- Company
- Care to Stay Home, a California-based in-home care provider
- Industry
- In-home care
- Core AI workflow
- Automatic transcript intake, AI categorization, transcript analysis, compliance-risk detection, and workflow routing
- Input
- Call transcripts automatically saved to a designated folder; Google Voice is one source of the underlying calls
- Broader systems
- Zenphi, Google Workspace, Google Voice, Salesforce, QuickBooks
- Results
- 700 hours saved per month, real-time visibility into compliance-critical calls, reduced legal exposure, and less manual data entry
Care to Stay Home did not just automate transcription. It automated what happens after a conversation becomes text. The workflow takes a newly saved call transcript, uses AI to understand what the conversation is about, categorizes it, analyzes it for defined signals and risks, and then uses workflow logic to decide what needs attention and who should receive it.
How Care to Stay Home uses automation in real operations
Parker Wells, COO of Care to Stay Home, explains how the team uses Zenphi to reduce administrative work, improve visibility, and scale operations without growing overhead at the same rate.
More than 1,000 calls a day — each carrying operational information
Care to Stay Home is a California-based in-home care provider. Phone conversations are part of the daily operating system: they can contain scheduling changes, caregiver reports, patient incidents, leave requests, and other information that may require action from different teams.
At more than 1,000 calls per day, the problem is not simply storing the recordings. The challenge is turning what was said into structured information quickly enough for the business to use it.
Critical information was trapped inside unstructured conversations
Before automation, Care to Stay Home’s team spent three to four hours a day listening to recorded calls and manually entering notes into internal systems. Across the organization, manual data-entry activity was estimated at 450–700 hours per month.
The volume created several problems at once:
Staff had to listen, interpret, and document calls one by one before the information could be used elsewhere in the operation.
A routine coordination call and a call mentioning a patient fall or caregiver injury could enter the same manual review process.
Compliance-sensitive information could be delayed or missed when detection depended on a person listening to the recording and recognizing the significance.
Information captured in conversations did not automatically become categorized, searchable workflow data that could be routed, monitored, and analyzed.
AI reads, categorizes, and analyzes each transcript before the workflow acts
The workflow starts when a call transcript is saved automatically to a designated folder. From that point, staff do not have to open every transcript and decide manually what it contains. Zenphi uses AI to interpret the unstructured text and turn it into structured information the rest of the workflow can use.
A transcript from a recorded call is saved to the designated folder and triggers the workflow. Google Voice is part of the upstream call process, but the automation begins with the transcript itself.
The transcript is passed to the AI step so the system can interpret the conversation in context rather than rely on fixed keywords or a rigid form structure.
The model assigns the transcript to the relevant operational category, turning an unstructured conversation into a structured workflow variable. This is the same pattern used in AI classification automation: AI interprets the input, then workflow logic acts on the classification.
The workflow looks for context that may require attention, including high-risk indicators such as “injury,” “fall,” or “leave request.” The goal is not simply to label a call, but to identify what the organization may need to do about it.
Calls requiring attention are flagged and escalated to the appropriate department. Routine transcripts can continue through the normal process without consuming the same level of human attention.
AI classification is most useful when the category is not the final output. In this workflow, categorization and analysis determine what happens next — whether a transcript can continue automatically or needs to be surfaced to a person. For a deeper explanation of that pattern, see AI Classification Automation.
From listening to every call to reviewing what actually needs attention
| Measure | Before Zenphi | With the AI workflow |
|---|---|---|
| Transcript review | Staff manually listened to calls and entered notes | Transcripts are processed automatically as they arrive |
| Understanding the call | A person interpreted every conversation | AI categorizes and analyzes the transcript |
| Manual workload | Estimated 450–700 hours per month of manual data entry | 700 hours per month saved across operations |
| Compliance signals | Potentially important information depended on manual review | Defined risk indicators are surfaced in real time |
| Routing | Staff had to recognize the issue and pass it on | Relevant calls are automatically flagged and escalated |
| Operational data | Information remained embedded in calls and manually entered notes | Conversation data becomes structured and usable inside workflows |
The value came from analysis and action, not transcription alone
Converting speech to text removed only the first layer of manual work. The larger operational gain came from using AI to interpret each transcript at scale: identifying what the call is about, detecting potentially important signals, and passing those results into an automated workflow.
That changed the role of the team. Instead of treating every call as something a person had to inspect, human attention could be concentrated on conversations that actually required judgment, follow-up, or escalation.
“We want to grow the organization without growing the cost of our operations at the same rate. Zenphi gives us one platform where we can handle healthcare workflows, call analysis, compliance visibility, and system integrations.”
The same platform supports the rest of the operation
Call analysis is the clearest AI use case in the Care to Stay Home story, but it sits inside a wider automation strategy. The organization also uses Zenphi for operational workflows and system integrations, including a Salesforce–QuickBooks synchronization designed to replace more fragile third-party connections.
The pattern applies to any high-volume stream of unstructured text
A call transcript is only one example. The same AI workflow pattern can start with an email, document, form response, support ticket, or other unstructured input: AI reads it, classifies it, extracts or analyzes what matters, and passes the result to deterministic workflow logic.
That distinction is important. The useful output is not a summary sitting in another tool. It is structured information that can trigger routing, escalation, human review, notifications, system updates, or another downstream process. Learn more about the model on the AI classification automation page.
Questions about this case study
What does AI do in the Care to Stay Home workflow?
AI reads each call transcript, categorizes the conversation, and analyzes the text for information or risk signals that may require action. The classification and analysis results then become inputs to the workflow logic that determines what happens next.
Does the workflow depend on Google Voice?
Google Voice is part of the source call process in this case, but it is not the core of the automation. The workflow begins when the call transcript is saved automatically to a designated folder. From there, Zenphi processes the transcript with AI and routes the resulting information.
What types of risk signals can the AI identify?
The Care to Stay Home case specifically mentions indicators such as “injury,” “fall,” and “leave request.” When relevant signals are detected, the workflow can flag the transcript and escalate it to the appropriate department.
How much time did Care to Stay Home save?
The reported total impact is 700 hours saved per month across operations. Before automation, manual data-entry activity associated with the process was estimated at roughly 450–700 hours per month.
How is this different from using a transcription tool?
A transcription tool produces text. This workflow continues from there: AI interprets and categorizes that text, analyzes it for defined signals, and passes structured results into workflow logic so the organization can route, escalate, log, or act on the conversation automatically.
Have unstructured data your team still reviews by hand?
See how Zenphi can use AI to classify and analyze incoming content, then connect the result directly to the workflow that acts on it.
Schedule a demo