How long does it take to organize a document? How many people must be asked to find an answer? Where does a case keep getting returned? These are often easier questions than “Should we adopt AI?” Start with one task, define inputs, completion criteria, and owners, then assess which part a tool could handle.
Choose a direction based on the work that is stuck
| Work challenge | Topic to explore |
|---|---|
| Teams have ideas but cannot prioritize investments | Adoption and transformation planning |
| Documents are organized, then repeatedly entered and assigned | Workflow automation and agents |
| Policies and product information are scattered and hard to search | Knowledge bases and RAG |
| Repeated inquiries are common and handoff to a person is unclear | AI customer-service planning |
| Tools work, but people still move results into the original system | System integration and customization |
| Account, data, and use permissions are unclear | AI governance and usage policies |
These areas can be planned together, but knowledge preparation, development, platform purchasing, and policy design should be scoped separately. Implementation and maintenance depend on confirmed requirements.
Check readiness for a pilot
Compare task frequency, current time, and the cost of errors. Then check available data, system interfaces, and who reviews results. For fixed-rule file conversion or notifications, ordinary automation may be sufficient; AI does not need to be the default.
The NIST AI RMF is a voluntary risk-management framework that can bring intended use and impact into planning. The pilot steps below are project-design suggestions, not a fixed procedure required by the framework.
Move from a small pilot to routine use
- Select a task with an owner and record current handling time, return reasons, and sample outputs.
- Confirm usable data, tool conditions, human decision points, and testing methods.
- Pilot regular and exception cases; separately record generation, review, editing, and delivery time.
- Decide whether to stop, adjust, or expand based on quality and cost, then plan accounts, document updates, and issue handoffs.
Plan data, access, and maintenance together
Confirm where data is sent, how long it is retained, and whether it may be used for provider training based on the actual service and contract. For system actions, separate query and edit permissions; define approval for external submissions, deletion, or material decisions.
For an initial discussion, prepare a process map, de-identified examples, user roles, volume, and system list. Do not put production passwords or keys in a requirements document. Possible deliverables include a requirements summary, pilot plan, acceptance records, and maintenance responsibilities. Development, deployment, and training need separately defined scope.
Estimating schedule and cost
Estimate data preparation, interface checks, prototyping, and testing, then ongoing model, platform, storage, and maintenance costs. Document quality, exception types, and limitations of existing systems may affect effort more than user count.
Frequently asked questions
We already bought a tool. Do we still need planning?
Review licenses, actual use, and unresolved tasks, then decide whether to adjust workflows, add training, or extend functions. Replacing the platform need not be assumed.
Can it connect directly to an ERP or CRM?
It depends on interfaces, data formats, test environments, and permissions. Products with the same brand can have different versions and licenses; check the actual environment.
Should we choose cloud or on-premises deployment?
Compare data conditions, usage, integrations, and maintenance capability. On-premises systems still need updates and computing capacity. For cloud services, review contract terms; deployment labels alone do not determine security or cost.
When is it suitable to expand use?
Results must be usable, and someone must update data, handle failures, and accept changes. A successful demo does not replace testing for routine operations.
For system governance, see AI management systems; for staff practice, see AI training.
Official references
These sources support the methods and scope. The workflow and deliverables in this page are project discussion suggestions.