Enterprise AI is creating real business value, but it is not reliably creating the clean cost reductions and time savings that many CIOs promised in their original investment cases. That is the uncomfortable conclusion emerging from the latest enterprise adoption data: organizations are getting better customer intelligence, sharper operational visibility, and more capable employees, yet the savings ledger often remains stubbornly unchanged.
This is not evidence that enterprise AI has failed. In many cases, it suggests the opposite. AI is delivering useful outcomes, but those outcomes are arriving as better decisions, faster discovery, improved interactions, and risk reduction rather than immediate cuts to headcount, software spending, or labor hours. The problem is that many AI business cases were built to measure the latter.
For CIOs, CFOs, and IT leaders, that mismatch is becoming a governance challenge. The question is no longer simply whether to deploy Microsoft Copilot, Google Workspace AI, Salesforce’s expanding AI portfolio, or other enterprise platforms. It is whether the organization can define, measure, and control the value those systems create before usage, cost, and operational risk scale beyond the original plan.
The early enterprise AI narrative was easy to understand. Employees would draft documents faster, summarize meetings, generate code, search knowledge bases, and automate repetitive workflows. Those capabilities would save hours. Saved hours would reduce costs. Reduced costs would create a straightforward return on investment.
In practice, the chain is rarely that direct.
An employee may save 30 minutes on a proposal, but the organization does not automatically eliminate 30 minutes of payroll expense. A sales team may use AI to create better account briefs, but the return may appear months later as improved retention, larger deal sizes, or fewer missed opportunities. A support organization may resolve routine questions more effectively, but it may choose to reinvest capacity into higher-touch customer service instead of reducing its workforce.
That does not make the AI deployment unsuccessful. It means the unit of value is different from the one used in the spreadsheet.
The most meaningful AI gains frequently take one of these forms:
That is why enterprise AI is entering a more demanding phase. The technology is increasingly embedded in daily work, but business leaders must now prove whether it is improving the enterprise as a system rather than merely making individual tasks feel faster.
If a finance analyst completes monthly variance commentary in half the time with an AI assistant, there are several possible outcomes. The analyst may review more business units, improve the quality of the analysis, investigate anomalies sooner, or spend additional time advising leaders. All may create business value. None necessarily lowers the finance department’s budget in the near term.
This is a familiar pattern in technology investments. Email did not reduce the amount of communication work employees performed. It increased the volume, speed, and expectation of communication. Collaboration platforms did not necessarily shrink meeting time. They often broadened access and accelerated coordination.
Generative AI and agentic AI can produce the same effect at a larger scale. They free capacity, but organizations frequently consume that capacity by raising standards, increasing output, or addressing work that was previously neglected.
An enterprise may have a fixed workforce, long-term licensing agreements, compliance obligations, service-level commitments, and teams that cannot simply be reduced after minor workflow improvements. Even when AI demonstrably speeds up work, the business may not be able—or willing—to translate that speed into immediate budget cuts.
This is particularly true in highly regulated sectors, customer-facing operations, and security-sensitive environments. Faster claims processing, legal review, threat triage, or healthcare administration may improve throughput and service quality, but human review requirements, staffing coverage, and accountability rules can remain intact.
The result is a persistent ROI tension:
Enterprise AI platforms are increasingly being used as a contextual layer over that information. Rather than requiring employees to search multiple applications manually, AI can summarize account history, identify trends in support cases, surface likely causes of delays, and help workers frame recommendations.
That produces an outcome that is genuinely valuable but difficult to quantify: decision velocity.
A sales manager who understands a customer’s issue before a renewal meeting may preserve revenue. A procurement team that spots unusual buying behavior may prevent a costly contract problem. An operations leader who sees a production risk early may avoid downtime. These results rarely fit neatly into a simple “hours saved per employee” model.
Microsoft Copilot capabilities within Microsoft 365, Google’s AI tools for Workspace, and Salesforce’s AI and automation offerings all aim to place assistance within the software employees already use. That matters because adoption rises when workers do not need to switch constantly between disconnected tools.
However, a better customer experience does not necessarily become an immediate cost saving. It may result in:
That gap matters because enterprise AI costs are broader than a monthly per-user license.
A pilot may rely on a small group of highly motivated users, a limited dataset, and a carefully controlled workflow. Production deployment means broader access, more varied prompts, more integrations, more exceptions, and more complex security requirements. The value may rise, but so can the operating burden.
High numbers of active users, generated documents, completed prompts, or automated summaries may demonstrate adoption. They do not prove that AI improved margins, customer outcomes, risk posture, or operating performance. In some cases, heavy usage can reveal the opposite: employees may be generating more content, creating more review work, or relying on AI where a simpler workflow would have been more effective.
The appropriate question is not, “How many employees used the assistant?”
It is, “Which business process changed, what outcome improved, what did that improvement cost, and would it have happened without AI?”
That is a harder question. It is also the one boards and finance teams increasingly expect CIOs to answer.
That capability can be transformative. It can also make spending and risk much less predictable.
Agentic AI is different. A single automated workflow may trigger repeated model calls, retrieve data from multiple systems, invoke APIs, generate documents, route approvals, and retry failed actions. When hundreds or thousands of tasks run concurrently, consumption can rise quickly.
This creates a new operating reality:
A chatbot that summarizes a policy document can be reviewed by an employee before use. An agent that creates a purchase request, modifies a customer record, resets access, sends an external message, or initiates a payment workflow needs much tighter controls.
The fundamental principle is straightforward: an AI agent should have only the minimum permissions necessary to perform its assigned task.
Organizations should avoid granting broad access simply because an agent needs to “work across systems.” An agent designed to summarize customer support trends should not have permission to alter customer billing records. A procurement assistant should not be able to approve its own purchase actions. A security agent should not be able to disable controls without explicit escalation paths.
This is not merely a technical design concern. It is an internal-control requirement.
Strong governance can accelerate deployment because it establishes the rules, visibility, and confidence needed to put AI into meaningful production workflows. Without governance, organizations can run impressive pilots while remaining unable to scale safely.
At a minimum, enterprise AI governance should include:
But AI also expands the attack surface.
Security teams must manage risks including prompt injection, data leakage, unsafe tool use, model manipulation, insecure integrations, excessive permissions, and unreliable outputs. An AI system that has access to internal knowledge bases, administrative tools, or security data can become a valuable target if controls are weak.
The use of AI in cyber defense should therefore follow the same principle as every other high-impact deployment: augment human judgment first, automate action only when controls and evidence justify it.
However, enterprises are discovering that persistent high-volume inference, large-scale data movement, and latency-sensitive workflows can create cost or performance challenges. Some workloads may be better served by a hybrid design that uses public cloud AI services while retaining selected data, inference pipelines, or operational systems in private environments or edge locations.
A hybrid strategy can support several objectives:
A low-risk employee writing assistant may fit naturally into a cloud-based productivity suite. A real-time industrial system, high-volume contact-center workflow, or regulated records process may demand more careful placement and tighter integration controls.
CIOs should resist one-size-fits-all AI architecture decisions. The more valuable the workload, the more important it becomes to evaluate data gravity, model access, latency, resilience, and total cost of ownership together.
The old procurement model—compare feature lists, negotiate a per-seat price, and select a platform—does not fully address agentic AI and embedded assistants.
Key evaluation areas include:
Microsoft Copilot may be compelling for organizations deeply invested in Microsoft 365, Teams, Azure, and related identity services. Google Workspace AI may be a natural choice for companies whose knowledge work lives in Gmail, Drive, Docs, and Meet. Salesforce’s AI capabilities may be particularly relevant where CRM data and customer workflows are the center of gravity.
Yet no platform should be treated as automatically optimal for every process. The best enterprise AI strategy often combines a governed core platform with carefully selected specialist tools for high-value workflows.
The standard should be interoperability, accountability, and measurable business impact—not brand loyalty.
For employee productivity tools, appropriate measures may include task completion time, output quality, rework rates, employee satisfaction, and capacity created. For customer-service tools, organizations may emphasize resolution rates, escalation rates, response time, retention, and customer satisfaction. For finance and operations, the focus may be forecast accuracy, working-capital improvement, exception detection, or avoided losses.
The key is to distinguish among:
A portfolio view allows leaders to stop underperforming experiments, fund successful workflows, and avoid the false expectation that every AI deployment will generate immediate savings.
It also makes governance more practical. Not every use case needs the same review process. A writing assistant and an autonomous finance agent should not be subject to identical controls.
That shift will require CIOs to work more closely with finance, operations, security, legal, procurement, and business-unit leaders. AI value cannot be owned by IT alone because the outcome is usually embedded in a business process, a customer journey, or a management decision.
The strongest organizations will do several things well:
The organizations that succeed will not be those that deploy the most AI tools. They will be the ones that understand where value actually lands, measure it honestly, control its cost, and build the governance required to turn promising assistance into dependable enterprise capability.
This is not evidence that enterprise AI has failed. In many cases, it suggests the opposite. AI is delivering useful outcomes, but those outcomes are arriving as better decisions, faster discovery, improved interactions, and risk reduction rather than immediate cuts to headcount, software spending, or labor hours. The problem is that many AI business cases were built to measure the latter.
For CIOs, CFOs, and IT leaders, that mismatch is becoming a governance challenge. The question is no longer simply whether to deploy Microsoft Copilot, Google Workspace AI, Salesforce’s expanding AI portfolio, or other enterprise platforms. It is whether the organization can define, measure, and control the value those systems create before usage, cost, and operational risk scale beyond the original plan.
Overview: Enterprise AI ROI Has Moved Beyond Simple Productivity Math
The early enterprise AI narrative was easy to understand. Employees would draft documents faster, summarize meetings, generate code, search knowledge bases, and automate repetitive workflows. Those capabilities would save hours. Saved hours would reduce costs. Reduced costs would create a straightforward return on investment.In practice, the chain is rarely that direct.
An employee may save 30 minutes on a proposal, but the organization does not automatically eliminate 30 minutes of payroll expense. A sales team may use AI to create better account briefs, but the return may appear months later as improved retention, larger deal sizes, or fewer missed opportunities. A support organization may resolve routine questions more effectively, but it may choose to reinvest capacity into higher-touch customer service instead of reducing its workforce.
That does not make the AI deployment unsuccessful. It means the unit of value is different from the one used in the spreadsheet.
The most meaningful AI gains frequently take one of these forms:
- Better quality in customer, product, and operational decisions
- Earlier identification of risks, bottlenecks, and anomalies
- More personalized customer engagement
- Faster access to institutional knowledge
- Improved consistency in routine business communications
- Higher employee capacity without a proportional increase in staffing
- Reduced rework, escalation, and avoidable process friction
That is why enterprise AI is entering a more demanding phase. The technology is increasingly embedded in daily work, but business leaders must now prove whether it is improving the enterprise as a system rather than merely making individual tasks feel faster.
The Core Disconnect: Productivity Is Not Automatically Cost Savings
Time saved is often reinvested, not removed
The distinction between productivity and cost reduction is central to the current AI ROI debate.If a finance analyst completes monthly variance commentary in half the time with an AI assistant, there are several possible outcomes. The analyst may review more business units, improve the quality of the analysis, investigate anomalies sooner, or spend additional time advising leaders. All may create business value. None necessarily lowers the finance department’s budget in the near term.
This is a familiar pattern in technology investments. Email did not reduce the amount of communication work employees performed. It increased the volume, speed, and expectation of communication. Collaboration platforms did not necessarily shrink meeting time. They often broadened access and accelerated coordination.
Generative AI and agentic AI can produce the same effect at a larger scale. They free capacity, but organizations frequently consume that capacity by raising standards, increasing output, or addressing work that was previously neglected.
The cost base is often fixed
Many CIOs also face a structural issue: labor, software, and infrastructure costs are not as elastic as early AI projections assume.An enterprise may have a fixed workforce, long-term licensing agreements, compliance obligations, service-level commitments, and teams that cannot simply be reduced after minor workflow improvements. Even when AI demonstrably speeds up work, the business may not be able—or willing—to translate that speed into immediate budget cuts.
This is particularly true in highly regulated sectors, customer-facing operations, and security-sensitive environments. Faster claims processing, legal review, threat triage, or healthcare administration may improve throughput and service quality, but human review requirements, staffing coverage, and accountability rules can remain intact.
The result is a persistent ROI tension:
For technology leaders, that means a business case based solely on hours saved is likely to understate value in some areas and overstate savings in others.
Why Business Insights and Customer Engagement Are Showing Up First
The strongest AI results are often appearing in domains where pattern recognition, summarization, search, and content generation can influence decisions quickly.AI turns fragmented information into usable context
Modern enterprises already possess enormous quantities of data: CRM notes, service tickets, product telemetry, emails, meeting transcripts, financial data, contracts, knowledge-base articles, operational logs, and supply chain records. The long-standing challenge has not been data collection alone. It has been making data useful at the moment a decision is made.Enterprise AI platforms are increasingly being used as a contextual layer over that information. Rather than requiring employees to search multiple applications manually, AI can summarize account history, identify trends in support cases, surface likely causes of delays, and help workers frame recommendations.
That produces an outcome that is genuinely valuable but difficult to quantify: decision velocity.
A sales manager who understands a customer’s issue before a renewal meeting may preserve revenue. A procurement team that spots unusual buying behavior may prevent a costly contract problem. An operations leader who sees a production risk early may avoid downtime. These results rarely fit neatly into a simple “hours saved per employee” model.
Customer experience gains can be commercially significant
AI is also well suited to customer interaction. It can draft responses, summarize conversations, classify requests, recommend next actions, and support personalization at a scale that would be difficult to achieve through manual processes alone.Microsoft Copilot capabilities within Microsoft 365, Google’s AI tools for Workspace, and Salesforce’s AI and automation offerings all aim to place assistance within the software employees already use. That matters because adoption rises when workers do not need to switch constantly between disconnected tools.
However, a better customer experience does not necessarily become an immediate cost saving. It may result in:
- Better customer retention
- Higher conversion rates
- Improved first-contact resolution
- Faster response times
- More consistent communications
- Greater account-manager capacity
- Better-informed renewals and upsell conversations
The Measurement Problem Is Now an Executive Problem
Many organizations can report that employees are using AI. Fewer can explain precisely which use cases are delivering financial value, how durable that value is, and whether it exceeds the full cost of deployment.That gap matters because enterprise AI costs are broader than a monthly per-user license.
The real cost of an AI deployment
A responsible enterprise AI cost model should include more than the vendor subscription. It must account for:- Platform licensing and consumption charges
- Cloud compute, storage, and networking
- Model and API usage
- Data preparation and integration work
- Security controls and identity management
- Governance, monitoring, and audit capabilities
- Employee training and change management
- Legal, compliance, and privacy review
- Process redesign and workflow ownership
- Human quality assurance and exception handling
- Vendor-management overhead
A pilot may rely on a small group of highly motivated users, a limited dataset, and a carefully controlled workflow. Production deployment means broader access, more varied prompts, more integrations, more exceptions, and more complex security requirements. The value may rise, but so can the operating burden.
Adoption is not a business outcome
Usage data has value, but it should not be confused with ROI.High numbers of active users, generated documents, completed prompts, or automated summaries may demonstrate adoption. They do not prove that AI improved margins, customer outcomes, risk posture, or operating performance. In some cases, heavy usage can reveal the opposite: employees may be generating more content, creating more review work, or relying on AI where a simpler workflow would have been more effective.
The appropriate question is not, “How many employees used the assistant?”
It is, “Which business process changed, what outcome improved, what did that improvement cost, and would it have happened without AI?”
That is a harder question. It is also the one boards and finance teams increasingly expect CIOs to answer.
Agentic AI Changes the Cost Equation
The next stage of enterprise AI complicates ROI further. Traditional copilots generally assist a person performing a task. Agentic AI systems can plan and execute multi-step work with varying degrees of autonomy, using tools, applications, workflows, and enterprise data.That capability can be transformative. It can also make spending and risk much less predictable.
From seat-based software to consumption-based execution
Traditional SaaS costs are usually understandable. A company buys a number of seats, assigns them to employees, and forecasts renewals based on workforce size and contract terms.Agentic AI is different. A single automated workflow may trigger repeated model calls, retrieve data from multiple systems, invoke APIs, generate documents, route approvals, and retry failed actions. When hundreds or thousands of tasks run concurrently, consumption can rise quickly.
This creates a new operating reality:
- Cost can scale with task volume rather than employee headcount.
- Autonomous retries can multiply usage unexpectedly.
- Poorly designed prompts or workflows can consume excessive resources.
- Integrations can create hidden transaction and API costs.
- Teams can deploy overlapping agents that duplicate work.
- Business units can create “shadow AI” expenses outside central IT budgets.
Agents require boundaries, not just access
Agentic systems also raise a higher bar for governance because they may perform actions rather than simply produce suggestions.A chatbot that summarizes a policy document can be reviewed by an employee before use. An agent that creates a purchase request, modifies a customer record, resets access, sends an external message, or initiates a payment workflow needs much tighter controls.
The fundamental principle is straightforward: an AI agent should have only the minimum permissions necessary to perform its assigned task.
Organizations should avoid granting broad access simply because an agent needs to “work across systems.” An agent designed to summarize customer support trends should not have permission to alter customer billing records. A procurement assistant should not be able to approve its own purchase actions. A security agent should not be able to disable controls without explicit escalation paths.
This is not merely a technical design concern. It is an internal-control requirement.
Governance Has Become a Prerequisite for Scale
Enterprise AI governance is often treated as a compliance exercise that slows innovation. That framing is increasingly outdated.Strong governance can accelerate deployment because it establishes the rules, visibility, and confidence needed to put AI into meaningful production workflows. Without governance, organizations can run impressive pilots while remaining unable to scale safely.
What a practical AI governance model should cover
A usable governance framework must connect business strategy, technology architecture, security controls, financial accountability, and operational ownership. It should define who can approve AI use cases, which data can be used, how models are evaluated, and who is accountable when a system makes a mistake.At a minimum, enterprise AI governance should include:
- Use-case classification based on business impact and risk
- Data controls covering sensitive, regulated, and proprietary information
- Identity and access management for users, models, tools, and agents
- Human approval requirements for high-impact decisions and actions
- Evaluation standards for accuracy, reliability, bias, and safety
- Logging and audit trails for prompts, outputs, tool calls, and decisions
- Spend monitoring by business unit, workflow, model, and application
- Incident-response processes for harmful outputs, data exposure, or agent failures
- Vendor governance for contracts, model changes, data handling, and portability
- Lifecycle management for retiring outdated models, prompts, and automations
Security teams face a double burden
AI is increasingly used in cybersecurity for alert triage, threat hunting, vulnerability analysis, phishing detection, and incident-response support. These applications can improve the speed at which defenders interpret a growing volume of security signals.But AI also expands the attack surface.
Security teams must manage risks including prompt injection, data leakage, unsafe tool use, model manipulation, insecure integrations, excessive permissions, and unreliable outputs. An AI system that has access to internal knowledge bases, administrative tools, or security data can become a valuable target if controls are weak.
The use of AI in cyber defense should therefore follow the same principle as every other high-impact deployment: augment human judgment first, automate action only when controls and evidence justify it.
Cloud Strategy Is Being Rewritten Around AI Workloads
AI is also forcing enterprises to reconsider cloud architecture. The assumption that all AI workloads should run in a public cloud is proving too simplistic for organizations with strict latency, cost, data-residency, or integration requirements.Why hybrid AI architecture is gaining attention
Public cloud platforms offer rapid access to powerful models, managed services, elastic infrastructure, and a broad ecosystem of AI tools. For many workloads, that flexibility remains essential.However, enterprises are discovering that persistent high-volume inference, large-scale data movement, and latency-sensitive workflows can create cost or performance challenges. Some workloads may be better served by a hybrid design that uses public cloud AI services while retaining selected data, inference pipelines, or operational systems in private environments or edge locations.
A hybrid strategy can support several objectives:
- Keeping sensitive data closer to controlled environments
- Reducing latency for operational applications
- Limiting expensive data movement
- Maintaining resilience across multiple environments
- Using specialized infrastructure for predictable workloads
- Avoiding excessive dependency on a single platform
Architecture decisions must follow the use case
The right question is not whether cloud, private infrastructure, or edge computing is “best” for AI. The right question is which architecture best supports a specific business process, data classification, performance target, and cost profile.A low-risk employee writing assistant may fit naturally into a cloud-based productivity suite. A real-time industrial system, high-volume contact-center workflow, or regulated records process may demand more careful placement and tighter integration controls.
CIOs should resist one-size-fits-all AI architecture decisions. The more valuable the workload, the more important it becomes to evaluate data gravity, model access, latency, resilience, and total cost of ownership together.
Rethinking Vendor Evaluation for the AI Era
The competition among Microsoft, Google, Salesforce, SAP, OpenAI, and other enterprise technology providers is driving faster product development and deeper AI integration. That can benefit customers, but it also increases the risk of buying overlapping capabilities from multiple vendors.The old procurement model—compare feature lists, negotiate a per-seat price, and select a platform—does not fully address agentic AI and embedded assistants.
What procurement teams should evaluate now
Enterprise buyers should examine the full operating model around an AI offering, not just the quality of its demonstrations.Key evaluation areas include:
- Data ownership and retention policies
- Model-training boundaries for enterprise content
- Integration depth with systems of record
- Identity, permission, and role-based access controls
- Audit logging and administrative reporting
- Consumption pricing and budget safeguards
- Portability of prompts, workflows, and agent configurations
- Model choice and ability to avoid vendor lock-in
- Support for human approvals and exception handling
- Security testing and incident-response commitments
- Product roadmap stability and licensing-change risk
Avoid the “one assistant for everything” trap
A broad enterprise AI platform can reduce complexity, but centralization should not be confused with universal suitability.Microsoft Copilot may be compelling for organizations deeply invested in Microsoft 365, Teams, Azure, and related identity services. Google Workspace AI may be a natural choice for companies whose knowledge work lives in Gmail, Drive, Docs, and Meet. Salesforce’s AI capabilities may be particularly relevant where CRM data and customer workflows are the center of gravity.
Yet no platform should be treated as automatically optimal for every process. The best enterprise AI strategy often combines a governed core platform with carefully selected specialist tools for high-value workflows.
The standard should be interoperability, accountability, and measurable business impact—not brand loyalty.
A Better Framework for Measuring Enterprise AI Value
Organizations need to replace simplistic “hours saved” reporting with a layered approach that measures both operational and commercial value.Start with a business baseline
Every material AI use case should begin with a baseline. Before deployment, the organization should document the current state of the process:- Define the business objective.
- Identify the process owner.
- Measure current cycle time, quality, cost, throughput, and error rates.
- Establish customer, employee, and risk metrics where relevant.
- Calculate the full expected cost of implementation and operation.
- Set thresholds for continuing, expanding, redesigning, or stopping the use case.
Measure the right category of value
Different AI use cases require different scorecards.For employee productivity tools, appropriate measures may include task completion time, output quality, rework rates, employee satisfaction, and capacity created. For customer-service tools, organizations may emphasize resolution rates, escalation rates, response time, retention, and customer satisfaction. For finance and operations, the focus may be forecast accuracy, working-capital improvement, exception detection, or avoided losses.
The key is to distinguish among:
- Realized savings: costs actually removed from the budget
- Cost avoidance: costs not incurred because capacity increased
- Capacity creation: time redirected toward higher-value work
- Revenue impact: growth, conversion, retention, or upsell gains
- Risk reduction: lower probability or impact of operational failures
- Quality improvement: fewer errors, better consistency, stronger compliance
Treat AI as a portfolio, not a single project
Enterprise AI programs should be managed like a portfolio of investments. Some low-risk tools will create broad but modest value across the workforce. Others will have narrow applicability but potentially transformative effects on a high-value process.A portfolio view allows leaders to stop underperforming experiments, fund successful workflows, and avoid the false expectation that every AI deployment will generate immediate savings.
It also makes governance more practical. Not every use case needs the same review process. A writing assistant and an autonomous finance agent should not be subject to identical controls.
The Strategic Shift CIOs Need to Make
The enterprise AI conversation must move from “How many hours did we save?” to “What operating advantage did we create, and can we govern it at scale?”That shift will require CIOs to work more closely with finance, operations, security, legal, procurement, and business-unit leaders. AI value cannot be owned by IT alone because the outcome is usually embedded in a business process, a customer journey, or a management decision.
The strongest organizations will do several things well:
- Tie AI investments to named business owners and measurable outcomes
- Separate productivity gains from actual budget savings
- Build cost visibility before autonomous usage expands
- Design agents with narrow permissions and clear approval paths
- Invest in data quality, integration, and process redesign
- Use governance as an enabler of scale rather than a late-stage review
- Measure customer, revenue, quality, and risk outcomes alongside efficiency
- Shut down AI use cases that generate activity without meaningful impact
The organizations that succeed will not be those that deploy the most AI tools. They will be the ones that understand where value actually lands, measure it honestly, control its cost, and build the governance required to turn promising assistance into dependable enterprise capability.
References
- Primary source: MarketScale
Published: 2026-07-25T13:27:00+00:00
Enterprise AI delivers insights, not cost savings, SAP surve
Enterprises report AI helps with insights and customer interaction but falls short on cost and time savings, per SAP. Governance and cost control gaps are widen
www.marketscale.com
- Related coverage: ciodive.com
AI ROI is rising, but not where companies expected | CIO Dive
Enterprises reported that AI is helping them find business insights and interact with customers, but not necessarily save money or time, an SAP survey found.www.ciodive.com
