Reconciliation at renewal
Client, benefits platform and insurer, crossed per insured with reversed surnames and abbreviations resolved.
orbit agents
/
Reconciliation · Health claims
/
In PoC
Two health insurance processes. One agent for each.
In a group health account, reconciling who is insured and checking whether an insurer's claim ruling holds up are two slow, manual jobs. orbit agents brings one agent for each: the reconciliation agent crosses the three insured lists and explains every difference, and the claims agent checks the ruling against the policy with cited evidence. It is currently a proof of concept with an insurance intermediary.
orbit agents
agents · insurance
01 · Context
Two jobs done by eye today.
At every renewal of a group policy there are three lists that should match: the client's, the one from its benefits platform and the insurer's. They don't. Surnames come in reversed, names abbreviated, and the same plan has a different name on each side, so someone crosses thousands of rows in Excel. Claims look similar: a ruling arrives asking for information or denying the case, and someone has to check whether that information was already sent or whether the denial holds up against the policy wording and its endorsements.
02 · What it is
orbit agents in one sentence
Two independent AI agents switched on per client: one reconciles the insured population across three sources with an AI-proposed matching key, fuzzy matching, plan mapping and an explainable score; the other classifies claim rulings and checks them against the policy with cited evidence, subject to human validation.
03 · How it works
Four steps. No magic.
Two short paths, each with a person validating at the end.
01
Sources are loaded
To reconcile, the three lists are uploaded as Excel or CSV. For a claim, the ruling is uploaded with the policy schedule, conditions, coverages, endorsements and the file already sent.
02
AI proposes, a person confirms
The reconciliation agent profiles the columns and proposes the matching key; the analyst accepts or adjusts it. The claims agent classifies the ruling: information request or denial, and its cause.
03
Matching and checking
Names are normalized, plans mapped, business rules applied, and each insured gets a score with its breakdown. For claims, the ruling is checked against the policy and every argument cites its passage.
04
Human review and output
Whatever falls below the threshold goes to review; the analyst accepts or rejects and the reconciled record is exported. For claims, the owner accepts, rejects or comments on the recommendation.
How it works · orbit agents live
What it does not promise
Neither agent decides on its own. The claims agent delivers a recommendation with a confidence level and never closes the case; if the documents aren't enough, it says so as insufficient information instead of forcing a conclusion, and it quotes nothing that isn't in the documents. The reconciliation agent only resolves high-score matches; everything else goes to review. Today both are built on group health insurance, and the reconciliation business rules are configured with our team, not from a screen.
04 · What it does
Where it goes to work.
What each agent does today in the proof of concept environment.
Classified inconsistencies
Missing from and extra at the insurer, possible duplicates, plan differences and rule exclusions, each with its reason.
Exportable reconciled record
An Excel file with one row per insured, its score, classification, matching method and review decision.
Information request rulings
The agent checks whether what the insurer is asking for is already in the file that was sent.
Denial rulings
Exclusion, pre-existing condition, waiting period or unauthorized treatment, checked against the wording and current endorsements.
Agents per client
Each client runs isolated, with its own branding, and only with the agents it has switched on.
05 · Benefits
What changes when you have it.
Five differences from crossing lists and reading rulings by hand.
Request demo1
The key isn't guessed
The agent profiles each source's columns and proposes how to match them; the analyst confirms before running.
2
The same plan, even under another name
Semantic mapping recognizes that two plan names are the same product before flagging a difference.
3
A score that explains itself
Every match shows how much each attribute contributed, and the threshold decides whether it passes or goes to review.
4
Cited evidence on claims
The insurer's position, the insured's argument and the analysis arrive with the passage from each document.
5
The decision stays human
Below the confidence threshold review is mandatory, and every recommendation is validated by an owner.
06 · Before and after
What changes, side by side.
An honest comparison between the manual work and the work with both agents.
Before / Without orbit agents
Today's day-to-day
- ✕ Three lists crossed by hand in Excel
- ✕ A reversed surname becomes a false missing record
- ✕ The same plan under another name is flagged as a difference
- ✕ The ruling is read against the policy page by page
- ✕ The conclusion doesn't say where it came from
After / With orbit agents
Day-to-day with the agents
- ✓ A matching key proposed and confirmed
- ✓ Fuzzy matching resolves names and abbreviations
- ✓ Plans are mapped before comparing
- ✓ The check against wording and endorsements arrives ready
- ✓ Every argument comes with its cited passage
07 · Industries
Where it applies today.
Both agents are built on group major medical insurance, which is where the proof of concept runs.
01 / Brokers
Group health accounts
The case the proof of concept was built on: reconciliation at renewal and advocacy for the insured on claims.
02 / Insurers
Health enrollments and claim rulings
The same reconciliation from the insurer's side, and a second review of rulings before they go out.
08 · Contact
Let's see it with your operation.
In 30 minutes we'll tell you whether orbit agents fits your process, what it integrates with and what it doesn't, and what a pilot looks like. If it doesn't fit, we'll tell you that too.
Run the discovery with rocky