Typed output, always
A field is a decimal, a date or an enum, never a paragraph of text. Line items come back as arrays so the next system can consume them without parsing.
invoice.total: decimal · due_date: date · line_items: arrayAbout Mapsy
Every invoice, contract and form already contains the data your systems need. It is locked in a layout meant for human eyes. Mapsy exists to read that layout, return typed fields with a confidence score on each, and make sure a person sees only the values the model is unsure about.
01 / 04 — The problem
Character recognition turned scans into strings, and left the hard part to people: find the total, decide which date is the due date, check that the tax adds up, paste it into the ledger. Mapsy was built for the step after the text.
“4,218.60 USD” and “USD 4.218,60” are the same figure. A text dump leaves your team to normalise currency, separators and dates by hand.
When every field arrives looking equally certain, reviewers either check everything or trust everything. Both are expensive.
Someone fixes a tax figure at month end. Weeks later an auditor asks who, when and from what value. The answer is a shrug.
Hand-written glue between the inbox and the ledger breaks when a vendor changes its template, and nothing tells you until a bill is missing.
02 / 04 — Principles
Each one is a testable property of the product, not a slogan.
A field is a decimal, a date or an enum, never a paragraph of text. Line items come back as arrays so the next system can consume them without parsing.
invoice.total: decimal · due_date: date · line_items: arrayEvery field carries its own confidence score. You set the threshold per field or per document type, and only what falls under it asks for a human.
auto_approve_if: confidence.min ≥ 0.90Rules such as subtotal + tax = total catch a misread digit that a confidence score alone might let through. A failed rule names itself in the review queue.
rule: subtotal + tax == total → else route(review)Each run stores the input, the model version, the workflow version, every edit as a before and after, and the approver, in UTC.
run.stamp: model v3.2 · workflow v14 · hash 9c41e703 / 04 — How it is built and supported
Mapsy is four stages (read, route, review, record), and each one exposes its inputs, thresholds and output so you can see why a document went where it did.
| Document | Amount | Model | Status | |
|---|---|---|---|---|
| Brightline LogisticsRun #4,306 · Nov 10 | $4,218.60 | v3.2 | Posted | |
| Okafor Print WorksRun #4,307 · Nov 10 | $1,876.00 | v3.2 | Posted | |
| Sable Office SupplyRun #4,309 · Nov 10 | $642.18 | v3.2 | Edited | |
| Corbell IndustrialRun #4,311 · Nov 10 | $12,905.00 | v3.2 | In review | |
| Lumen Cloud HostingRun #4,318 · Nov 11 | $2,310.90 | v3.2 | Posted |
Pick a ceiling
Plans are sized by monthly volume, so the cost tracks your inbox rather than your headcount.
See the three ceilings04 / 04 — Where it is going
No dates promised here: each item is a direction we are working toward, listed so you can judge the fit.
Models
Better handling of skewed scans, multi-page line-item tables and handwritten approvals, with confidence scores that stay calibrated as accuracy improves.
OngoingWorkflows
More condition types in the builder, reusable rule sets across workflows, and approval steps that escalate when a reviewer has not responded.
In designConnectors
Additional accounting, CRM and storage connectors alongside the 40+ available on Automate, each stamped into the run history.
RollingHave a document type you need read? Write to contact@mapsyai.com with a description of the fields you need.
“What convinced us was the audit log. We could open one run and see the input, the model version, the edit and who confirmed it. That is a conversation with an auditor that takes a minute.”
Next
Pick the volume ceiling that fits your inbox, connect it, and publish your first workflow.
Choose your plan