BucketAI · Population health
You have 22,000 patients. Who do you call first?
Most hospitals sit on years of patient records and treat them as an archive. BucketAI reads that archive as a worklist — sorting every person into health buckets, ranking who needs contact this week, and telling the front desk exactly what to offer them.
Your patient list is a waiting room. BucketAI decides the order.
47 patients need attention this week. Reaching all of them is ₹4.2 L in addressable care.
See all 14 cohorts →
How it works
Four steps, and then it does it again.
Every patient runs the same loop — and the loop closes, which is the part most population-health tools skip.
Chart, labs, prescriptions, lifestyle inputs, family history, age and location — whatever the record actually holds for that person.
Each patient gets a 0–1 score for every possible health bucket, from cardiovascular risk to a sleep study to an overdue dental check.
The top-scoring buckets become that patient's care plan, with the specific service to offer and a reason a human can read aloud on the phone.
When the service happens, the result is labelled and fed back. Confirmed buckets reinforce; refuted ones correct. The list gets sharper every cycle.
Explainability
Every bucket comes with its reasons.
A score nobody can interrogate is a score nobody should act on. Open any patient and BucketAI shows which facts pushed them into a bucket, and by how much — so a doctor can overrule it in ten seconds if it's wrong.
- Contributing factors, weighted — the actual chart values that moved the score, each with its own contribution.
- A plain-language next action — not "risk score 0.88", but "schedule a cardiology consult this week, and here is why".
- Confidence on the face of it — low-confidence cases are routed to a clinician for review rather than dialled by the call centre.
- Care plan you can hand over — the whole bucket set for a patient, printable, ready for the follow-up call or the family conversation.
What to do next. Schedule a cardiology consult this week — high LDL, stage-2 hypertension and a family history of MI.
The buckets
Twenty-seven ways a patient can need you next.
A bucket is not a diagnosis — it is a reason to make contact. Each patient is scored 0–1 against every one of them, and the ones that clear the threshold become their care plan. Most people land in three to seven.
The loop that closes
Most population tools stop at step three.
Finding cohorts is the easy half. What separates a working programme from a dashboard is whether the outcome of the call comes back and changes the next list.
Chart, labs, prescriptions, lifestyle inputs, family history, age. Whatever the record actually holds — no separate data warehouse.
268 featuresA 0–1 score against all 27 buckets, with the contributing chart values weighted and kept for display.
27 scores / patientTop buckets become a care plan with a specific service and a reason a receptionist can say out loud.
3–7 buckets typicalOutreach books a real appointment on the real calendar against the real UHID. No CSV, no re-keying.
Booked in HISService rendered, outcome labelled, fed to the next training cycle. Confirmed reinforces; refuted corrects.
Back to step 1Every registered patient scored, not a sample. The cohorts you never knew you had are the point.
Tracked per cohort, so you learn which reasons people actually say yes to — and which scripts to retire.
Care actually rendered, traced back to the cohort that generated it. Everything before this is a promise.
Guardrails
A recall engine is one bad prompt away from being a sales list.
Population health done badly is upselling with a stethoscope on. These are the rules that keep it clinical.
Model operations
It tells you when it is getting worse.
Any model degrades as your population changes. The difference is whether it admits it. BucketAI watches its own drift, queues the cases it isn't sure about, and asks a clinician instead of guessing.
A fleet, not a black box
Eight specialty models behind one router, each versioned, each with its own accuracy tracked separately — so "the model is accurate" can be broken down to which bucket, this month — and which one is drifting.
Drift monitoring
Population shift is measured per bucket against the training distribution. Cross the threshold and the model is flagged for retraining before its recommendations quietly rot.
Ambiguous cases go to a human
Cases the model is genuinely unsure about are queued for clinician labelling rather than pushed to a call list. Those labels are also the most valuable training data you have.
Outcomes close the loop
After the service is rendered, the bucket is confirmed or refuted. That confirm rate — reported per bucket, not as one flattering average — is how you judge whether to keep trusting it.
The funnel
Identified. Booked. Actually delivered.
Population health is judged on care delivered, not on cohorts discovered. BucketAI reports the whole funnel, including the part that didn't convert.
Sorted into buckets across the full panel, with the addressable care quantified per cohort.
Outreach from the recommendation itself — appointment booked, care plan sent, no re-keying.
The only number that counts, tracked back to the cohort that generated it — and fed to the next training cycle.
Interface vignettes show illustrative data, not real patients or records. Cohort sizes and values shown are examples, not results from a named hospital.
Works with
The panel it reads, and the care it triggers
BucketAI is the only engine that looks at everybody rather than the patient in front of you — but it acts through the same systems.
Every registration, encounter, lab and prescription HIS has recorded is the training and scoring surface. No export, no separate data warehouse.
Explore → 📅Hands off to It books back into HISA cohort is worthless if acting on it means re-keying. Outreach creates real appointments on the real calendar, against the real UHID.
Explore → 📱Hands off to Care plans travel homeThe bucket set for a patient becomes something they can actually read — trends, what needs attention, and the date they are expected back.
Explore →Book a 20-minute demo
See your hospital run on Harvis.
No slide decks. We'll walk you through a live OPD consult, a discharge and a billing cycle — on your specialty, in your language.