Explain-Only Model
The AI Insight model receives a computed metric with its definition, cohorts and comparisons; it cannot calculate, score or write — by architecture, not by policy.
BR Insights: Every Benefits Number, Explained in Plain Language — With the Records Behind It
A coverage gap of 4.8% is a number. “1,214 eligible employees have no medical election; 71% are Plant 2 hourly who waived three months running; 22% are new hires still inside their window” is an insight. BR Insights writes the second sentence for every figure on the platform.
BR Insights is the AI Insight model of the platform: it reads a metric the analytics engine has already computed, the cohorts and records that produced it and the comparison that gives it meaning, and explains it — in a sentence a CFO, a benefits manager or an employee can act on, with the sources linked. It never calculates and never edits, and it explains SuccessFactors records and Workday, UKG, ADP, Dayforce or Paycom records the same way.
Explains computed metrics onlySources on every explanationWhy this match · why this riskGuided across the benefits cycleAny HRIS record
A Number Nobody Can Explain Is a Number Nobody Will Act On
Benefits teams are not short of figures; they are short of the sentence that goes with each one. BR Insights is that sentence, written by a model that is allowed to do exactly one thing: read a metric the engine has already computed, along with the records and comparisons behind it, and say what it means. It cannot calculate a figure, cannot change a record and cannot cite a source it did not read — which is why the explanation can sit next to the number on a board pack, an employee's plan recommendation or an auditor's screen.
ComputeThe engine produces the figure firstTraceRecords, cohorts and definitions gatheredExplainWhat, why, who, compared to whatGuideAcross the cycle: enroll → verify → reconcile → communicate → complyAny HRISSame explanation for every source
1job for the AI Insight model: explain a computed metric — it cannot calculate, score or write
5questions every explanation answers: what it is, what moved it, who is behind it, what it compares to, what to do
100%of explanations carry linked sources — the records, the definition and the comparison used
2audiences for the same engine: administrators on any workspace, employees on “why this match” and “why am I being asked this?”
One Number, Five Stages to an Explanation
Follow the 4.8% coverage gap from the engine that computed it, through the records Insights gathers, to the sentence a benefits manager reads — then the same for an employee's plan recommendation and for an entity on Dayforce.
The engine goes first — Insights explains only what exists.
Nothing is explained until it has been computed. The analytics orchestrator produces the coverage gap as of the evaluation date — eligible from the benefit program and rules, enrolled from the dated election records, pending kept apart — and stores the figure with its definition, its context and its cohorts. Only then is the AI Insight model called, with that computed result as its input. It does not see raw tables, it does not run queries, and it has no way to produce a number the engine did not.
What Insights receives
- The computed metric, its definition and the evaluation date
- The Analytics Context — filters and selection — the screen is showing
- The cohorts behind the figure, already suppressed below the minimum
- The comparison set the engine chose: prior period, prior year, peer band
Why it matters
An explanation can only be wrong about meaning, never about arithmetic — the arithmetic was done before it was asked.
The same separation lets Insights sit on ACA figures: it may explain exposure; it may never compute or change it.
Records, cohorts and definitions — gathered, not guessed.
Before writing a word, Insights assembles what it may cite: the definition of the metric from the orchestrator, the cohort breakdown with small groups already folded, the prior-period and prior-year values computed on the same definition, the records that changed since the last period — hires, terminations, waivers, plan moves — and any open Spotlight item or case that touches the figure. Everything in that bundle is a link. Anything not in it cannot appear in the explanation.
What the trace bundle holds
- The metric definition and the evaluation date it was computed for
- Cohorts and their sizes, suppressed below the minimum
- What changed since the last period — hires, exits, waivers, moves — as counts with links
- Related Spotlight items, cases and campaigns already open
Why it matters
Every clause in an explanation can be clicked back to the record or the definition that justifies it.
Role scoping applies to the bundle: an explanation for a manager without employee-level access is written from cohorts alone.
What, why, who, compared to what, what next — in one paragraph.
The explanation follows a fixed shape so it reads the same on every workspace: what the figure is and as of when; what moved it since the last period, with the counts; who is behind it, by cohort; what it compares to; what is not part of it; and what to do, as a proposal for the owning module. Every clause carries its source. The tone is plain — no jargon an employee could not follow — and the length is a paragraph, not a report. An administrator can ask a follow-up in the same place; the answer is written from the same bundle.
The shape of every explanation
- What it is, as of when, by which definition
- What moved it — counts of hires, exits, waivers, moves
- Who is behind it — cohorts, never individuals unless the role allows
- Compared to what — prior period, prior year, peer band when one exists
- What to do — a proposal drafted in the owning module
Why it matters
A manager, a CFO and an auditor read the same paragraph and each finds the clause they need.
Explanations are versioned with the model and prompt template that wrote them; re-running on the same bundle reproduces the same clauses.
Guided across the benefits cycle — the same explanation, five moments.
Insights is one model with many seats. At enrollment it is the “why this match” under an IntelliSuggest recommendation — the family, care pattern, prescriptions and HSA eligibility behind the score. At verification it is “why am I being asked this?” on a requirement. At reconciliation it explains a four-way variance — which lives, which period, which route. In communication it explains why a campaign is proposed and what it measured. In compliance it narrates an ACA month — offered, affordable, and the three exposure figures — without ever computing them. Same shape, same sources rule, different owner.
Where the same model explains
- Enroll — IntelliSuggest “why this match” for the employee, plan-mix shifts for the administrator
- Verify — “why am I being asked this?” on requirements; IntelliVerify outcomes for reviewers
- Reconcile — the variance behind an invoice, the leakage behind a period
- Communicate & Comply — why a campaign is proposed; what an ACA month means
Why it matters
One explanation engine means an employee's “why this plan?” and a CFO's “why this cost?” are held to the same standard.
Employee-facing explanations never expose internal confidence scores, table names or workflow states.
The explanation does not know which HRIS the record came from.
Insights reads the canonical model, so the source of an employee record is one more attribute, not a different pipeline. A gap on an SAP SuccessFactors entity and a gap on a Dayforce entity are explained by the same clauses from the same definitions; when the source matters — a hire date missing from a Workday feed, a termination not yet confirmed by HRIS Sync — the explanation says so and links the sync record. Multi-entity leaders get one paragraph for the group and one per entity, each with its own cohorts and comparisons.
How sources show up in an explanation
- Source named only when it explains something — a late feed, an unconfirmed field
- Group and per-entity explanations from the same definitions
- Cohort suppression and role scoping per entity
- Employee-facing “why” identical on every HRIS sign-in and language
Why it matters
A leader with a mixed HRIS estate reads one kind of explanation, not one per system.
Quebec French and Spanish explanations are written from the same bundle in the employee's language.
Core Capabilities of BR Insights
Built to explain benefits figures to people who must act on them — not a chat window bolted onto a chart.
Trace Bundle & Sources
Every clause is written from a bundle of linked records, definitions and comparisons assembled first; anything outside the bundle cannot be said.
Fixed Explanation Shape
What it is, what moved it, who is behind it, compared to what, what is not part of it, what to do — one paragraph, the same on every workspace.
Why This Match · Why This Risk · Why Asked
The same engine behind IntelliSuggest recommendations, Spotlight items and employee requirement prompts — without exposing internal scores or states.
Guided Across the Cycle
Enrollment, verification, reconciliation, communication and compliance explained by one model held to one standard, with the owning module named.
Governed & Reproducible
Role scoping and cohort suppression on the bundle; every explanation logged with model and prompt version; re-runs reproduce the clauses; inventoried under NIST AI RMF.
“AI Summaries” vs. BR Insights
Most analytics tools now offer a generated summary next to a chart. The difference is what the model was allowed to read, what it was allowed to do, and whether you can click on what it said.
| The moment | A generated summary | BR Insights |
|---|---|---|
| The number itself | The model may recompute or round it. | Computed by the engine first; the model receives the result and cannot change it. |
| “Why did it move?” | A plausible paragraph. | Counts with links — hires, exits, waivers, moves — from the change log. |
| “Who is behind it?” | Names, sometimes. | Cohorts, suppressed below the minimum and scoped to the reader's role. |
| “Compared to what?” | Whatever the model remembers. | The comparison set the engine chose — or an explicit “no peer band for this plan type”. |
| An employee's plan recommendation | A badge that says Recommended. | “Why this match” in plain language, no internal scores exposed. |
| An auditor asks how it was written | Nobody knows. | Model, prompt version and bundle logged; re-running reproduces the clauses. |
| A Dayforce entity | A different tool with a different summary. | The same paragraph shape from the same definitions, with the source named only when it matters. |
Insights inside Athena
Ask a Follow-Up and Get the Same Standard
When an administrator asks Athena about a figure, it is the Insight model that writes the answer — from the same trace bundle, with the same sources rule. A follow-up narrows the bundle; a “what should we do” hands a draft to the owning module. The employee gets the same engine under a plan recommendation or a requirement, in their own language.
- “Explain the employer cost per capita” — definition, what moved it, which plans, compared to last year
- “Only Plant 2 — and only the waivers” — the bundle narrowed; the clauses rewritten from it
- “Is the peer band real?” — where it came from, or an honest “no band for this plan type”
- “Why was PPO Family recommended to me?” — family, care pattern, prescriptions, HSA eligibility — no scores shown
- “What would you do?” — a campaign, a case or a review drafted, never sent
Explain the $812 employer cost per capita. Is that up, and why?
BR Insights · from the computed metric$812 is the employer-paid premium per enrolled employee per month as of 09/01, employee contributions excluded. It is up $31 (4.0%) on June: $19 of that is the 412 employees who moved from PPO to HDHP with a higher employer HSA seed, $9 is the July renewal on the dental plan, and $3 is mix — 98 new family-tier enrollments. Compared with the same date last year it is up 2.1%, inside the 2–6% band for your plan types. Cohorts under five are folded into “other”.
Sources: Benefits Participation · cost definition · Election Change Intelligence · plan moves · Renewal record · dental, July
Show by planShow last year
Only Plant 2, and only the HDHP movers.
BR InsightsNarrowed: Plant 2 has 176 of the 412 HDHP movers. Their employer cost rose $58 per head because the HSA seed is paid in full in the first month; from October it settles to $22 above their PPO cost. Nothing else moved in Plant 2. If you want, I can draft a note for Finance explaining the one-month spike — nothing is sent until you approve it.
Draft the noteShow the 176
Ask why any number is what it is…
Why Benefits, Finance and HR Leaders Choose BR Insights
Meaning, Not Arithmetic
The model explains a figure the engine computed; it cannot produce a number of its own.
Click Every Clause
Definitions, cohorts, changes and comparisons are links — remove a source and the clause goes with it.
One Shape, Every Metric
What, why, who, compared to what, what next — a paragraph that reads the same on every workspace.
Employees Get the Same Engine
“Why this match” and “why am I being asked this?” are written to the same standard, in the employee's language.
Reproducible by Design
Logged with model and prompt version; a re-run on the same bundle reproduces the clauses — inventoried under NIST AI RMF.
Any Record, Any HRIS
SuccessFactors or any other HRIS: the source is named only when it is the reason.
Where Insights Appear — Under Every Number on the Platform
Insights is a layer, not a page. It is the “Why?” on every Essential Insights tile, the paragraph beside each Spotlight item, the “why this match” under an IntelliSuggest recommendation, the “why am I being asked this?” on an employee requirement, the narrative on a four-way variance case and the explanation of an ACA month — one model, one sources rule, everywhere a number needs a sentence.
- Essential Insights — “Why?” on every KPI tile, chart and table row, written from the workspace's own context
- BR Spotlight — the reason paragraph on every ranked item — the rule or model, the records, the threshold
- IntelliSuggest and Employee Home — “why this match” under a recommendation, and “why am I being asked this?” on a requirement
- Carrier Billing and 4-Way Reconciliation — the narrative on a variance: which lives, which period, which route, what is unexplained
- ACA Intelligence — what an employee-month means and what the three exposure figures are — explained, never computed
Bring the Number Your Board Always Asks About. Read the Explanation.
Pick the figure that gets the most questions — employer cost per head, the waive rate, the ACA exposure, the plan mix. We'll compute it on your records, let Insights write the paragraph, click through every clause to its source, and then narrow it the way your CFO would.
Explains computed metrics onlySources on every clauseModel + prompt version loggedCohorts under 5 suppressedNIST AI RMF
BR Insights FAQs
The explanation layer of Benefits Reimagined analytics. Its AI Insight model reads a metric the analytics engine has already computed — with its definition, cohorts and comparisons — and writes a plain-language paragraph: what the figure is, what moved it, who is behind it, what it compares to, what is not part of it and what to do, with a linked source on every clause.
No. The analytics orchestrator computes every figure first; the Insight model receives the result and cannot run a query, recompute a number or change a record. That separation — the LLM interprets, Python calculates, the database supplies the data — is what allows an explanation to sit beside a board-pack figure or an ACA exposure.
Every clause is written from a trace bundle assembled before the model runs: the metric definition, the cohorts with small groups folded, the prior-period and prior-year values on the same definition, the changes since the last period and any related open item. Anything outside the bundle cannot be said, every clause links to its source, and the explanation is logged with the model and prompt version so it can be reproduced.
Spotlight decides what needs attention — it ranks signals by consequence and time left and gives each role a short list. Insights explains why a figure or an item is what it is. On a Spotlight item the reason paragraph is written by Insights; on a workspace the “Why?” under a tile is Insights too.
Yes, through the same engine held to the same standard: “why this match” under an IntelliSuggest plan recommendation — family, care pattern, prescriptions, HSA eligibility — and “why am I being asked this?” on a requirement. Employee-facing explanations are written in the employee's language and never expose internal confidence scores, table names or workflow states.
The trace bundle is assembled under the reader's role-based permissions and cohort suppression: a manager without employee-level access gets an explanation written from cohorts alone, and any group below the configured minimum is folded into the next level before the model sees it.
The explanation says so. If the engine has no peer band for a plan type or no prior-year value on the same definition, the comparison clause is omitted and the omission is stated — rather than a plausible figure being invented.
Yes. Insights reads the canonical data model, so a record from Workday, UKG, ADP, Dayforce or Paycom through HRIS Sync is explained with the same clauses and definitions as a SAP SuccessFactors record. The source is named only when it is the reason — a late feed or an unconfirmed field — and multi-entity leaders get a group paragraph and one per entity.
The Number, and the Sentence That Goes With It.
Computed by the engine, explained by a model that cannot change it, sourced on every clause — for SAP SuccessFactors records and every record from another HRIS.