Case Studies

See the framework in practice

These examples show how the same three-part filter and four risk levels play out in real operational work. The goal is not to copy each example literally, but to help you recognize the same pattern inside your own organization.

Start with pattern recognition, not technology

Every story below maps to the same filters taught in the frameworks: what is repetitive, what is intern-level, and what is not getting done. Each one also shows a risk level so you can see where to begin and how cautiously to expand.

Review the Frameworks

Scan the examples

Use the labels to scan by filter, risk, and category before reading the full stories.

Repetitive
Level 2: Monitoring & summarizing
Healthcare operations
Voicemail Summarization

AI transcribes, summarizes, and routes high-volume voicemail so teams can focus on follow-up instead of listening and sorting.

Jump to case study
Intern-Level
Level 4: External / end-user facing
Healthcare operations
Lead & Intake Routing

AI calls people back, gathers the few key facts, and routes each lead to the right team faster than manual follow-up.

Jump to case study
Not Getting Done
Level 4: External / end-user facing
Nonprofit
User Communication (E-Stam)

A weekly workflow turns product updates into clear bilingual user communication so an important message finally goes out consistently.

Jump to case study
Intern-Level
Level 1: Helping you work
Leadership
Data Analysis for Leadership Decisions

A leader used a free browser-based AI tool to challenge polished vendor reporting with a faster fact-based analysis.

Jump to case study
Repetitive
Level 3: Internal team interaction
Operations
The Multiplier Effect (Software Development)

Using AI for one step accelerated output, but real gains came only after the next bottleneck was identified and addressed too.

Jump to case study
Repetitive
Level 4: External / end-user facing
Healthcare operations
Bonus
Compliance Calls

AI handles routine reminder calls on schedule, then records responses for human review so repetitive outreach does not slip.

Jump to case study

Full case studies

Each example uses the same structure so you can compare the work, the implementation, and the nonprofit translation quickly.

Repetitive
Level 2: Monitoring & summarizing
Healthcare operations
1. Voicemail Summarization

The problem

Departments like HR, intake, and staffing were all receiving heavy voicemail volume. Each person had to listen to every message, decide what was relevant, and figure out next steps. The work consumed dozens of hours each week and was almost entirely repetitive.

What we did

We met with each department to understand the kinds of messages they receive, what details matter, and what a useful summary should include. Then we set up AI-powered transcription and summarization so each voicemail produced a clean summary with key points, follow-up actions, and clickable phone numbers or email addresses.

The result

Relevant information and likely next actions now land directly in each team member's inbox. Teams no longer spend time listening through everything just to sort and route messages, and dozens of hours of repetitive work were removed each week.

Time to set up

About an hour of conversation per department, plus technical setup.

Key insight

The value is not transcription by itself. The value is that each person gets the relevant information and possible actions delivered where they already work, instead of digging through every message manually.

Your version

Think about voicemails or inbound messages at a bikur cholim, chesed organization, development office, or family support program. If someone has to listen, interpret what is needed, and route it to the right person, the same pattern applies.

Intern-Level
Level 4: External / end-user facing
Healthcare operations
2. Lead & Intake Routing

The problem

A team member was spending 15 to 20 hours a week calling inbound leads back to figure out what kind of lead they were, whether they were patients or caregivers, and what they actually needed before routing them to the correct department.

What we did

We began implementing an AI system that calls leads back automatically, has a natural intake conversation, collects the key information, identifies the lead type, and routes the person to the right team.

The result

Around 20 hours per week of staff time are freed up, and leads receive faster callbacks, which improves service quality as well as operational efficiency.

Time to set up

Several weeks of technical work.

Key insight

This is classic intern-level work: it matters, but it does not require deep judgment. It mainly requires a short conversation, a few questions, and reliable routing.

Your version

This same pattern shows up in food pantry intake, volunteer coordination, event RSVPs, donor inquiry follow-up, and many other nonprofit workflows where someone is manually gathering a few facts and routing the request.

Not Getting Done
Level 4: External / end-user facing
Nonprofit
3. User Communication (E-Stam)

The problem

E-Stam needed to keep its users informed about changes, fixes, and improvements, but those updates were not going out. The development team was busy building, and no one had the bandwidth to gather what happened, write it clearly, translate it, and send it.

What we did

Developers only mark relevant items in the project management system. Once a week, an automated workflow gathers everything marked that week, uses AI to draft a clear update, translates it into Hebrew, and sends it for review before release.

The result

A communication that mattered but was not happening at all now goes out every single week. The organization serves its users better without pulling developers away from core work.

Time to set up

A couple of hours to build the workflow.

Key insight

This is the not-getting-done category in action. AI did not replace anyone. It made an important recurring communication possible where there was previously no bandwidth.

Your version

Monthly newsletters, donor updates, volunteer thank-you messages, and program impact summaries often already exist in scattered operational systems. If the pattern is stable, AI can help turn that raw material into consistent communication.

Intern-Level
Level 1: Helping you work
Leadership
4. Data Analysis for Leadership Decisions

The problem

A vendor's reporting looked polished and positive, but the people on the ground felt something was off. Historically, proving that meant spending a full day pulling numbers, building spreadsheets, and looking for patterns, or hiring outside help.

What we did

Four relevant reports were pulled into Claude in a browser. The relationships between the reports were explained, the questions were made explicit, and the desired output was specified.

The result

A complete analysis came back quickly, followed by about 30 minutes of spot-checking and review. What would have taken a full day was done in under an hour, giving leadership concrete evidence instead of a gut feeling.

Time to set up

About 45 minutes total, including input and review.

Key insight

For internal decision-making, Level 1 work can be remarkably accessible. A clear problem, a few reports, and a free tool can often replace a much heavier analysis process.

Your version

Board meetings, grant reporting, impact reviews, and vendor management all create moments where leaders know something is wrong or need a clearer case but do not have analyst capacity. This is a practical starting point you can try immediately.

Repetitive
Level 3: Internal team interaction
Operations
5. The Multiplier Effect (Software Development)

The problem

AI started helping produce software much faster than a person could type, but the additional code still needed review and testing. The bottleneck shifted from writing code to reviewing it safely and thoroughly.

What we did

After recognizing the new constraint, AI was introduced to support review and testing as well, so the entire workflow improved instead of just one step.

The result

The same team, working the same hours, reached roughly three to four times the output because the full chain was improved rather than only the first task.

Time to set up

Iterative, with weeks of refinement.

Key insight

When AI speeds something up, ask what pressure it puts on the next step. If the downstream system cannot absorb the new volume, you have only moved the pile.

Your version

Volunteer intake, donor acknowledgment, matching, reporting, scheduling, and follow-up all have the same end-to-end dynamic. The real question is not only what to automate first, but what constraint that change will expose next.

Repetitive
Level 4: External / end-user facing
Healthcare operations
Bonus
6. Compliance Calls

The problem

Routine compliance reminders about deadlines, renewals, and document submissions had to be delivered consistently. The task was necessary, repetitive, and easy to fall behind on because no one enjoyed doing it.

What we did

AI was configured to place the calls, deliver the reminder in a natural conversation, record the response, and place the results in a spreadsheet for human review.

The result

The calls now happen consistently and on time, while humans only need to review the outcomes instead of making every reminder call manually.

Time to set up

Setup plus scripting the call flow.

Key insight

Scheduled outreach that follows a standard script is often a strong AI candidate, but relationship risk still matters. Internal stakeholders may be appropriate; primary client relationships may not be.

Your version

Volunteer shift reminders, paperwork follow-up, donor renewal prompts, and board meeting nudges all fit this pattern when the outreach is routine and the script is predictable.

Cross-cutting themes

The value of these examples is not only the stories themselves. It is the recurring pattern that helps you spot strong, lower-risk starting points.

The three-part filter

Repetitive

Same work every week, requiring time more than judgment. Examples here include voicemails, routine outreach, and workflow bottlenecks that respond well to systematization.

Intern-Level

Important work that still does not require high-level expertise. Intake routing and leadership analysis both fit because the process can be explained clearly and reviewed afterward.

Not Getting Done

Work that matters but keeps slipping because no one has bandwidth. AI is often most valuable when it makes the important-but-neglected possible again.

The four risk levels

Level 1

AI helps you work directly, such as reviewing reports for internal decisions.

Level 2

AI monitors and summarizes incoming information, such as voicemail processing.

Level 3

AI supports internal team communication or workflows, such as internal reminders and internal bottleneck reduction.

Level 4

AI touches external people or critical relationships, such as lead callbacks or scripted outreach. These uses need the most caution and the slowest rollout.

Core principle

Start at Level 1, build confidence, and move upward deliberately. When AI touches people you serve, begin with a small group, watch it closely, fix issues, and only then expand.

Go back to the frameworks