How Thompson performed 260 credit analyses with AI in 20 days

With the support of the Skyone platform, the consulting firm developed Clara, an AI agent that performed approximately 260 credit analyses in 20 days, reduced costs associated with credit bureau inquiries, and expanded the team's operational capacity.

Analyzing financial documents, calculating indicators, and assessing credit risks are activities that require time, attention, and consistency. When the volume of requests increases, these tasks can compromise the pace of operations and even delay the release of sales orders.

This was one of the challenges identified by Thompson, a technology consulting firm that develops solutions to optimize business processes. Based on its experience with automation and with the support of Skyone Studio, the company created Clara, an artificial intelligence agent focused on credit analysis.

In approximately 20 days of operation, Clara performed around 260 analyses. Of that total, between 45% and 46% were conducted without incurring new costs to obtain data from a credit bureau, thanks to the reuse of previous queries.

The project demonstrates how the combination of artificial intelligence, systems integration , and knowledge of business processes can expand operational capacity without requiring a proportional increase in resources dedicated to the activity.

About Thompson

Thompson Management Horizons operates in consulting, technology, process management, and the development of digital solutions. Since 2012, Helder Satin, the company's technology director, has been overseeing the evolution of this area, which began by supporting consulting projects and has expanded to developing its own products to address recurring client problems.

Throughout its history, the company has participated in over 800 projects in sectors such as industry, healthcare, services, and oil and gas. Its areas of expertise include the WTMH process management platform, systems integration, and process automation using RPA and artificial intelligence.

The experience gained from working with different operations also helps Thompson identify opportunities to apply technology to activities that require a lot of manual effort, have bottlenecks, or could be performed more consistently.

The challenge: to make credit analysis more agile and consistent

Prior to Clara, Thompson had already worked on credit analysis projects based on RPA, a technology that automates repetitive tasks. While this approach helped execute process steps, there were still limitations in analyzing documents and information that required interpretation.

Credit assessment involves various sources of information, such as articles of incorporation, balance sheets, income statements, financial reports, and consultations with credit bureaus. It may also require the calculation of indicators and the application of specific rules to identify risks.

In operations with a high volume of requests, manual work demands constant dedication from the team. According to Helder Satin, in certain scenarios, one or two people needed to dedicate themselves full-time to these analyses.

The time spent on this activity also impacted other stages of the business. Since credit analysis can determine the approval of a sales order, delays in this process can affect the progress of the commercial operation.

There was also the challenge of maintaining the quality of the evaluations. Under pressure from volume and deadlines, more detailed analyses could end up being carried out with less rigor, increasing the risk of inconsistencies.

The opportunity, therefore, was to expand automation beyond task execution. Thompson was looking for a solution capable of interpreting documents, relating information, and applying analysis criteria consistently, while maintaining the possibility of human validation.

Read also: AI Agents: Transforming data into intelligent decisions with autonomy and scale

The solution: Clara, an AI agent for credit analysis

Developed with the support of Skyone Studio and Thompson's expertise in process automation, Clara was created to take on a significant portion of the document analysis work.

Instead of requiring an analyst to open different files, consult reports, and calculate indicators individually, the agent gathers the necessary information and produces a structured analysis, taking into account the rules defined by the company.

Among the documents that Clara is able to analyze are:

• Social contracts;
• Balance sheets and income statements (DRE);
• Financial transaction reports;
• Credit bureau reports;
• Records of invoices and payment slips, to verify payment behavior and the occurrence of delays.

Based on this data, the agent organizes the indicators and presents a summary to support the analyst's evaluation. This allows the team to focus their time on validating information and making decisions, instead of dedicating most of their effort to manually reading and consolidating documents.

Clara's autonomy depends on the rules established by the company and the maturity level of the process. The solution can support human decision-making and, under predefined conditions, evolve to execute certain decisions autonomously.

This approach allows automation to be scaled up according to the criteria and controls established by the business itself.

How Skyone Studio makes the operation viable

Clara was developed as a multi-agent system that utilizes different Learning Language Models (LLMs) to leverage distinct analytical capabilities and balance quality and cost.

Skyone Studio provides the framework to connect the agent to the systems and channels used by the client, organize workflows, and store information about the analyses performed.

Currently, the solution can receive requests via email, through Thompson's process management platform, and through the Sankhya ERP system. This allows requests to be routed from different points in the operation, without depending on a single channel.

The platform also allows users to record information about the processes performed, creating a database for tracking indicators such as:

• Number of analyses performed per period;
• Credit volume analyzed;
• Number of reviews per supplier;
• Trends of approval and disapproval.

With the Creator functionality, available in Skyone Studio, Thompson is also working on building an interface for clients to monitor the agent's performance and the results produced.

This visibility is part of the implementation strategy. For Thompson, putting an agent into operation is just one step in the project. It's also necessary to measure its output, track its performance, and understand the return it provides to the operation.

The results: approximately 260 analyses in 20 days

Clara's initial results demonstrate the potential of artificial intelligence to enhance credit analysis capabilities and reduce the effort dedicated to operational activities.

In approximately 20 days of operation, the agent performed around 260 credit analyses, applying the rules defined by the company and organizing the necessary information to support decision-making.

The gains are concentrated on three fronts:

1. Greater operational capacity. Automation allows for processing a larger volume of requests without requiring the team to dedicate proportionally more time to reading documents and calculating indicators.
2. Greater consistency in analyses. By applying the same rules in a standardized way, the agent helps reduce the risk of missed steps or inconsistent calculations, especially during periods of higher demand.
3. Cost reduction for credit bureau reports. One of the implemented features allows for the storage and reuse of reports from credit bureaus for a period defined by company policy. This means that certain analyses can utilize previously consulted data, avoiding charges for issuing a new report.

The results also helped demonstrate the value of the solution to the client, opening up opportunities for Thompson to explore other automation needs.

A partnership that began with integrations

The relationship between Thompson and Skyone began before the development of Clara, stemming from a recurring challenge in the consultancy's projects: integrating enterprise systems.

In many projects, the team relied on resources provided by the client or other suppliers. When documentation, APIs, or available professionals to develop the necessary connections were lacking, deadlines and costs could increase.

Thompson also found in Skyone Studio an alternative for developing integrations with greater agility and flexibility. The low-code approach also helped reduce development effort and brought security and traceability features to the projects.

The experience began with Thompson as a user of the tool and evolved into an integration partnership, with participation in Skyone's client projects.

Since then, the collaboration has expanded to encompass new possibilities for the platform, including data resources and artificial intelligence. The creation of Clara is an example of this evolution, combining technological infrastructure with knowledge of business processes.

Thompson also participated in other projects with Skyone, including the development of an agent to monitor daily publications in the Official Gazette of the Union and the states of São Paulo and Paraná. The solution identifies mentions of strategic topics for a retail company and sends reports with relevant excerpts and links for further information.

In another project, in the pharmaceutical sector, the consulting firm supported the migration of an integration tool to Skyone Studio. The work involved eight workflows between systems, including sales orders, billing, CRM, e-commerce, and operational information related to IoT. Even with limited initial documentation, the team delivered working integrations and produced detailed documentation to allow the client to perform maintenance independently.

These experiences reflect a common approach to Thompson's projects: using technology to solve concrete problems and delivering solutions that can be monitored and managed by the client.

From credit analysis to new AI applications

Clara was Thompson's first step in creating vertical artificial intelligence agents, developed to solve specific problems that can also arise in different companies and segments.

Credit analysis was chosen because it combines a recurring need, a significant volume of operational work, and criteria that can be structured to support automation.

Based on this experience, the company intends to identify other opportunities in the financial, operational, and human resources areas, leveraging the knowledge gained in consulting and mapping clients' processes.

The goal is to find activities where artificial intelligence can enhance team capabilities, improve task execution, and fill operational gaps that arise when competing demands make it difficult to perform certain routines.

Technology applied to business results

Thompson's case study demonstrates how the combination of artificial intelligence, systems integration, and process knowledge can transform a labor-intensive, manual activity into a more scalable operation.

With Clara, the consulting firm was able to automate part of the credit analysis, conduct approximately 260 assessments in 20 days , and reduce the need for new paid consultations with credit bureaus in a significant portion of cases.

The project also reinforces the importance of integrating agents into existing systems and monitoring their results, so that technology does not operate in isolation, but becomes part of the business operation.

With Skyone Studio, Thompson found a foundation to connect systems, structure automation flows, and develop new artificial intelligence applications. The partnership continues as a path to transforming identified customer needs into solutions that increase efficiency and support business decisions.