Toli

Toli turns your data and processes into calibrated decisions

Toli answers typed questions with a calibrated confidence, and keeps every reading countable. Your agents act when they can and hand over when they must: and you can analyse both.

Documents arrive by the hundred, filed by whoever sent them. Toli reads each one, proposes its type, and says how sure it is. What is clear files itself; what is not waits for a person, with the reason attached.

One reading, in full
QuestiondeclaredreadProbability per optioncertaintyRoute
the typePleadingsJudgment58%ESCALATE
This document sets a deadline96%92%ACT
The order, and nothing else
declared / read
declared
read
SummonsPleadingsJudgmentExhibit
Summonsn = 229
16873%
4118%
62.6%
146.1%
Pleadingsn = 330
3711%
21465%
5818%
216.4%
Judgmentn = 217
41.8%
2913%
17681%
83.7%
Exhibitn = 197
199.6%
2613%
115.6%
14172%
What was declared, against what Toli read. Cells are tinted by share of their row, so rows of different sizes compare; the outlined diagonal is agreement.

The same thing to plug in, whatever the field

What changes from one domain to the next is the wording of the questions and where you cut the thresholds. The shape of the answer, and the analysis you run on it, never change: which is why this list has no end.

Technical supportStatus pagesDocument classificationHospital pathwaySickle-cell and rare diseaseContent moderationFraud reviewCompliance and KYCApplication screeningLead qualificationInsuranceAlert triageCustomer feedbackSupplier intakeTailoring and garment makingProperty and housingAgricultureMicrofinanceLogistics and transportEducation and exams+ any other domain

It proposes, a person decides

On this path the ACT route does not exist, not switched off: absent. An agent built here cannot act alone, whatever its configuration says.

What is availableToli readsProbabilitiesCONFIRMESCALATEReport

There is no third route here: ACT is absent from this path.