AI Data Intelligence:
Ratios, lookups, and document answers, calculated from your data with clear methodology and verifiable results.
Bind Ratio by Broker — 2025
| Broker | Submissions | Bound | Ratio |
|---|---|---|---|
| Coastal Risk Partners | 214 | 176 | 82.2% |
| Summit Underwriting Group | 158 | 121 | 76.6% |
| Harborview Insurance Services | 96 | 84 | 87.5% |
Methodology: Bind Ratio = Bound Policies ÷ Total Submissions per broker. Bound = active PolicyStatus (In Force, Renewed, Bound, Active, Written).
Why general AI falls short
Companies run on data spread across operational systems, spreadsheets, and document archives. An AI layered on top still has to get the math and the sourcing right, every time.
Policy exports, claims files, location schedules, and document archives live in separate systems. Getting a straight answer usually means exporting, joining, and reconciling by hand.
"What's our bind ratio by broker" or "loss ratio for this class of business" requires the right data, formula, and methodology. Plausible answers are not enough when results inform reports and decisions.
Spreadsheets, exports, and system records contain exact values that must be retrieved and calculated correctly. AskHandle delivers precise answers instantly and at scale, with no room for error.
Teams across insurance, finance, and other data-intensive functions need every figure tied to its source data, formula, and methodology. Unverified outputs cannot support critical reporting, compliance, or business decisions.
What's built for enterprise
Ask questions over your spreadsheets and CSVs in plain language. When you follow up — "explain that," "which of those is highest?" — the agent answers from the exact verified rows it already found instead of losing context or re-guessing.
A hybrid retrieval engine combines keyword ranking with multilingual semantic search, so exact codes, IDs, and terminology are matched precisely — and a question asked in English can still find the answer inside a document written in another language.
Convert PDFs into individual page images or process up to 100 files at once, including documents with hundreds of pages. Image conversion reduces exposure to embedded PDF elements, while dedicated infrastructure handles high-volume workloads beyond standard serverless limits.
A pluggable analytics engine encodes your domain's own vocabulary and math — whatever ratios and metrics run your business — computed correctly from joined datasets, with the methodology behind every number spelled out. Our deepest example so far: bind ratio, loss ratio, and open-claims analysis for insurance.
INSURANCE USE CASE · AI DATA INTELLIGENCE
Ask questions across policy and claims data, generate precise ratios and calculations, and trace every result to its source. AskHandle connects related records and applies your business methodology with speed, consistency, and scale.
“What's our 2025 bind ratio by broker?”
Bound policies ÷ total submissions, grouped by broker — with enterprise code, submission count, and bind ratio side by side.
“Loss ratio for Lee Wesley Restaurants?”
Scopes to the named insured, joins premium to incurred claims by policy number, and reports total premium, total incurred, and the resulting ratio.
“Which brokers have a bind ratio above 80% and an open claim?”
A cross-metric query — filters brokers by bind-ratio threshold AND at least one open/reopened claim on a bound policy.
“List open claims for Broker X”
Claim number, adjuster, loss date, status, description, incurred, and reserve — scoped to that broker's policies via a policy-number join.
This isn't an insurance product — it's an AI data intelligence engine with an insurance example. New industries are supported by configuring their data structures and business rules, without changing the core engine.
Loss ratio looks like simple division. In practice it's one of the easiest figures to get quietly, confidently wrong — because premium and claims almost never live at the same level of detail.
Loss Ratio by Class of Business — 2025
| Class of Business | Premium | Incurred | Ratio |
|---|---|---|---|
| Commercial Office Buildings | $1,860,400 | $412,900 | 22.2% |
| Casual Dining Restaurants | $3,140,700 | $920,200 | 29.3% |
| Light Manufacturing | $2,205,000 | $573,300 | 26.0% |
| COMBINED (3 classes) | $7,206,100 | $1,906,400 | 26.5% |
Methodology: Total Premium = sum of policy-level billed premium across all bound/in-force policies in the class. Total Incurred = sum of ALL claims — open and closed — joined on policy number. Loss Ratio = Incurred ÷ Premium.
One premium base, top and bottom
Numerator and denominator must be measured the same way. We deliberately use policy-level billed premium — not per-location premium — as the base for both, because mixing granularities silently inflates the ratio.
Each claim counted once, via a real join
Claims are matched to policies on policy number and attributed exactly once, even when a policy has multiple claims or appears across several location or class rollups.
Open AND closed claims — always
Total Incurred always sums every claim for the period, open and closed. A separate "open claims" view exists for adjuster follow-up, but it is never substituted into the loss ratio calculation.
The methodology ships with the number
Every result carries a plain-language explanation of exactly how it was computed, so when someone asks "how did you get this?" the answer is precise — not improvised.
Why the premium base matters
Early on, using per-location premium as the denominator against policy-level claims inflated one class's loss ratio to 229.7% instead of the correct 29.3% — nearly an 8x error, on a class that was actually profitable.
Follow-ups without losing the thread
Verified search results travel with the conversation. When your next message is a follow-up, the agent answers from the data it already found — instead of re-running a search that might come back empty.
No re-search required
"Explain that" or "put it in a table" is answered from the same verified rows.
Every figure stays traceable
Methodology travels with results, so the agent can explain how a number was calculated.
Never falls back to guessing
When a search finds nothing, the agent says so — it will not substitute general knowledge.
Follow-up, grounded end to end
01
Upload spreadsheets, exports from your operational systems, and PDF archives — policy and claims files for insurance, or the equivalent for your industry. Auto-detection matches your column headers to the right analytics profile.
02
Whatever ratio or metric runs your business — bind ratio by broker and loss ratio by class for insurance, or your own equivalents — plus straight lookups across your document library. No report-writing required.
03
Every figure carries its methodology. Every document search is hybrid-ranked for accuracy. Every large PDF batch runs on infrastructure built to handle it.
Insurance is one example, not the product. The underlying engine is a domain-agnostic "dataset profile" plugin architecture — each profile teaches the platform a vertical's vocabulary and math (which columns mean "broker" or "premium," how to compute a given ratio, which query phrasing triggers which calculation). Insurance (policy + claims data from RMS-style systems such as Nexsure and Filehandler exports) is our most fully built-out profile because it's a strong proof point for how deep this can go — but new verticals are added by registering a new profile, with no changes to the core search or join logic.
Every analytics result carries a plain-language methodology string describing exactly how it was computed — the premium base used, the join key, which rows counted as "bound," and what's included or excluded. That methodology travels with the answer into the AI's context, so it explains its math truthfully instead of improvising. The same discipline applies to document search: if nothing relevant is found, the agent says so rather than answering from general knowledge.
Yes. Verified results from a search are carried forward in the conversation, so when you ask a follow-up — "explain that," "put it in a table," "which of those is cheapest?" — the agent answers from the exact data it already retrieved instead of running a fresh, possibly empty search or falling back to guesswork.
Individual PDFs up to thousands of pages, and up to 100 files in a single bulk request. Large jobs run asynchronously on a dedicated, horizontally scaled worker fleet — you keep working while conversion completes in the background, with progress and per-page credit usage visible the whole time.
Yes — search combines keyword ranking (best for exact codes, IDs, and terminology) with multilingual semantic embeddings (best for matching meaning across languages), so a question asked in one language can retrieve the right passage from a document written in another.
Yes. Every search, join, and analytics computation is scoped to your organization's files. Team members can share access within an org, but data never crosses organizational boundaries.
Connect your data and documents, and get grounded answers — from custom ratio analytics to bulk PDF conversion — in one platform.
