01
Complaint complexity
A pilot can define how uncertain outputs reach a reviewer. Confidence-based routing to reviewers is not built.
Workflows / human review
Specialised inference for financial-service operations
Complaint teams need category suggestions that match their own rules. Lending operations need transaction labels and summaries that reviewers can check before use. Gyde can test one complaint or statement task on representative cases and count material errors. A pilot can define how uncertain outputs reach a reviewer.
Pilot candidates, not lending-decision automation. The institution owns consequential decisions and jurisdiction-specific rules.
GYDE / INFERENCEILLUSTRATIVE PILOT FLOW
DESIGN PATH / BUYER-SPECIFIC
01WorkloadTask + test cases
02ModelQualified candidate
03PolicyAllowed route
04ServingApproved capacity
05EvidenceOutcome + cost
Decision: proceed, redesign or stop
01Complaint operations
Define a reviewer step for category suggestions.
Pilot design: each regulator's categories can be mapped to a shared analytical taxonomy where useful, while the native category remains in the workflow. The US model and India preview have separate evidence populations. Confidence-based routing to reviewers is not built.
01Complaint→︎
02Native category suggestion→︎
03Reviewer step to define→︎
04Case system
02Statement analysis
Design a review of labels and calculated summaries.
Pilot design: a small exploratory line-label test exists. Statement-level income, obligation and borrower outcomes are not yet proven. A pilot must test material EMI and bounce events and the lender's income-error tolerance.
01Statement→︎
02Transaction labels→︎
03Calculated summary→︎
04Reviewer step to define
03Exception design
Escalate cases that need judgment.
02
Transaction ambiguity
Unfamiliar merchants, unclear income and material obligation lines need manual checks.
03
Regulatory variation
Taxonomy, language, location, retention and review rules are set for each institution and jurisdiction. A demo does not establish compliance.
04Pilot design
Measure the case that reaches the reviewer.
Use consented, de-identified representative data in an approved location, blind domain adjudication, a lender baseline and a frozen test split. Measure cost per correctly processed case. Do not send customer records to the marketing form or an unapproved labeller.
See Gyde's broader financial-services work?Buyer questions
Questions to settle before a pilot.
Answers reflect the current prototype and scoped engagement approach.
Can the model make a credit decision?
No. This pilot scope covers categorisation, extraction and triage suggestions with human review. The lender owns decision rules and consequential actions.
Do you have proven accuracy on our lender data?
No. That is a primary question for the pilot's representative, permissioned test set.
Can we keep our own complaint categories?
Yes. The pilot starts with your native taxonomy and maps it to a shared analytical layer only where useful. Native regulatory categories remain available in the workflow.
What is the evidence for statement analysis?
The current evidence is a small exploratory volunteer-statement line-label check with weak lending-critical coverage. Statement-level income, obligation and borrower outcomes still need a lender pilot.
Gyde Inference pilot
Start with a workload you can measure.
Describe the task, data boundary and decision you need to make. We will scope the test and operating conditions together.
Discuss a workload Do not submit sensitive records through the marketing form.