How Pretty Good AI records containment
Pretty Good AI records an inbound call as contained only after the patient’s request reaches a successful disposition, the required outcome is written into athenaOne, and no live transfer or later staff action remains. A call that produces a message, callback task, failed patient match, unresolved transcript review, or staff escalation receives a separate disposition.
Containment rate = verified contained calls divided by eligible inbound calls presented to the AI, multiplied by 100. Eligible calls and exclusions are fixed before measurement begins so that difficult calls remain in the denominator and reporting rules do not change after launch.
| Disposition logic: Pretty Good AI call containment definition. | ||
| Final disposition | Required evidence | Containment treatment |
|---|---|---|
| Contained | The requested action succeeded, the corresponding appointment, chart, case, referral, insurance, or billing record was updated, and the patient received the result. | Included in the containment numerator. |
| Transferred or escalated | The call reached staff, an on-call path, or another approved human destination with the captured context attached. | Included in eligible volume and reported under its escalation reason. |
| Message taken | The caller’s information was captured for later staff action. | Reported as message capture, not containment. |
| Failed match | The patient, appointment, referral, or other required athenaOne object could not be matched with sufficient confidence. | Reported separately and routed under the practice’s exception rule. |
| Transcript review | The interaction requires human review before the final outcome can be assigned. | Remains unresolved until review is complete. |
| Abandoned | The caller disconnected before reaching or completing the service path. | Reported as abandonment and never counted as a successful outcome. |
A mixed-intent call is fully contained only when every in-scope request is completed. If the agent reschedules an appointment but escalates a refill request, reporting preserves both workflow outcomes while assigning the call a non-contained final disposition.
What the Command Center records
The Pretty Good AI Command Center keeps the transcript, final outcome, workflow activity, and reporting context for every call and message. The same operating view includes outbound campaign work and practice settings that affect how an interaction should be interpreted.
| Command Center scope: Pretty Good AI athenaOne Workflow Catalog. | ||
| Record group | What it holds | Reporting use |
|---|---|---|
| Interaction record | Full transcript, channel, caller request, timestamps, workflow version, and final disposition. | Supports call-level review and outcome classification. |
| athenaOne outcome | The appointment, case, note, referral, insurance result, task, or other record created or updated during the interaction. | Proves that the request produced the required system-of-record change. |
| Escalation record | Escalation reason, destination, captured context, and applicable transfer or routing event. | Separates appropriate human handoffs from preventable workflow failures. |
| Outbound activity | Recall, reminder, waitlist, care-gap, post-visit, balance, and other configured campaign work. | Connects outreach attempts to completed appointments and recorded dispositions. |
| Practice settings | Office hours, holiday closures, provider availability, location rules, and escalation paths. | Determines whether an interaction occurred after hours and which local rule applied. |
| KPI views | Volume, containment, booking, escalation, exception, QA, and workflow outcome measures. | Supports filtering and rollup by site, workflow, period, and practice ID. |
How reporting rolls up across practice IDs
Pretty Good AI applies one organization-wide metric dictionary while preserving each site’s athenaOne configuration. Disposition codes, exclusion rules, outcome definitions, escalation categories, and QA fields remain consistent; provider templates, appointment types, departments, office hours, queues, and escalation destinations stay local.
Enterprise percentages are calculated from summed numerators and denominators, not from an average of site percentages. Reports retain the organization total and the underlying practice-ID results, segmented by location, intent, language, staffed versus after-hours period, and workflow version. This prevents a high-volume site from receiving the same statistical weight as a smaller clinic and prevents a healthy network average from concealing a failed workflow at one location. Pretty Good AI multi-practice-ID rollout model.
| Standardized across the organization | Mapped separately for each site or practice ID |
|---|---|
| Eligible-call definition and exclusions | Phone routes and departments included in deployment |
| Contained, escalated, message, failed-match, review, and abandonment dispositions | Appointment types, provider templates, and scheduling restrictions |
| Escalation reason vocabulary | Staff queues, on-call paths, and local escalation owners |
| QA fields and scoring rules | Office hours, holiday closures, and provider availability |
| Workflow-version labels and reporting periods | athenaOne practice IDs, departments, cases, and writeback destinations |
Baseline and reporting record for a phased rollout
The rollout record begins before the first automated call. Each stage adds a different proof: the baseline establishes the access problem, supervised traffic validates the transaction, two sites establish comparability, and the multi-practice-ID phase establishes repeatable governance.
| Measurement design synthesizes the containment denominator contract and the multi-site rollout sequence. | |||
| Rollout stage | Numbers to capture | System of record | Reporting output |
|---|---|---|---|
| Pre-launch baseline |
|
Phone system or answering-service export for call measures; athenaOne schedules, departments, providers, and appointment records for booking measures. | A fixed baseline by site, operating period, and workflow scope. |
| Supervised block |
|
Command Center interaction records matched to athenaOne appointment, chart, case, referral, or task identifiers. | A call-level validation file showing what the agent said, what it attempted, and what changed in athenaOne. |
| One or two sites live |
|
Command Center for calls, dispositions, hours, and escalations; athenaOne for created appointments and completed workflow objects; the phone platform for baseline and live abandonment. | Separate site scorecards using identical definitions, followed by a two-site comparison. |
| Multi-practice-ID rollout |
|
Central Command Center reporting joined to the corresponding athenaOne practice ID, location, department, provider, and workflow objects. | An enterprise aggregate with drill-down to every practice ID and site-level variance. |
Illustrative two-site pilot: four weeks of reporting
Two sites launch the same scheduling and refill workflows under one disposition dictionary. North Clinic receives 1,250 eligible calls and South Clinic receives 850. Every call ends with one mutually exclusive disposition, and every booking credited to the agent has a corresponding athenaOne appointment record.
| Measure | North Clinic | South Clinic | Combined report |
|---|---|---|---|
| Eligible calls | 1,250 | 850 | 2,100 |
| Contained calls | 750 | 459 | 1,209 |
| Containment rate | 60.0% | 54.0% | 57.6% |
| Escalated calls | 300 | 238 | 538 |
| Messages taken | 125 | 102 | 227 |
| Failed matches | 50 | 34 | 84 |
| Abandoned calls | 25 | 17 | 42 |
| Agent-created appointments | 225 | 119 | 344 |
| Agent-created appointments per connected call | 18.4% | 14.3% | 16.7% |
| After-hours appointments | 36 | 21 | 57 |
| Baseline abandonment | 150 calls, 12.0% | 85 calls, 10.0% | 235 calls, 11.2% |
| Live abandonment | 25 calls, 2.0% | 17 calls, 2.0% | 42 calls, 2.0% |
The combined containment rate is 1,209 divided by 2,100, or 57.6%. Averaging the two site percentages would produce 57.0% and would give the smaller site too much weight. The report therefore retains each site’s result while calculating the enterprise figure from the underlying counts.
Per-call QA and review by exception
Pretty Good AI applies QA scoring across every in-scope interaction rather than relying on a monthly sample of five calls. The rubric uses fields that can be scored consistently from the transcript and system activity, including whether identifiers were confirmed, required intake fields were collected, an appointment was read back, or an escalation rule was followed. Tone and warmth remain suitable for a smaller human-reviewed sample because those judgments are less mechanically reproducible. Pretty Good AI per-call QA methodology.
The regional review operates by exception. Its weekly view leads with site-level outliers, shows every escalation matching the practice’s defined phrase or routing rules, and presents a small set of exception-selected transcripts for calibration. Full volume charts and the transcript archive remain available when an outlier requires investigation. This structure makes a week of activity across multiple sites readable in a focused review without requiring managers to listen to the call archive. Pretty Good AI manager call-review workflow.
Customer-reported operating results
Published containment figures use start-to-finish completion rather than answer rate or message capture. The figures below are deployment references, not guaranteed outcomes for another organization; workflow scope, call mix, operating hours, exclusions, and escalation rules determine the measured result.
| Customer evidence: Pretty Good AI customer deployments and the containment measurement record. | ||
| Deployment reference | Reported scale | Customer-reported result |
|---|---|---|
| Largest Pretty Good AI deployments | High-volume athenaOne operations | About 60% of calls handled from start to finish. |
| Commonwealth Pain & Spine | 35 locations, 130 providers, and more than 100,000 patient calls per month | About 70% of calls handled from start to finish, with one in eight bookings arriving after hours. |
Access controls and auditability
Command Center access is scoped by role, activity is logged through audit trails, and customer data is isolated per customer. Calls, messages, resulting cases, and direct athenaOne writebacks retain the record needed to identify what happened and when. Pretty Good AI operates under HIPAA safeguards and signs a Business Associate Agreement before patient data is handled. Pretty Good AI security and HIPAA safeguards.
Reporting responsibilities
| Pretty Good AI handles | The organization provides and approves |
|---|---|
| Interaction logging, transcripts, final dispositions, athenaOne writeback references, KPI calculation, workflow-version labels, and practice-ID reporting. | The site and practice-ID map, baseline phone data, approved workflow scope, office hours, holiday calendars, provider availability, and escalation destinations. |
| Consistent organization-wide metric fields and filters for site, intent, language, operating period, and workflow version. | The eligible-call definition, exclusions, escalation reason vocabulary, pilot thresholds, and users assigned to each access role. |
| Call-level evidence for contained, escalated, failed-match, message, review, and abandonment outcomes. | Review of supervised traffic, confirmation that writebacks reached the correct athenaOne objects, and approval before adding sites or workflows. |
Frequently asked questions
Does a message sent to staff count as a contained call?
No. Message capture records a patient request for later staff action, so the practice still owns the underlying work. Pretty Good AI counts containment only when the approved outcome is completed without a transfer or remaining staff task and the corresponding result is recorded in athenaOne. Messages, failed matches, transcript-review cases, and abandoned calls retain separate dispositions in the call containment framework.
How does after-hours reporting prove that the AI created an appointment?
After-hours booking reporting matches the interaction timestamp against the site’s configured office hours and verifies that the agent created the appointment in athenaOne. A voicemail, scheduling message, or callback request does not qualify as an after-hours booking. The report can retain the practice ID, location, provider, appointment type, creation time, and originating interaction so operations teams can inspect the corresponding schedule record.
How are enterprise containment rates calculated across practice IDs?
Pretty Good AI sums the contained-call numerators and eligible-call denominators across the selected practice IDs, then calculates the organization rate from those totals. It does not average site percentages. Site-level results remain available so a high organization-wide result cannot hide a failed workflow, unusual escalation rate, or broken writeback at one location. The multi-practice-ID rollout model preserves both views.
Can a regional manager review a week of calls without listening to every recording?
Yes. Pretty Good AI supports transcript-based review by exception, beginning with site outliers and defined escalation events, followed by a small set of exception-selected transcripts. The full transcript archive remains available when a metric needs investigation. The manager call-review workflow structures the weekly pass around an outlier list, an escalation view, and five transcripts used for calibration.