A public library patron wants a book. The copy in the catalog is checked out. Three other people are waiting. Another library in the region may have it, and the title is inexpensive enough to buy. But the requester is not a local taxpayer, and local policy matters.
Librarians make decisions like this every day. Cost matters. Demand matters. Service priorities and context all play a role. None of that changes with AI. What can change is how much time is spent on the work around the decision: checking systems, verifying availability, moving between screens, and repeating the same background steps.
That is where AI is most likely to have an immediate effect. There is strong evidence that it reduces time spent on routine internal work. The more difficult question is what happens next. Saving time is not the goal on its own. What matters is how that time is used and whether the work itself changes.
Workflow gains are real, but they are only the first step
Recent studies point to consistent gains from AI in internal workflows. Research from St. Louis Fed found that workers using generative AI saved an average of 5.4 percent of their work hours in the prior week, or about 2.2 hours in a 40-hour week. In the same survey, more than one in five users reported saving four or more hours in a week.
Other research points in the same direction. Bain reported productivity gains in the 10 to 15 percent range in software development. In customer support, field research summarized by NBER found that agents using AI handled 13.8 percent more inquiries per hour. These gains are often tied to documentation, summarization, internal search, and routine support tasks.
This pattern is familiar territory for libraries. Staff time is often absorbed by reporting, recurring communications, documentation, knowledge lookup, and the many small handoffs required to run services across multiple systems. AI can reduce that effort. It can draft, summarize, retrieve, and organize.
But that is where many organizations stop. They introduce a tool and assume the benefit will follow.
The second decision is where value is captured
Saving time inside a workflow does not, by itself, improve service. Choosing to introduce AI is one decision. Deciding how the work changes afterward is another, and it is often left implicit.
If roles, expectations, handoffs, and review steps remain the same, the saved time tends to be absorbed by more internal activity: additional reporting, more documentation, more maintenance. The process speeds up, but the public experience does not change much.
Libraries have seen this pattern before. Self-check machines existed for years without meaningfully changing circulation. The shift came when libraries redesigned spaces and workflows around them, making self-service visible, expected, and easy to use. The technology mattered, but the change in workflow and expectations mattered just as much.
AI presents a similar challenge. The benefit depends less on whether the tool works and more on whether leaders decide how work should be done differently.
Freeing time is not the same as using it well
It is easy to assume that saved time will naturally turn into better service. Experience suggests otherwise.
Consider meeting summaries. Many organizations already record internal meetings. Few of those recordings are revisited. AI-generated summaries may make those meetings easier to process, but they do not answer a more basic question: should the meeting exist in that form, at that frequency, with that level of documentation?
Without deliberate choices, AI risks making existing patterns more efficient without making them more useful. Faster summaries of low-value work are still low-value work.
The same applies across other tasks. If staff save several hours a week on drafting, search, reporting, or recurring preparation, those hours need a destination. More one-on-one patron support. Stronger outreach. Faster follow-up with partners. More time for instruction, programming, or local collection work.
Those outcomes do not emerge on their own. Someone has to decide that they matter and adjust workflows to support them.
Staff effort can move upward, but only if planned
One way to think about AI is that it shifts where effort is spent. Routine handling becomes easier. The remaining work depends more on judgment, context, and relationships.
In many service settings, AI supports lookup, routing, and summarization, while people focus on interactions that are harder to standardize. Libraries are no different. Routine processing can be accelerated. Relationship-building, instruction, partnerships, and public-facing decisions still depend on staff.
The opportunity is not just efficiency, but reallocation. Without planning, organizations often get the first and miss the second.
Information quality still sets the limit
There is a related constraint that AI does not solve on its own. It can search and summarize effectively, but it cannot reliably determine whether the underlying information is current or authoritative.
Most organizations carry a backlog of outdated or duplicated material. Libraries see this in legacy web pages, superseded policies, shared folders with overlapping versions, and documentation that no one clearly owns. AI makes that content easier to surface, which can also make inconsistencies more visible.
If multiple versions of guidance exist, an AI tool may draw from all of them unless there is clear ownership and a defined source of truth. Faster retrieval does not improve accuracy without curation.
Libraries already understand this principle. Collection maintenance has always involved selection, deselection, and authority control. The same discipline applies to internal information if AI is going to be useful.
Some of the most practical use cases are behind the scenes
Early results across sectors suggest that internal friction is often the best place to start. In one large field experiment, AI reduced time spent on email by about a quarter. In customer support, it has improved response speed by helping staff find and summarize information more quickly.
For libraries, that points to areas where capacity is quietly lost: board reporting, repeated documentation, website updates, policy questions, and coordination across disconnected systems. These are not high-visibility services, but they shape how much time staff have for public work.
Reducing this background load can have a real effect, provided the surrounding workflow changes with it.
Take internal reporting. Many libraries spend hours pulling together usage data, program attendance, circulation metrics, and digital activity from multiple systems. That work is necessary, but much of it is manual: exporting data, reconciling numbers, formatting reports, and checking accuracy.
If AI reduces the time required to gather and prepare that information, the workflow itself has not yet changed. The same report still exists. Staff simply have time they did not have before.
What happens next depends on how the library uses that time. In one approach, the saved hours are absorbed into more internal work. In another, the workflow is reshaped. Staff spend less time compiling data and more time interpreting it. They identify trends, adjust services, or bring clearer insights to board discussions and funding conversations.
The technology makes the first step possible. The change in how the work is done is what makes the difference visible.
What leaders should plan for
Library leaders can start with a practical sequence:
- Identify where staff time is being lost to repetitive, low-value work.
- Introduce AI in limited, well-understood tasks such as drafting, summarization, routing, documentation support, or internal search.
- Clean up the information behind those workflows by clarifying ownership, removing outdated content, and establishing a source of truth.
- Decide in advance where the reclaimed time should go and adjust roles and expectations accordingly.
That final step is the most important. It is also the easiest to skip.
Plan for success
Libraries do not need to resolve every question about AI before they begin. They do need to be clear about what they expect it to change.
The practical question is no longer whether AI can make work faster. That point is already well supported. The question is whether leaders are prepared to act on that change: to reshape workflows, adjust responsibilities, and direct the time that is freed toward visible improvements in service.
AI can make existing processes more efficient. Better outcomes depend on what follows.