McKinsey published its 2026 State of AI survey on August 25, and the findings are useful precisely because they are not pure AI hype. Organizations are using AI more broadly, including agentic AI and coding agents, but the financial impact is still concentrated. McKinsey reports that 80 percent of respondents say AI has improved their individual productivity, while 37 percent report positive EBIT impact from AI use, about the same share as last year.
That gap matters for business owners. A team can feel faster, write drafts sooner, answer routine questions more quickly, and still leave the company without a clear return on the next AI subscription, consultant project, or software build. Productivity is real, but it does not automatically become profit, margin, better service, or fewer bottlenecks.
For New Jersey businesses, schools, nonprofits, healthcare practices, manufacturers, and professional services firms, the lesson is practical: AI ROI needs more than access to tools. It needs a workflow owner, a measurable result, and a budget line that someone actually watches.
The productivity story is only half the story
McKinsey's survey says AI use is expanding across functions, and agentic AI adoption is rising fastest among large enterprises. Smaller organizations are not absent from the trend, but the report says their agentic AI scaling stayed flat at 22 percent while large organizations moved ahead faster.
That should not push smaller businesses into panic buying. It should push them into clearer selection. The right question is not whether the company has the newest AI feature. The better question is whether the feature changes a process that already matters.
If AI helps an employee summarize notes but the same customer handoff still waits three days for approval, the bottleneck survived. If a coding assistant helps build an internal tool but nobody budgets support, security review, documentation, or ownership, the build-versus-buy decision may only move the cost to a different shelf.
AI costs are becoming a management issue
The report also notes that about one in five respondents say AI-related operating costs, including token costs, constrained AI use. That is an owner-level issue, not just an IT setting. AI spending can spread through subscriptions, embedded software features, API usage, consultants, data cleanup, training, governance, and support time.
Those costs are not automatically bad. A higher bill can be justified when it replaces manual work, improves service quality, speeds revenue, or reduces avoidable risk. The problem is approving growth in AI spend without agreeing on what the spend is supposed to change.
This is where AI budget planning needs to become more ordinary. Not dramatic. Not a 90-page policy. Just clear enough that the business knows which use cases are experiments, which are production workflows, and which are expensive curiosities wearing a productivity badge.
Questions to ask before expanding AI spend
- Which workflow will change? Name the process, not just the tool. Examples include intake, quoting, scheduling, documentation, support triage, month-end reporting, inventory review, or proposal drafting.
- What result will be measured? Choose a business measure such as cycle time, rework, response time, close rate, error rate, documentation completeness, support backlog, or avoided software cost.
- Who owns the workflow after launch? AI cannot be owned only by the person who found the tool. Assign responsibility for training, quality checks, access, cost review, and process changes.
- What data is allowed into the tool? Decide whether customer records, patient information, contracts, financials, credentials, employee details, or confidential vendor material are permitted.
- How will operating cost be monitored? Track licenses, usage, API consumption, premium features, implementation fees, and support time before the cost becomes normal background noise.
- Are we buying software or building it ourselves? McKinsey reports that some organizations are deciding against purchases because coding agents can help build functionality internally. That choice still needs a maintenance, security, and ownership plan.
The practical next step
Before approving the next AI tool or renewal, ask for a one-page AI value plan. It should identify the workflow, the current pain point, the expected business result, the owner, the data boundaries, the monthly operating cost, and the review date.
Keep the first version simple. For many smaller organizations, the best AI planning starts with three buckets: approved everyday use, controlled pilots, and blocked or not-yet-approved uses. That gives employees room to work while keeping sensitive data, surprise costs, and unsupported internal tools from drifting into the business quietly.
The strongest takeaway from McKinsey's State of AI report is not that every company needs more AI. It is that AI value shows up when the business changes how work gets done. The workflow is where the ROI is hiding, and it is not going to file a ticket by itself.
Sources and further reading
