Indico Data

A one-sentence summary

Indico Data is an AI-based platform for order processing and task scheduling, designed for insurance companies. It transforms the underwriting, claims handling, and policy-related materials that are scattered across emails, attachments, and document repositories into structured tasks that have been verified and can be directly fed into the core systems.

What is Indico Data?

At present, Indico Data focuses on supporting insurance operations, rather than providing general OCR or chatbot services to ordinary users. The platform handles large volumes of insurance documents that are in inconsistent formats and require auditing, through three processes: ingestion, completion, and orchestration.

The product integrates pre-configured agents, custom agents, visual workflows, human-machine review, business rules, and enterprise integration. Customers can start with existing insurance processes and then gradually expand them by using their own data structures, prompts, validation logic, and downstream systems.

Core working method

Main functions

Insurance document acquisition and unpacking

The platform can automatically retrieve materials from shared email accounts, document repositories, and object storage, as well as process nested emails and their attachments. Complex submission packages are split according to the document type, which reduces the need for manual opening, naming, and archiving.

Multi-format document understanding

It can process structured, semi-structured, and unstructured information, including email bodies, PDFs, tables, Excel files, ACORD forms, loss records, SOV files, images, compressed archives, and handwritten content. Different scan qualities, formats, and languages still need to be verified using actual samples.

Extraction Agents extraction agents

It extracts the relevant fields from insurance-related documents using proxy recognition techniques, and combines the dispersed information into a unified structure. The platform places emphasis on specialized training for insurance purposes as well as pre-built capabilities, rather than simply applying general large-scale models to such documents.

Enrichment Agents – completion agents

Complementary agents can categorize details, add context, answer questions from underwriters, generate summaries, or retrieve additional data through external APIs. Each completion must be governed by business rules, permissions, and data contracts.

Agent Gallery agent library

Agent Gallery offers hundreds of pre-configured agents and workflows, enabling insurance teams to bypass the need to train everything from scratch. Customers can deploy them directly, copy them to create new versions, or use these ready-made agents as a starting point for custom solutions.

Agent Studio Agency Studio

Agent Studio enables the configuration of agent behavior, input/output handling, data structures, prompts, and field-level validation logic. Versioned agents allow for isolated testing, followed by controlled deployment to a shared agent library.

Visualizing Agentic Workflows

Users can arrange multiple agents in a drag-and-drop canvas, and define conditional branches, sequencing, and data routing. Workflows are ideal for combining task submission and reception, field validation, error handling, manual review, and downstream data writing into a complete decision-making process.

Human-in-the-loop human verification

Fields with low confidence, cases of verification failure, and business exceptions can be sent for manual inspection; after modifications are made, the process can continue. The platform integrates manual verification into the production workflow, thereby preventing all AI-generated results from being used in high-risk insurance decisions without any oversight.

Data lineage, confidence, and explanation

The platform keeps track of the relationship between the fields and the original documents, as well as of manual and automated actions, versions, and pipeline logs. Auditors can determine which document a particular output value comes from, what processes were applied to it, and why an exception was triggered.

Enterprise system integration

Indico adopts an API-first architecture that supports REST, GraphQL, Webhook, messaging, and polling mechanisms. Customers can have the platform output data in JSON or table format independently, or they can send the processed results to insurance core systems, data warehouses, and automation platforms.

Headless and embedded deployment

The Headless mode enables companies to integrate Indico’s capabilities into their existing user interfaces and workflows, without requiring business users to access a separate front-end. Teams that need a full-featured workspace can also make use of the platform’s own proxies, workflow, and review interfaces.

Supported insurance documents

Current capacity scale

Four key business scenarios

Underwriting submission and order acceptance

Indico can separate the files from brokerage emails and submission packages, extract fields such as the policyholder, risk, loss, location, and SOV, and pass any missing or uncertain information to underwriters for further review. The structured data resulting from this process can be fed into the underwriting workstation or the core system.

Claims Processing and FNOL

The platform handles initial loss notifications, emails, attachments, images, and related materials; it extracts information from these elements and classifies, completes, and assigns them according to established rules. The claims handlers remain responsible for confirming the key facts and making the final decisions.

Mid-term adjustment of the policy

Changes in addresses, assets, personnel, or coverage amounts within emails, PDFs, and attachments can be identified and routed to the appropriate service teams. This helps to reduce duplicate entries and backlog, but it is necessary to ensure consistency with the rules of the existing policy management system.

Broker reconciliation

Inconsistent remittance lists and reconciliation packages can be structured, standardized, and matched with downstream records. The finance team can process cash applications more quickly, while still keeping track of any anomalies and carrying out manual verifications.

Implementation process

1. Select an insurance process with clear volume of business, available data, and a quantifiable baseline.
2. Collect real, anonymized samples representing different brokers, regions, languages, formats, and quality levels.
3. Define target fields, validation rules, exception types, and the data structure of downstream systems.
4. Select a nearby agent and workflow from the Agent Gallery, or create a custom version.
5. Configure the upstream email account, document repository, or object storage, and connect it to the target system.
6. In Agent Studio, adjust the data structure, prompts, field rules, and confidence thresholds.
7. Use the isolated version to test accuracy, missed detections, false positives, and abnormal routing.
8. Configure manual review and escalation procedures for fields with low confidence and high risk.
9. Start with a proof of concept or a pilot project, and then gradually introduce production traffic.
10. Continuously monitor the version, processing volume, rate of manual modifications, latency, and business outcomes.

From pilot testing to the acceptance methods for production

1. Calculate the field accuracy, document classification accuracy, and the pass rate for complete work packages separately.
2. Create test sets with common, edge, low-quality, and malicious inputs in various proportions.
3. Check whether each field can lead to the original document and page.
4. Determine whether to enter the security queue in cases of failed validation rules, proxy failures, and system unavailability.
5. Compare the processing time, rework rate, and backlog volume of the manual baseline.
6. Determine how manual modifications will affect the current records, subsequent versions, and the model.
7. Conduct tests for permissions, auditing, data export, backup and recovery, as well as deletion.
8. Verify the retry and idempotency of REST, GraphQL, Webhook, or message interfaces.
9. It goes live after being signed off by the business leader, the IT team, the security team, the legal team, and the model risk team.

Connection between the upstream and downstream sections

Deployment method

The platform operates on a single-tenant model, with each customer’s account being isolated from those of other customers. While this single-tenant approach facilitates isolation, it is still necessary to conduct a comprehensive assessment taking into account factors such as the network, keys, roles, logs, backup procedures, and operational permissions.

Prices and procurement models

Indico Data does not disclose a fixed monthly fee nor offers pre-defined packages that can be purchased based on the number of seats; instead, each customer receives a customized quote. The price is determined primarily by the number of submissions processed each year, as well as the number of users and administrators, along with the specific workflow, deployment, and service requirements.

Commercial terms in the open license template

The public templates for corporate licenses indicate that the subscription fee, activation fee, and consultation fees will be specified in the order or project description; these fees are generally non-refundable, except in cases where the platform fails to fulfill its obligations, in which case a refund may be given for the remaining period. The actual customer contract may stipulate different terms, and it is the document signed by both parties that shall prevail.

Service availability

The public template sets a monthly target of 99.95% for the Indico hosting environment, and offers different service discounts depending on whether the achievement level is below 99.95%, 99%, or 5%. Requests for these discounts must be submitted within 30 days after the end of the relevant month, and they represent the main remedy for instances of unavailability.

Launch time and support

Security and Governance

Data retention and privacy boundaries

• The standard order-handling approach allows for direct processing, without storing business data on the platform for an extended period.
• Options such as Underwriting Clearance and Triage allow data to be stored for a period selected by the customer.
• For managed environments, a selection of regions is available; self-managed options allow for further adjustments to the boundaries of data control.
• The platform claims that it does not use customer data to train models.
• Customers still need to confirm third-party models, logs, caching, backups, and access to technical support.
• The website’s privacy policy was last revised in 2023; when placing orders, it is necessary to obtain the current DPA as well as a list of sub-processors.
• Some contact addresses listed in the website’s privacy policy are in internal placeholder format and should not be used as the sole means of communication.
• Insurance documents often contain information regarding health, finances, identity, and social security, and should be handled at the highest level of sensitivity.

Data security incidents in 2026

Indico disclosed publicly that it became aware of a cybersecurity incident on May 7, 2026; investigations revealed that its online repository may have been accessed without authorization, and it contained information related to its corporate clients. The personal information at risk includes names, addresses, and social security numbers.

The company states that it has initiated an incident response plan, hired external cybersecurity experts, changed access credentials, enhanced monitoring and tightened access controls. It also provides credit monitoring and identity restoration services for some of the affected individuals. A public announcement says that the review of the data is still in progress, and customers should ask whether their organization has been affected and what the results of the corrective actions are.

How should companies assess this incident?

• Inquire about the affected systems, time frame, attack path, and the extent of the data leakage.
• Verify whether the company’s data, test samples, logs, backups, and support attachments are involved.
• Request third-party verification for credential rotation, enhanced monitoring, and access control improvements.
• Reassess repository permissions, service accounts, key lifecycle, and development environment isolation.
• Include in the contract the time limits for notifying of incidents, cooperation in collecting evidence, as well as responsibilities and credit monitoring.
• Decide whether additional audits are needed based on the insurance coverage, regulatory requirements, and customer commitments.
• Do not skip due diligence regarding an incident based solely on a SOC 2 or single-tenant statement.

Which organizations are suitable?

• Commercial insurance companies and reinsurance institutions that process a large number of broker-submitted packages each year.
• It is necessary to accelerate the operations of the FNOL, classification, and case allocation teams responsible for claims processing.
• The underwriting team that manually processes SOV, Loss Runs, and ACORD forms.
• Companies that need to transfer emails and attachments to Guidewire, Duck Creek, or Salesforce.
• Regulated industries with strict requirements for human-machine verification, data lineage, and auditing.
• Insurance groups that require single-tenant, regional hosting, or customer self-hosting.
• Teams that have in-house development and integration capabilities and require REST, GraphQL, or SDKs.
• The platform operations team hopes to reuse agents and workflows across multiple product lines.

Situations that are not very suitable

• Individuals, freelancers, and ordinary users who only need OCR occasionally.
• Small teams that wish to start using the service with a fixed monthly fee as soon as they complete the online registration.
• Projects that lack sufficient real-world samples, business rules, and definitions for downstream fields.
• Companies that only want to acquire general chat or content writing skills.
• Teams that demand complete open source and the ability to modify the core code of the platform on their own.
• Organizations that lack resources for security, legal affairs, model risk, and integration.
• Projects that require custom procurement, have long implementation cycles, and involve ongoing version management are not acceptable.
• Companies are prohibited from adding new suppliers until the special assessment of security incidents for 2026 is completed.

Main advantages

• Moving beyond general document processing, it focuses more on the semantics and entities related to insurance operations.
• It covers the entire process chain of ingestion, classification, extraction, completion, verification, and routing.
• Pre-built agents and workflows reduce the cold start cost for the first insurance use case.
• Agent Studio supports fine-grained control over structures, prompts, and field rules.
• Human-machine review, confidence levels, data lineage, version control, and auditing are suitable for high-risk operations.
• It supports REST, GraphQL, Webhook, messaging, polling, and various SDKs.
• Options include managed, regional, self-managed, Headless, and standalone platforms.
• Single-tenant, RBAC, and model governance meet the basic control requirements of enterprises.
• It can be connected to major insurance systems, data platforms, and RPA tools.

Usage restrictions and precautions

• There is no fixed public price; the budget must be determined through sales, scope, and contract evaluations.
• The manufacturer’s claims regarding accuracy, absence of illusions, and production success rates cannot replace customer testing.
• The first use case usually still requires 6 to 12 weeks; it is not a plug-and-play consumption tool.
• More than 70 languages and over 900 types of documents do not mean that every field will yield the same results.
• Custom proxies, integrations, and self-hosting increase the complexity of implementation and maintenance.
• The data retention rules for direct processing and persistence solutions are different.
• The public privacy policy is outdated, and some contact details need to be reconfirmed.
• Security incidents in 2026 that involve potentially sensitive personal information must be included in the procurement review.
• The open-source nature of the SDK does not mean that the enterprise platform or insurance models are also open source.
• High-risk underwriting and claims decisions still require clear human responsibilities and appeal processes.

APIs, SDKs, and GitHub

The official GitHub organization hosts a large number of public repositories, many of which contain projects that are no longer actively maintained. There is still ongoing maintenance for Python, C# and Java clients, but it is necessary to check the platform version, authentication methods and support details before using these SDKs.

Open-source status

• The Indico Data enterprise platform, Agent Studio, the agent database, and insurance models are not open-source products.
• The Python, C# and Java clients are made available under the MIT license.
• Some RPA connectors, examples, and research code are also available under different licenses.
• The historical indico.io client should not be confused with the current insurance platform.
• The fact that a warehouse is in an open state does not mean it is still supported by the current products.
• Self-hosted deployments are subject to corporate licensing and ordering restrictions, and cannot be freely redistributed.
• Platform access fees, implementation fees, and consultation fees are not waived just because the SDK is open source.

Basic information

Frequently Asked Questions

Is Indico Data a general-purpose OCR?

It’s not just OCR; it is designed for insurance order processing and operations, bringing together functions such as document acquisition, classification, field extraction, completion, verification, manual review, and downstream workflow management in a single platform.

Which files are supported?

It can process email bodies, PDFs, tables, Excel files, ACORD forms, Loss Runs, SOVs, images, compressed files, and handwritten text. The specific size limits, number of pages, and quality constraints must be determined based on the customer’s environment.

Is it necessary to write code?

Business teams can use pre-built agents, code-free configuration, and drag-and-drop workflows, so programming is not always necessary. Complex integrations, external data integration, and headless applications usually still require developers.

How long before it can go live?

The typical production timeline for the first use case provided by the platform is 6 to 12 weeks. Factors such as sample preparation, system integration, security reviews, field complexity, and acceptance criteria all influence the progress.

What is the price?

There is no fixed public price. Quotations are determined primarily based on the number of submissions per year, as well as by the user and administrator, along with factors such as agents, workflows, deployment, integration, and the scope of services.

Is a trial version available?

Companies can discuss concept validation or pilot projects, but there are no immediate free accounts available for individuals. Whether a pilot project is charged, its duration, the amount of data involved, and the expected outcomes must be confirmed in writing by both parties.

Will the insurance documents be kept?

The standard order-processing scheme can be retained in a direct manner without any long-term storage, while certain underwriting-related cleanup and routing schemes are kept for the period selected by the customer. Logs, backups, sub-processors, and self-hosted rules require separate confirmation.

Will the model be trained using customer data?

The current security architecture specifies that customer data will not be used to train models. Companies should still include this restriction, the requirements regarding third-party models, and the procedures for verification before deletion in the contract.

Is self-hosting supported?

Customers are allowed to deploy their own instances in accordance with the environment specifications provided by Indico. Self-hosting entails additional responsibilities regarding infrastructure, security, upgrades, and version support, and it does not mean that the platform is open source.

Are APIs provided?

REST and GraphQL endpoints are provided, in addition to support for Webhook, messaging, and polling integrations. The specific interfaces, permissions, rates, and available capabilities depend on the customer’s environment and the contract terms.

Is the platform open source?

The platform itself is not open source, but the official Python, C# and Java SDKs are made available under the MIT license. The older clients, research repositories and RPA components do not reflect the licensing terms of the core platform.

What are security incidents?

In May 2026, the company identified a cybersecurity incident: the online database might have been accessed without authorization, and the information at risk included names, addresses, and social security numbers. Purchasers should request information on the scope of the impact, the measures taken to address it, and any third-party verification.

Summary

Indico Data is suitable for large and medium-sized insurance companies that need to convert large volumes of insurance-related emails, submission packages, SOVs, loss records, and claims documentation into data that can be used in subsequent processes. Its key advantages include specialized insurance handling tools, end-to-end workflow management, human-machine verification, tracking of data origins, and various options for enterprise deployment.

When making procurement decisions, it is necessary to consider factors such as the accuracy of data fields, abnormal routing patterns, integration costs, customized pricing, the boundaries between single-tenant and self-hosted solutions, old privacy policies, and security incidents that may occur by 2026. While the SDKs and interfaces offer comprehensive functionality, the platform remains a proprietary commercial system subject to corporate contracts.