OPP-002 / Patient access / Research
The Opportunity:
Service Information Assistant
A Service Information Assistant that helps patients understand approved service information, identify the right administrative next step, and reach a person when they need one.
Simulated Source Consolidation: This is not a Healthcare System project or endorsement. Google Docs, Granola Notes, Slack Channels, and FinOps Dashboards are the assumed source systems for this demonstration. No records were actually retrieved from these systems; source contributions are fictional mappings of the synthetic workbook. Patient voices and impact figures are illustrative.
The patient experience
Purple = assistant support“I found the service page, but I still couldn’t tell which number to call.”
“Each person asked me to explain what we needed all over again.”
“I wanted a simple explanation—and a person if I was still unsure.”
Business Case
Problem StatementStaff spend 5,600 hours annually on information contacts. With 40% repeat contacts, repeated explanations consume 2,240 hours and $100,800 per year. The customer effort score is 5.5% reporting high effort.
ObjectiveWithin 12 months of rollout, reduce annual repeat-contact cost by 50%: $100,800 → $50,400. Return 1,120 staff hours / $50,400 in capacity annually. Lower customer effort from 5.5% to 3.8% and increase appointment completion from 6.0% to 8.5%.
Measurement basis
Synthetic annual planning estimates across six campuses.
- Contact volume: 105,000 users × 0.4 contacts = 42,000 contacts; 8 minutes each = 5,600 staff hours.
- Repeat-contact baseline: 40% of contacts = 16,800 repeats, 2,240 hours and $100,800 at $45 per hour.
- 50% reduction goal: Avoid 8,400 repeat contacts, returning 1,120 hours and $50,400 in employee capacity each year. Remaining repeat-contact cost: $50,400. This is capacity value, not cash savings.
- Customer effort: Reduce high-effort survey responses from 5.5% to 3.8%. Use the same question and comparable patient groups; report sample sizes.
- Timing: The six-week test informs a measured rollout; assess these goals after 12 months.
Opportunity Decision
The revised six-criterion framework produces 83/100. Evidence, evaluation, workflow adoption and operating ownership are mandatory requirements, not weighted criteria.
| Investment criterion | Weight | Rating / 5 | Points |
|---|---|---|---|
| Patient / workforce benefit | 25% | 5 | 25 |
| Strategic alignment & urgency | 20% | 4 | 16 |
| Economic value & financial sustainability | 20% | 3 | 12 |
| Data & technical readiness | 15% | 4 | 12 |
| Delivery feasibility / time to value | 15% | 5 | 15 |
| Reuse & partnership value | 5% | 3 | 3 |
| Total | 100% | 83 |
Points = weight × rating ÷ 5. The economic rating is a provisional assessment; other ratings are retained from the synthetic dataset. Delivery feasibility at 5/5 is optimistic and requires challenge. The score is a prioritization aid, not an approval.
Mandatory pilot checklist
- ✓Named owners and sponsorMet*
- ✓Security reviewMet*
- ✓Authorized data accessMet*
- —Clinical review for administrative scopeNot applicable*
- □Privacy reviewOpen
- □Data-use agreementOpen
- □Safety and escalation acceptancePartial
- □Evaluation and acceptance planPartial
- □AI versus non-AI comparisonPartial
- □Operating and adoption readinessPartial
*States in fictional records. No real approval is represented. Production additionally requires acceptance of OPP-019, the approved-content foundation, and vendor patient-data terms.
Funding recommendation
Seek approval for the $45,000 staged seed test described in the Business Case. Use the results to choose a non-AI guide, a constrained AI workflow, or no further investment. Re-estimate staffing and implementation costs after testing; no seed or pilot funding is currently approved.
Proposed seed money · not approved
Test the problem and two solution approaches before committing to implementation.
What the seed money buys
| Work package | Budget |
|---|---|
| Research & comparison design · 60 hours | $7,200 |
| Content preparation & ownership · 60 hours | $7,200 |
| Guide + RAG prototypes · 120 hours | $14,400 |
| Evaluation, accessibility & safety · 60 hours | $7,200 |
| Sandbox hosting, search & model usage cap | $3,000 |
| Participant recruitment & incentives | $3,000 |
| Contingency | $3,000 |
| Total seed envelope | $45,000 |
Bottom-up illustrative estimate: 300 hours at a $120 blended loaded hourly rate, plus $9,000 in expenses and contingency. Labor may be internal capacity rather than incremental cash. Assumes an existing approved platform, one service line, one campus content set and an English-language prototype. No EHR integration, live booking, patient records or production rollout. New procurement or platform setup requires re-estimation.
Weeks 1–2: validate the problem and prepare approved public content. Weeks 3–4: build a clearer guide and a constrained RAG prototype. Weeks 5–6: compare them in supervised usability sessions and a separate safety test set. Recruit up to 60 consenting participants using standardized tasks, without collecting medical histories. Session size supports directional learning, not proof of population-level conversion uplift.
Seed release & stop/go decision
Funding: $15,000 for discovery; $30,000 after scope, ownership, safeguards and evaluation approval. Patient-facing deployment requires all mandatory gates.
Proposed test targets: ≥80% task success, ≥10-point improvement over the non-AI guide, ≥95% factual accuracy, valid citations and no critical unsafe responses observed. Confirm targets before testing; assess accessibility, escalation and cost.
Decision: Choose the guide if results are comparable. Stop for content or safety failures. Success supports a controlled pilot—not full-scale approval.
Full-scale economics · existing planning case
| Total one-time implementation envelope | $180,000 |
| Annual AI operations | $85,000 |
| Annual contribution before AI costs | $249,375 |
| Annual contribution after AI costs | $164,375 |
| First-year net after implementation & operations | −$102,906 |
The $45,000 seed envelope replaces the former unscoped request in this page and is included within the $180,000 planning envelope, leaving $135,000 provisionally for later work. These allocations are assumptions, not approved commitments. Re-estimate after testing; the non-AI option needs its own cost case.
Annual contribution = 140,000 eligible opportunities × (8.5% target − 6% baseline) × 75% adoption × $250 net collectible revenue × 38% contribution margin. First-year net = annual contribution × 65% realization − $85,000 operations − $180,000 implementation. Productivity capacity below is excluded to avoid double counting.
Solution fit assessment
Does this need AI?
Plain-language pages, FAQs and a guided route to the right contact. No AI required; test this first.
Return approved source passages. Keyword search may suffice; semantic search adds AI retrieval if wording varies.
Retrieve approved information → draft a cited answer → check scope → answer or hand off. Useful if patients need explanations across several pages.
The task does not require independent planning or actions. No autonomous booking or changes to patient records.
No demonstrated need for custom model training or risk prediction. An ungrounded chatbot is unsuitable for current service facts.
Why these options overlap
RAG (retrieval-augmented generation) is how an answer uses source content; a workflow defines its controlled steps. They can be combined. An agent chooses its own sequence of actions. These are architectural choices, not mutually exclusive product labels.
Pattern references: Anthropic: workflows and agents; Microsoft: RAG. The fit assessment is our judgment for this demonstration.
Potential system impact
Less time explaining.
More time helping.
A shared information experience, adapted to each campus. Select a hospital to explore its contribution to the scenario.
Across the six-campus scenario
Impact assumptions & calculation
New scenario assumptions, not hospital operating data: 140,000 eligible annual opportunities × 75% adoption = 105,000 people reached, assuming one opportunity per unique person and no cross-campus overlap. The allocation across hospitals is illustrative, not based on hospital size or reported demand.
Baseline: 0.4 information contacts per user, of which 40% are repeat contacts. Target: avoid 50% of repeat contacts. Avoided contacts = people reached × 0.4 × 40% × 50% = 8% of people reached. At 8 minutes per contact and $45 per staff hour, the six-campus total is 8,400 avoided contacts, 1,120 hours and $50,400 in returned capacity annually. Baseline repeat-contact cost is $100,800; the target remaining cost is $50,400. This is capacity returned, not cash savings; implementation and AI costs are excluded, and this value is not added to appointment contribution.
Illustrative Healthcare System campuses: Orlando · Altamonte Springs · Celebration · Winter Park · East Orlando · Waterman. Campus names are illustrative; no adoption is implied.