confirmation at n=1,006 across 32 industries.

Factor 5: Scalable Infrastructure

Organizations using elastic computing resources (cloud-native, variable-cost compute) report shorter model training times and reduced operational costs, with higher overall returns. Infrastructure inflexibility is a compounding drag: it delays retraining, forces deployment compromises, and creates incentives to keep underperforming systems running past their effective lifespan.

For mid-market: this factor is the easiest to address via SaaS AI platforms (which abstract infrastructure) and the most costly to address when building proprietary systems. The architecture decision made at deployment creates a compounding cost or compounding advantage for 3-5 years.

Factor 6: Continuous Monitoring and Evaluation

Companies with real-time monitoring and automated retraining sustain value. Static deployments — systems deployed and left running without feedback loops — experience a 15% effectiveness decline within 6 months due to data drift.

This is a budget and governance finding, not a technology one. The companies that treat AI deployment as an event (capital project, implementation, go-live) capture initial gains and then watch ROI decay. The companies that treat it as an operational capability (budget for monitoring, retraining cycles, performance audits) sustain and compound gains.

The Forrester TEI model (207% three-year ROI for customer service AI) assumes sustained optimization. The 15% effectiveness decay figure shows what happens when that optimization is absent.

Factor 7: Integration with Existing Systems

Embedding AI into existing ERP, CRM, and analytics platforms — rather than deploying alongside them — reduces adoption friction, accelerates decision cycles, and amplifies economic impact per algorithm. AI that surfaces insights where users already work produces measurable behavior change; AI that requires users to visit a new interface produces low utilization.

This corroborates the embedded AI finding in the Workday/SAP/ServiceNow research: the highest-adoption AI deployments in mid-market are the ones inside tools employees already use daily. The point solution strategy (best-in-class AI deployed standalone) loses to the integrated strategy on adoption, even when the point solution has superior model performance.

The Six-Stage AI Economic Maturity Model

The research introduces a proprietary six-stage progression model showing how organizations advance from initial experimentation to high-value returns. The full stage definitions are in the HBR article (paywalled). The public description is: “a research-based roadmap showing how organizations progress from pilots to high-value returns,” with detailed breakdowns by industry, company size, and geography.

The model’s practical implication is that AI value accumulates nonlinearly — early stages produce modest returns, but organizations that progress through all six stages compound gains that earlier-stage peers cannot match. This is structurally consistent with the MIT CISR 4-stage maturity model (Stage 1: −12.6pp financial performance vs. peers; Stage 4: +17.1pp) and the BCG AI Radar finding that only 5% of organizations reach “substantial gains.”

Key Data Points

What This Means for Your Organization

The 90% “getting value” figure will confuse boards that are familiar with the McKinsey 6% high-performer number. They are not contradictory. Davenport and Srinivasan measure self-reported value at any level (including “small” benefit at 9%). McKinsey measures financial performance impact at the EBIT level. Most organizations get some value from AI. Very few get the value that moves EBIT. The Davenport research tells you what separates those two outcomes: it is not technology selection. It is whether the organization measures rigorously, manages actively, and redesigns processes rather than layering AI onto old ones.

The practical diagnostic is simple. Ask your AI program team these seven questions, one per factor:

1. Where is our data unambiguously clean enough to trust AI outputs on it today? (Start there; don’t start with the messy data.)
2. Who is the person — named, accountable — who bridges our business process and our AI system?
3. What specific business outcome, with a number attached, does this AI project need to produce?
4. What is the training and communication plan before go-live — not after problems appear?
5. Is our compute infrastructure elastic enough to retrain when data patterns shift?
6. Who reviews AI performance after go-live, on what cadence, and what triggers a retraining cycle?
7. Does this AI surface outputs inside the tools users already open daily, or does it require a new workflow?

If you cannot answer all seven before deploying, the 15% effectiveness decay at 6 months is the likely outcome — not a technology failure but an operational design gap.

The Challenger, Gray & Christmas March 2026 data (AI now the #1 cited reason for U.S. job cuts at 25% of all announcements) is the pressure creating urgency around these seven factors. Boards and CFOs are asking for visible, auditable, compounding AI value — not because AI is controversial, but because other companies are now cutting headcount while citing AI, creating competitive and governance pressure on every organization that has not yet demonstrated returns.

If these questions surface specific gaps in your program design, that is a conversation worth having — brandon@brandonsneider.com.

Sources

1. Davenport, Thomas H. and Srinivasan, Laks (Return on AI Institute) — “7 Factors That Drive Returns on AI Investments, According to a New Survey.” Harvard Business Review, March 17, 2026. URL: https://hbr.org/2026/03/7-factors-that-drive-returns-on-ai-investments-according-to-a-new-survey — MEDIUM-HIGH credibility (paywalled; primary methodology confirmed; HBR peer-reviewed academic/practitioner publication; Babson College + Return on AI Institute, independent of platform vendors; n=1,006, 11 countries, 32 industries; self-reported value measures). Temporal tier: TIER 1.

2. Challenger, Gray & Christmas — “Challenger Report: March Cuts Rise 25% From February, AI Leads Reasons.” April 2026. URL: https://www.challengergray.com/blog/challenger-report-march-cuts-rise-25-from-february-ai-leads-reasons/ — HIGH credibility (independent outplacement firm, long-running monthly tracking methodology, employer-stated reasons, no vendor interest in AI adoption narrative). Temporal tier: TIER 1.

3. IBM IBV “The Tech Debt Reckoning” (n=1,300, Nov 2025) — 29% higher ROI when tech-debt accounting is included; corroborates the data quality / Factor 1 relationship. Available in corpus: research/04-consulting-firms/ibm-ibv-tech-debt-reckoning-2026.md.

4. KPMG Global AI Pulse Q1 2026 (n=2,110, March 2026) — talent investment correlates with 4x more value (77% vs. 20%); corroborates Factor 2. Available in corpus: research/04-consulting-firms/kpmg-global-ai-pulse-tech-report-2026.md.

5. BCG “AI at Work 2025” (n=10,635) — 10-20-70 value framework; corroborates Factor 4. Available in corpus: research/01-ai-native-landscape/bcg-ai-at-work-2025.md.

6. Futurum Group 1H 2026 AI ROI Survey (n=830) — ROI metric shift from productivity to revenue; corroborates Factor 3 outcome clarity. Available in corpus: research/01-ai-native-landscape/futurum-enterprise-ai-roi-1h-2026.md.

See also (wiki): ai-roi-evidence · productivity-rcts

Brandon Sneider | brandon@brandonsneider.com April 2026