AI creates business value not simply because people use it, but because it changes processes that affect revenue, costs and cash flow. A PYMNTS Intelligence survey found that companies leading in AI adoption were twice as likely as those in the bottom group to report annual revenue growth above 15%. That is an important correlation, not proof that the technology caused the growth.
For enterprise leaders, the practical question is not how many employees already use AI. It is whether that use has changed how the organization sells, delivers, collects payments and handles exceptions.
- 24% of AI leaders reported revenue growth above 15%.
- 12% of companies in the bottom group reported growth above that threshold.
These figures describe the share of companies reporting strong growth, not their average growth rate. The survey does not say AI leaders grow twice as fast. It says fast-growing companies are more common in that group.
What the survey tells us, and what it does not
The PYMNTS Agentic Edge Index, published in November, compares groups of companies by their maturity in implementing AI. It also found a profitability gap: 21% of companies in the top third reported profit-margin increases above 5%, compared with 11% in the bottom third. According to the report, the revenue and profitability differences are statistically significant.
But statistical significance does not establish what caused what. Companies with strong management, well-organized data and resources to spare may both implement AI more effectively and perform better. Successful implementation may contribute to performance. Both directions may be at work.
Not a return forecast: The gap between these groups does not promise a similar outcome from your AI investment. Investment decisions need to rest on your own process data and a phased assessment of benefits.
That distinction matters for management. The survey warrants a serious examination of how deeply AI is embedded in the business, not automatic approval for a larger budget. The business case still has to be built inside the organization.
The value lies between tasks
The revealing question is not just who uses more AI, but where they use it. Among the leaders, 46% use AI to generate contracts and quotes, compared with 19% in the bottom group. For invoice processing and accounting reconciliation, the figures are 46% versus 26%. For fault detection and incident-response prioritization, they are 61% versus 32%.
Customer-service chatbots, by contrast, are almost equally common in both groups. That does not make them worthless. It means that having a chatbot does little to distinguish a company that has embedded AI deeply from one that has added a new interface to an old process.
Our management takeaway: the right unit of investment is a business process, not an isolated task.
- Sales: A standalone tool drafts a quote; process integration covers preparation, approval and follow-up. Measure time to signature.
- Finance: A standalone tool extracts invoice data; process integration includes reconciliation and exception handling. Measure processing cost per invoice.
- Operations: A standalone tool summarizes an alert; process integration prioritizes it and routes it for action. Measure downtime.
- Collections: A standalone tool drafts a reminder; process integration identifies payment blockers and manages follow-up. Measure time to collect.
These are implementation options, not outcomes measured by the survey. The difference is accountability for the result. A system that drafts quotes quickly does not necessarily shorten the path to revenue if those quotes still wait for manual approval in an overloaded inbox.
Collections requires another distinction: collecting payments faster may improve cash flow and working capital without increasing revenue. When those metrics get mixed together, it becomes easy to claim success that is hard to find in the financial statements.
Keep humans in the loop without making them a bottleneck
Sales, finance and operations involve decisions that cannot all be reduced to fixed rules: a missing document, an unusual commercial term, a supplier with a complicated history or an incident that needs interpretation. These are areas where AI can offer substantial value, but also introduce risk.
The answer is neither to give an agent unrestricted authority nor to require an employee to approve every action. Human oversight should focus on risk, exceptions and the overall quality of the process.
- Low-risk actions can run automatically within defined permissions and rules.
- Financial or legal commitments need approval paths that follow organizational policy.
- Missing information, conflicting system records or rule violations should trigger escalation to a person.
- Automatically approved actions still need sampling and failure monitoring.
The goal is for one manager to oversee many processes through a clear exception queue, rather than redo the agent's work. That requires decision records, a way to stop execution and the ability to correct a mistaken action. Asking a model to be careful is not a control mechanism.
For the CFO: time saved is not necessarily money saved
According to the report, leaders invested an average of $14.5 million in new AI technologies over the past year, compared with $9.1 million in the bottom group. Those are figures from the sample, not a recommended spending threshold.
A midsize organization does not need to copy that budget. It needs measurement discipline that separates different kinds of benefit:
- Realized savings: Actual reductions in expenses, overtime or supplier payments.
- Freed capacity: The ability to handle more work with the same team, backed by a clear plan to use that capacity.
- Commercial improvement: Shorter sales cycles, lower churn or better conversion, verified in practice.
- Cash-flow and risk improvement: Faster collections, fewer incorrect payments or less downtime.
Set those benefits against the full cost: model usage, integration, maintenance, testing, information security, human oversight and error correction. If a team saves time on data entry but spends it reviewing outputs, the work may simply have changed form.
Turn the findings into an operating plan
Do not start by choosing a model. Start with a process that has an owner, a business problem and enough data to establish whether anything improves.
- Choose a process with business impact: Find a recurring bottleneck that affects cost, revenue, cash flow or business continuity.
- Establish the baseline: Record handling times, rework, exceptions and total cost before making changes.
- Define authority and controls: Specify what the agent may do, when approval is required and how to stop an incorrect action.
- Test within a limited scope: Compare against current performance while assessing quality, cost and effects on employees and customers.
- Scale on proven value: Expand only after the improvement justifies operating and oversight costs.
This requires collaboration among the business process owner, finance, IT, information security and AI specialists. It calls for deep domain knowledge, a relevant academic foundation and practical implementation experience. The ability to deliver an impressive demo is no substitute for understanding an accounting process or being accountable for a production system.
Organizations also need to advance on two tracks: AI literacy for employees and the internal capability to build and manage agents. A personal AI tool can require substantial changes in working habits. An agent integrated well into an existing system can work in the background and reduce the change required of users. It still needs a proper foundation of permissions, monitoring, testing and version control.
License counts tell you how much access to technology you bought. Process performance tells you what the organization got from it.
At the next executive meeting, ask for fewer lists of AI tools and more answers to concrete questions: Which process changed? What improved? What did it cost to operate? Who is accountable when the system gets something wrong? That is where you can begin to distinguish adoption that looks good in a presentation from a measurable business advantage.
