Benki

A one-sentence summary

Benki is an explainable AI platform designed for teams involved in M&A, due diligence, risk analysis, and financial compliance; it focuses on generating auditable transaction documents, updating financial models, and tracking the contributions of both human staff and AI.

Tool Introduction

Benki combines transaction data, corporate knowledge, role-based permissions, and financial AI agents to create virtual analysis teams. Its purpose is not to serve as a tool for individual investors to select stocks, but rather to function as a trading workstation for investment banks, M&A consultants, private equity firms, corporate development teams, and compliance departments.

As of August 20, 2026, the official website still offers joining the waiting list as the main way to access the service; there is no official product control panel, fixed prices, or comprehensive help documentation available. The features listed below are based on the product offerings shown on the website at present, and they do not mean that all users can use them immediately.

Current open status

What problems does Benki solve?

• Centralize the transaction-related materials that are scattered across data rooms, tables, memos, and internal knowledge bases for unified processing.
• Assist in creating drafts of CIM, due diligence memorandums, risk lists, and other transaction documents.
• Synchronize new financial data into the model and files to reduce redundant manual updates.
• Document the processes of AI generation, manual editing, review, and publication to create an audit trail.
• Restrict access to knowledge sources and documents by transaction, team, and role.
• Select larger models that better suit the institution’s risk preferences through financial task evaluations.

Main functions

Dynamic transaction files

The Benki website shows the command-based workflow for generating transaction documents, updating profit and loss data, and issuing CIMs. It is possible to view the changes before they are published, and to export them in formats such as PDF, DOCX, or HTML.

Financial model update

The functions of this platform include extracting data from tables or transaction records and updating existing financial models. AI can help in locating the relevant data and filling in drafts, but valuation assumptions, accounting principles, and the accuracy of formulas still need to be reviewed by professionals.

Tracking human and AI contributions

Benki emphasizes the need to keep track of both human and AI contributions to documents, in order to support internal approvals and compliance checks. This traceability helps answer questions such as where the content comes from, who has made modifications, and when it was published.

Financial AI Agents

The team can create AI agents in the model customization environment as described on the official website, and assign to them knowledge sources, access rights, and roles. Multiple virtual analysts can handle tasks such as data organization, financial analysis, risk identification, and document generation.

Knowledge sources and permissions

• Connect data rooms, tables, documents, and internal knowledge by transaction items.
• Assign roles to different analysts, managers, compliance officers, and AI agents.
• Limit the range of transactions, directories, and knowledge that agents are allowed to read.
• Retain the locations of citations or evidence in the output for easy reference.
• Add manual verification before publishing to prevent drafts from becoming official documents directly.

Publication of transaction documents

The examples on the official website allow users to view, modify, or publish the generated content, and to download the files in PDF, DOCX, or HTML format. The public pages do not provide detailed information regarding features such as the template system, collaborative annotation, version restoration, and electronic signing; therefore, it is not possible to infer these features on one’s own.

FinBench Arena

FinBench Arena is a financial large-model evaluation project made publicly available by Benki on GitHub; it is used to compare the performance of various models in real-world financial services tasks. This repository was developed based on the concepts of FastChat and Chatbot Arena, and it is licensed under the Apache-2.0 license.

The workflow from data to transaction documents

1. The transaction lead defines the project boundaries, determines the type of transaction, the scope of analysis, and the deliverables.
2. Import data room files, financial tables, and internal templates that have undergone permission verification.
3. Configure the minimum access permissions for analysts, reviewers, compliance officers, and virtual agents.
4. Select a model and agent configuration that is suitable for the task and the organization’s risk tolerance.
5. Generate initial drafts of CIM, due diligence questions, risk summaries, or financial models.
6. Check each item regarding references, financial metrics, assumptions, formulas, omissions, and conflicts of interest.
7. Official documents should be published or exported only after the changes have been approved by authorized personnel.
8. Retain the input version, model information, prompts, manual edits, and publication records.

Tutorial for pending application

1. Verify that the team is indeed engaged in M&A, due diligence, transaction advisory, financial risk, or compliance work.
2. Provide the organization’s email address, team size, typical transaction process, and a description of the areas that need improvement.
3. Submit your email address at the waiting entry on the official website, and wait for the product team to get in touch.
4. When communicating, ask about the available regions, status of deployment, methods of deployment, model providers, and data processing rules.
5. It is required to use masked materials for the demonstration; real confidential transaction documents should not be uploaded during the first communication.
6. A pilot on a limited scale will be carried out only after obtaining safety, legal, and procurement approvals.

Pilot implementation tutorial

1. Select a completed or fully anonymized historical transaction as a test sample.
2. Define metrics such as file generation time, reference accuracy, data extraction accuracy, and the amount of manual modification.
3. Create read-only data sources and separate testing accounts, without connecting to the production data environment.
4. Use the same task to compare different models, prompt phrases, and permission settings.
5. Have analysts, legal experts, finance professionals, and compliance officers review the results independently.
6. Check the processes for exporting, logging, deleting, revoking permissions, and switching models.
7. Only after meeting the predetermined standards is it possible to gradually move on to actual transactions.

Which users are it suitable for

• Investment banking team: Assists in drafting CIMs, transaction memorandums, and client materials.
• M&A advisor: Organizes the data room, list of issues, and findings from due diligence.
• Private equity firms: Handle investment screening, business due diligence, and preparation of materials for the investment committee.
• Business Development Department: Compares target companies, synergy assumptions, and transaction risks.
• Financial analyst: Extracts data from documents and maintains draft models.
• Risk and Compliance Team: Reviews the basis for checks, permissions, changes, and records of manual audits.
• Head of Financial AI: Utilize evaluation frameworks to establish strategies for model selection and deployment.

Typical use cases

• Generate a draft of CIM or an internal transaction memo from the materials in the data room.
• Update the latest figures for profits and losses, assets and liabilities, and cash flows in the analysis model.
• Identify due diligence issues from contracts, operational documents, and financial records.
• Create a tracking list for risk clauses, abnormal data, and information gaps.
• Provides research and writing support for junior analysts, including the locations of relevant evidence.
• It records AI-generated content and manual modifications, and supports internal quality checks.
• For various financial tasks, use larger models and establish a list of organization-level models.

Product advantages

• It focuses on mergers and acquisitions and financial transactions, rather than being a simple version of a general-purpose chatbot.
• Place file generation, financial models, knowledge sources, and permissions within the same workflow.
• It emphasizes the tracking of human and AI contributions, which aligns with the financial teams’ focus on auditability.
• It allows virtual analysts to be organized by role, as well as to define access levels.
• It allows users to view the changes before their official release and to include manual verification.
• FinBench Arena provides a publicly available reference implementation for the evaluation of financial models.
• The output formats include PDF, DOCX, and HTML, facilitating integration with existing delivery processes.

Usage restrictions and precautions

• The official website still shows mainly a list of candidates; it cannot be assumed that the product is already widely in use.
• There are no publicly stated fixed prices, free trials, service levels, or standard contract durations.
• The public page lacks detailed help documents, API documentation, and deployment architecture.
• AI may incorrectly extract numbers, misinterpret accounting principles, or generate non-existent references.
• The formulas, valuation assumptions, and scenario analysis in the financial model must be reviewed by professionals.
• The transaction documents contain highly confidential information, and access to them requires legal, security, and customer authorization.
• Audit trails can only enhance traceability; they cannot automatically meet all regulatory requirements.
• The open-source nature of FinBench Arena does not mean that the Benki commercial platform can be deployed on its own.
• The official website indicates the copyright information for 2024; it is necessary to verify the current maintenance status and product roadmap before making a purchase.
• Different judicial jurisdictions have varying requirements regarding cross-border transaction data, financial advice, and AI records.

Financial exports must be manually reviewed.

Price and availability

Benki’s official website does not disclose information regarding monthly fees, prices per seat, corporate packages, or free trials. The current way to obtain access is by joining the waiting list; future costs may be determined by factors such as the number of seats, volume of transactions, storage needs, use of models, data integration, deployment methods, and support services.

Questions to ask before making a purchase

• Is the product already in official commercial use, or are some of its functions still in the demonstration or testing phase?
• In which regions are the data stored? Is it possible for customers to specify a particular region or to opt for private deployment?
• Can it be specified in the contract that customer documents shall not be used for model training?
• Which data rooms, tables, documents, and identity management systems are supported?
• Which audit fields are retained for citations, versions, hints, models, and manual modifications?
• How to implement transaction-level isolation, role-based permissions, key rotation, and emergency revocation of privileges?
• How should the original files, vectors, logs, model caches, and backups be deleted after the contract comes to an end?
• Does the price include costs for model invocation, storage, export, support, and implementation?

Supported platforms and outputs

Permission and audit design

• Create separate spaces for each transaction; default sharing of knowledge across transactions is prohibited.
• Set minimum permissions for AI agents that are the same as or stricter than those of employees.
• Record the input file version, model version, prompt, parameters, and generation time.
• Log the manual modifications, comments, approvals, and publishing actions in logs that cannot be altered arbitrarily.
• High-risk outputs must be approved by a designated responsible person and cannot be released automatically by AI.
• Regularly review the access rights of employees who leave the company, those who are reassigned, projects that have been completed, and external consultants.

Data security and privacy

M&A data rooms typically contain confidential information such as financial details, employee data, customer information, contracts, intellectual property rights, and transaction terms. The public pages of Benki do not provide sufficient details regarding the data handling procedures, so companies cannot rely on claims of ‘auditability’ to assess its security.

• During the pilot phase, historical or synthetic materials are used, and personal and commercially sensitive fields are removed.
• Details on encryption, authentication, access logging, backup, and vulnerability management are required.
• Confirm the underlying model provider, sub-processors, and cross-border transmission arrangements.
• Service providers are prohibited from using customers’ content to train general-purpose models unless they have written permission.
• Set requirements for data retention periods, legal retention, deletion proofs, and event notifications.
• Add categories, access controls, and measures to prevent accidental sending for file downloads.

Detailed Explanation of FinBench Arena

The FinBench Arena repository applies the dialogue arena approach to financial services scenarios, enabling teams to compare models through repeatable tasks. It is more suitable for researchers and evaluators, and is not a free alternative to the Benki business platform.

How to conduct evaluations of financial models

1. Tasks involving understanding, calculation, and writing are derived from real-world documents that have had their sensitive information removed.
2. Define criteria for correctness, referencing, stability, cost, and latency for each task.
3. Compare candidate models using the same input and tool conditions.
4. It is rated independently by at least two reviewers with expertise in the field.
5. Record the model version, date, prompts, parameters, and failure type.
6. Thresholds are set based on business risk, rather than merely considering the average ranking.
7. After going live, regular regression testing with new tasks should be carried out, and a manual rollback plan should be in place.

GitHub and the open-source status

The official Benki GitHub organization makes available FinBench Arena, the code for its website, as well as various branches and experimental repositories. FinBench Arena is licensed under the Apache-2.0 license, while some of the other repositories use the MIT license or licenses from upstream projects.

The complete source code for Benki’s core M&A workspace, transaction document system, permission management platform, and business services is not made public. The open-source status indicated in the catalog should read as ‘Core products are not open source; projects such as FinBench Arena are open source’.

Basic information

Recommendation score

The comprehensive recommendation score is 3.8 out of 5 points. Benki offers a clear approach to evaluating merger and acquisition documents, permissions, audit trails, and financial models; it is suitable for teams that wish to establish controlled financial AI processes.

However, the information available about these products at present is mainly conceptual in nature; prices, documentation, deployment details, security aspects, and the timeline for their official availability are all unclear. At this stage, it is more appropriate to add them to the list of candidates and conduct further due diligence, rather than integrating them into key business processes before they have been verified.

Frequently Asked Questions

What is Benki tool?

It is an AI workspace designed for M&A, due diligence, and financial compliance teams, with a focus on transaction documents, financial models, and auditable processes.

Can Benki be registered directly now?

The official website currently provides only a link to the waiting list; it does not offer a complete self-registration option or a formal control panel.

How much is Benki?

The official website does not disclose the prices; it is necessary to wait for the team to get in touch and determine the costs based on the company’s requirements.

What types of files can be generated?

The official website displays transaction documents such as CIM, and supports export in PDF, DOCX, and HTML formats.

Will the financial model be updated automatically?

The areas related to the product include updating financial models, but the formulas, calculation methods, units, and assumptions must be reviewed by professionals.

What is an auditable transaction document?

It emphasizes documenting the processes of AI generation, manual editing, approval, and publication, so that the team can trace the basis for the content and identify the responsible parties.

Can Benki ensure compliance?

No, the audit and permission functions can only serve as a support for governance; actual compliance still depends on policies, configurations, manual reviews, and applicable laws.

Are APIs provided?

The official website does not provide verifiable API or SDK documentation.

Is Benki open source?

The core business platform is not open-source, but the official GitHub releases projects such as FinBench Arena.

What is FinBench Arena?

It is an open-source evaluation platform used to compare the performance of large-scale financial models, and it is licensed under the Apache-2.0 license.

Can real transaction documents be uploaded?

Real and confidential transaction data should not be uploaded until the security, legal, customer authorization, and data processing reviews have been completed.

Is it suitable for individual investors?

It is not suitable; it is designed for professional trading and institutional workflows, rather than serving as a stock recommendation service or personal investment advisor.