AI ROI Gains Are Widespread, but Few Firms Are Explaining AI Decisions
FICO survey: 85.1% say AI met ROI goals, but only 5.2% are very confident explaining its decisions. Explaining AI decisions is the enterprise gap.
Enterprises report strong returns from AI, yet few can explain the decisions those systems make. In a FICO survey of 1,004 senior technology, risk and data leaders, 85.1% said AI has met or exceeded its initial ROI expectations. Only 5.2% said they are very confident explaining AI-driven decisions to regulators or customers. That gap between reported payoff and demonstrable control is the central finding of the research, and it is where compliance and liability costs are likely to accumulate.
FICO commissioned the study, and the figures are self-reported by respondents, so they measure what leaders believe rather than audited outcomes. The sample extends beyond financial-services executives to senior leaders across seven industries.
Where the ROI claims come from
The 85.1% headline is built from several tiers. Roughly 5% of respondents said AI significantly exceeded ROI expectations, more than 37% said it met them, and about 43% said it somewhat exceeded them. The payoff claim is therefore broad and moderate rather than concentrated among a few standout deployments.
Other surveys point the same way. Sonatafy Technology, citing a survey of more than 200 enterprise tech leaders, reports that only 31% of AI spend can be tied to specific business outcomes, even though 51% of leaders express high confidence in their ability to measure ROI. A separate study from Larridin found that 58% of senior leaders say there is no clear ownership of AI in their organisation and 75% lack AI governance, despite high confidence in AI returns. A Collibra survey with The Harris Poll, published in 2025, recorded 86% of decision-makers confident that agentic AI will deliver adequate ROI. Enterprises are reporting value while a much smaller share can trace that value to a documented decision path.
Pressure is part of the picture. CFO Dive reports that 92% of CFOs and top finance staff feel pressure to show ROI from AI, which raises the incentive to report returns before the supporting records are in place.
The explanation gap in numbers
| Metric | Share | Source |
|---|---|---|
| AI met or exceeded initial ROI expectations | 85.1% | FICO survey, 1,004 leaders |
| AI significantly exceeded ROI expectations | About 5% | FICO survey |
| Very confident explaining AI-driven decisions to regulators or customers | 5.2% | FICO survey |
| Customer-impacting decisions made without AI | 71.5% | FICO survey |
| AI spend attributable to specific business outcomes | 31% | Sonatafy Technology |
The 71.5% figure shows where decisions still sit. Human judgement accounts for 33.3% of decision-making, rules-based systems for 21.4%, and standard analytics for 16.8%. Those three legacy approaches together make up the 71.5% share, so most customer-facing decisions are not yet governed by the systems that drive the ROI claims. Firms that report AI returns are measuring a narrower slice of their decision load than the headline suggests.
Finance functions show the same bottleneck
Finance is where the explanation gap becomes a reporting problem. Datarails reports that just 5% of finance leaders trust AI to produce board-ready financial reports without human review, and only 4% trust it with month-end close. CFO Dive, summarising a survey of finance staff, found that 44% of respondents have only partial confidence that they could account for an AI agent's actions when questioned by an auditor or regulator. The same survey found that 76% say their organisation lacks the in-house expertise to understand how its AI operates.
Avalara's survey of Indian finance leaders reaches a similar conclusion. Its authors argue that confidence in explaining automated tax and compliance actions needs to catch up with deployment speed, and that closing the gap depends on trusted data, audit trails and documentation built in from the outset. Both findings point to the same bottleneck: review and documentation, not model capability.
Why platform consolidation is part of the answer
TheStreet, summarising the FICO research, reports that more than 75% of surveyed executives believe closer collaboration between business and IT leaders, combined with a shared AI platform, could drive ROI gains of 50% or more. That claim points to a structural fix. A single platform gives one place to log inputs, versions and decision rules, which is the record a regulator would ask for. Firms running many disconnected tools would need to reconstruct that record from several systems.
Masterofcode's analysis of enterprise AI returns makes a related point. Early initiatives typically deliver modest efficiency gains, and larger financial impact appears only after workflows, decision rights and governance models are reorganised around the AI. The governance layer is therefore a precondition for the larger returns, not a cost that follows them.
Trade-offs in explaining AI decisions
Enterprises face two options. They can keep expanding AI deployments and accept thin documentation, or they can slow rollouts to build audit trails and explanation capability first. The first path preserves near-term speed, but the FICO figures suggest most firms would struggle to defend individual outputs to a regulator today.
The technical risk compounds the governance gap. A model may not alert the organisation when the data feeding it has degraded, and its output can look equally confident whether the input was sound or not. Without logged inputs and documented decision rules, a company cannot separate a correct output from a confident error after the fact. The second option costs more up front but avoids reconstructing decisions under regulatory pressure.
TrustedTech founder Julian Hamood has argued that staff who know how to use AI can still lack the guardrails needed for safe adoption, so skills and controls must be funded separately. That distinction matters for budgeting: training raises usage, while audit tooling and documented decision rules are what allow a firm to explain a result.
Verdict
The evidence favours governance as the binding constraint on enterprise AI. The 85.1% ROI figure is a self-reported, vendor-commissioned perception measure, while the 5.2% explanation figure describes what firms can actually defend. Companies using AI in customer-facing or regulated decisions should prioritise decision logs, input-data monitoring and documented decision rules before expanding those deployments. Firms that do this now will be able to answer a regulator's request for an individual decision's reasoning. Firms that do not will have to rebuild that record under pressure.
Why this matters
Most enterprises are reporting AI returns without the records needed to explain how those returns were produced. When a regulator or customer asks for the reasoning behind an automated decision, firms with the weakest documentation carry the liability. The FICO data suggests that group is large, which turns explainability from a research topic into a budget line.
Related Articles
- Enterprise AI Agent Adoption Tripled, Yet ROI Proof Lags
- AI Customer Service ROI: Why 3 in 4 Deployments Fail
- OpenAI enterprise revenue overtakes consumer, ROI data lags
✔Human Verified
Researched and cross-referenced against primary sources by the Bytevyte editorial team. This article was generated with the assistance of artificial intelligence and reviewed by the Bytevyte editorial team.