Apollo Pharmacy transforms last-mile with Shipsy AI: Attendance AI, Selfie AI, OFE Assist, Address Intelligence
Apollo Pharmacy’s last-mile operation now runs on four purpose-built Shipsy AI agents that collectively hit a 62% rider attendance response rate, deliver 70% attendance prediction accuracy in pilot, auto-detect 17 incident types, and auto-process 3,000+ rider selfies per shift — with SLA breach warnings firing 30 minutes early. For an operation with 3,000+ riders, 150+ Ops Field Executives (OFEs), and two shifts per day, that AI layer is the difference between managing by scramble and managing by foresight.
Customer: Apollo Pharmacy. Industry: Pharma retail / last-mile. Region: India. Shipsy modules deployed: Attendance AI, OFE Assist (Live Incident Alerts), Selfie AI (Rider Compliance), AI Address Intelligence. Headline metric: 62% attendance response rate, 70% prediction accuracy, 17 auto-detected incident types, 3,000+ selfies auto-processed/shift, 30-min early SLA breach warning.
The challenge: four manual bottlenecks, every shift, every day
Apollo Pharmacy runs 3,000+ riders across two shifts daily, coordinated by 150+ Ops Field Executives (OFEs). At that scale, four specific bottlenecks were quietly eating ops capacity on every shift.
Rider no-shows were discovered only after shift start. Cluster managers had no visibility into likely attendance until riders either showed up or didn’t — by which point SLA risk was already baked in and there was no time to rebalance.
OFEs manually scanned dashboards for delays. 150+ OFEs, each watching their own cluster’s live dashboard for exceptions, was an enormous duplication of effort and still missed signals.
Every rider selfie was manually reviewed. Grooming and uniform compliance checks — thousands of photos per shift — went through human reviewers, which was completely unscalable.
Poor address data drove failed deliveries. For a pharmacy operation where medication timing matters, a failed delivery isn’t just an SLA miss — it’s a patient-impact event.
Apollo needed targeted AI for each of these, not a generic “AI platform.”
The solution: four purpose-built AI agents
Shipsy deployed four discrete AI components — each solving a specific last-mile bottleneck.
Attendance AI is the shift-foresight agent. A WhatsApp and Voice agent contacts every rider 30 minutes before shift, confirming attendance. Cluster managers get a live deficit report before SLA risk materializes — enough lead time to call in backup, rebalance routes, or activate floating riders. Pilot data shows 62% rider response rate and 70% attendance prediction accuracy, meaning leadership knows, with meaningful confidence, what the shift looks like before it starts.
OFE Assist (Live Incident Alerts) replaces the eyes-on-glass model. Instead of 150+ OFEs each watching their dashboards, Shipsy’s AI auto-detects 17 categories of incidents — delays, SLA-at-risk trips, exception patterns — and alerts the right OFE with context. The platform routinely surfaces issues 30 minutes before they become SLA breaches, giving OFEs the runway to intervene.
Selfie AI (Rider Compliance) is the grooming and uniform compliance agent. A grooming-score engine combines face-match verification with uniform check logic. Anything scoring above the 70% threshold is auto-approved — meaning 3,000+ selfies per shift flow through without human review. Reviewers only see the edge cases, which is the only scalable model at 3,000+ riders across two shifts.
Address Intelligence tackles the failed-delivery root cause. The system AI-scores every delivery address at CN creation, fills data gaps via LLM-based inference (landmark references, floor numbers, apartment identifiers), and if address confidence is still low, the AI calls the customer directly to resolve the gap before dispatch. That’s the full closed loop — not just validation, but active remediation of low-quality address data.
Together, the four agents form a purpose-built AI layer on top of Shipsy’s last-mile stack, each solving a discrete operational problem with a discrete mechanism.
The outcome: autonomous-ops signatures across four domains
62% rider attendance response rate. In pilot, nearly two-thirds of riders responded to Attendance AI’s pre-shift contact, producing usable attendance signal ahead of shift start rather than after it.
70% attendance prediction accuracy. The combination of responses, historical patterns, and no-response inference gives cluster managers a confident projection of shift staffing — enough to actually take action on, not just watch.
17 incident types auto-detected. OFE Assist replaces manual dashboard scanning with proactive alerts, and those 17 categories cover the bulk of what OFEs previously caught through eyes-on-glass monitoring.
3,000+ selfies auto-processed per shift. Selfie AI scales to the whole rider base across two shifts without proportional reviewer headcount. Reviewers now only see the edge cases the model flags for judgment.
30-minute early warning on SLA breaches. OFE Assist’s alerting cadence gives OFEs a meaningful window to intervene before customer impact — the operational signature of a system of action, not just a system of record.
For a pharmacy last-mile where timing is tied to patient outcomes, these aren’t productivity metrics — they’re reliability metrics.
What’s next
Apollo continues expanding the AI footprint across more clusters and refining the predictive models powering Attendance AI and OFE Assist. Near-term priorities include deeper Address Intelligence coverage, expanded incident-type libraries, and tighter integration between the four agents for compounding effects across the last-mile stack.