Decisions that can be reconstructed years later.
Public-sector decisions get challenged, sometimes years later. AnantState records the evidence, the reasoning and any disagreement behind each conclusion, so a decision is defended with a record rather than with memory.
In plain termsDecisions you can reconstruct and defend.
A case, from arrival to a defensible decision
The entities are the things a service is delivered through: a case, a claimant record, an entitlement, a service point, a payment, a referral. The platform keeps their state and the team keeps the decision, and the record connects the two.
A case, from arrival to a defensible decision
- A case arrivesLane: Case teamLeads to: Backlog and risk state
- Backlog and risk state: Updated as events arriveLane: AnantStateLeads to: Ranked by value at stake
- Ranked by value at stake: With evidence and conflictLane: AnantStateLeads to: Verifies, refers or decides
- Verifies, refers or decides: A person, with the case in front of themLane: Case teamLeads to: Reasoning on record
- Reasoning on record: Evidence, decision, who, whenLane: AnantState
What makes a decision defensible
- Evidence, provenance and conflict travel with every conclusion, which is what makes a decision challengeable and defensible.
- Receipts and attribution cover decisions taken and decisions declined.
- Priced interventions. Verify, refer, request evidence or escalate, each with a cost, compared against the cost of a wrong payment or a delayed entitlement.
What moves in a service
| State | Baseline drivers | Unexpected drivers |
|---|---|---|
| Backlog pressure | Arrival rate, staffing, policy cycle | Policy change, surge, system outage |
| Error and fraud posture | Historical rates, verification depth | Organized activity, insider patterns |
| Service load | Population, season, channel mix | Migration, local economic shock |
| Payment integrity | Award rules, cycle | Duplicate payments, undeclared change of circumstance |
Where it goes wrong today
- Decisions are held to account that cannot be reconstructed, so the organization is defensible only by assertion.
- Backlog is managed by adding people, because the shape of the queue is unknown: throughput is visible, but which cases are about to become difficult is not.
- Fraud and error work is retrospective, with intervention arriving after payment, which is both more expensive and more politically fraught.
A worked example
Ten thousand cases, and verification applied to the ones most likely to go wrong.
Round, illustrative figures for cases: they show the arithmetic, leave out the subscription, and are for you to change in the calculator.
| Input | Value | What it stands for |
|---|---|---|
| Cases in scope | 10,000 | Everything the team is responsible for |
| Value at stake per one | $2,000 | Exposure on one case from a wrong or delayed decision |
| Cost of the action per one | $80 | Verifying evidence or referring the case |
| Share acted on | 12% | The cases most likely to become difficult: 1,200 of 10,000 |
| Modeled, USD | Pessimistic | Expected | Optimistic |
|---|---|---|---|
| Share of value at stake avoided | 5% | 12% | 25% |
| Cost to act, in total | $96,000 | $96,000 | $96,000 |
| Avoided loss | $120,000 | $288,000 | $600,000 |
| Net position | $24,000 | $192,000 | $504,000 |
Break-even is 4.0%: each action costs $80 against $2,000 at stake, so it has to avoid that share of the loss to pay for itself.
Procurement notes
- Data residency and tenancy: see governance and control and cloud native.
- Security and controls: talk to us through the trust page. We state which controls exist and which attestations we do not claim.
What exists today
Public-sector worlds exist as catalog templates (the catalog carries budget to report, assess to treat and similar value chains). They are not a live simulated world.
See a decision reconstructed from its record
Bring a decision that was challenged. We will show what the receipt, the evidence and the provenance would have given you.