BVDLC Case Studies

Challenge

• Went from serving 1K → 10K families overnight; intake and scheduling collapsed under the spike
• Volunteers ran ad-hoc spreadsheets with zero traceability or compliance trail for donor and regulator review
• No way to onboard surge volunteers fast — every new helper needed days of shadow training before they could safely take a shift

BVDLC Approach

• Phase 0 locked the KPI on "meals delivered per day" instead of features shipped — pulled the team out of the shiny-tool conversation
• Phase 1 prototyped SMS request flows + dispatcher dashboards in three days; killed two weaker designs before architecture started
• Phase 2 architecture emphasized auditable flows so donors and regulators could trace every delivery without manual stitching

Outcome

• Delivery cycle time shrank from 5 days → 12 hours
• Compliance-ready reports auto-generated from the context folder — no quarterly scramble
• Surge volunteers onboarded in 15 minutes using the handover docs Phase 4 produced

Artifacts to copy

• Phase 0 template tuned for social-impact KPIs
• Context-folder structure for highly regulated nonprofits
• SMS + dispatcher console pattern with retry-and-escalate flows

Challenge

• Manual matching capped capacity at ~600 students per semester; demand was nearly double that
• Compliance rules (background checks, parental consent) added weeks to every release
• Auditors couldn't follow the chain from "policy section" to "matching rule that ran in production"

BVDLC Approach

• Phase 0 set the KPI "students matched within 48 hours" plus the strict policy constraints up front
• Phase 2 architecture captured every decision as an ADR linked to the policy section it implemented — built for auditors, not for engineers' egos
• Phase 4 paired engineers with AI prompts for matching logic and consent forms; human review owned the policy edge cases

Outcome

• Matching capacity doubled (1,200 students) with the same headcount
• Audit package creation dropped from 3 weeks → 3 days
• Stakeholders traced every rule back to the policy section it came from — in minutes, not a special-request email

Artifacts to copy

• ADR template referencing policy sections
• Phase 3 acceptance criteria mapping user stories to compliance tests
• Monitoring dashboards focused on equity + turnaround time

Challenge

• Every animal needed an engaging bio before listing — staff wrote each one by hand, 15–25 minutes per animal
• Intake spikes (post-holiday surrenders, shelter transfers) created multi-day listing backlogs while animals waited unseen
• Bio quality drifted by author — some warm and specific, others copy-pasted templates that buried the animal's personality

BVDLC Approach

• Phase 0 locked the KPI on "hours from intake to live listing" — not "bios written" — because the point is getting animals seen by adopters
• Phase 1 prototyped phone-photo → AI vision tags (breed cues, age, coat, posture) → LLM drafts a bio in the shelter's voice from the context folder
• Phase 4 used the three-touch rule: AI drafts → staff reviewer edits one line → system logs who approved what changed, for accuracy audits

Outcome

• Bio creation dropped from ~20 minutes to ~30 seconds of staff time per animal
• Median intake-to-live-listing time went from 2.4 days to under 6 hours; adoption inquiries rose 38% in the pilot quarter
• Volunteers went back to walking dogs and socializing cats — the work they signed up for

Artifacts to copy

• Context-folder template for shelter voice + brand-safe language rules
• Phone-capture → AI-tag → LLM-draft prompt chain with prohibited-phrasing list embedded
• Three-touch approval log schema for accuracy audits and adopter complaints