Machine Learning workshop for a FinTech company

Client Profile

The client is a company that has developed a platform for accounting professionals. The platform's primary function is to automate data extraction from invoices and integrate it into ERP or finance management systems. The client had multiple ideas for leveraging machine learning.

Challenge

The client faced several challenges:

Legacy system limitations:

Their current AI system for data extraction only supported one language, despite a key business requirement for multilingual support. This system was also difficult to maintain, with minor improvements requiring significant effort.

Lack of ML experience:

The company had no prior experience with machine learning projects and needed guidance on how they differ from traditional IT projects and what key elements require special attention.

Defining success and ROI:

The client needed to understand how to measure success, both technically and from a business perspective. It required a cost analysis to estimate the Total Cost of Ownership (TCO) and calculate a potential ROI.

Risk mitigation:

The client wanted to validate their ML ideas and identify potential limitations early in the process to reduce the risk of project failure and maximize long-term savings.

Solution

We organized and ran a machine learning workshop at the client's office.

Pre-workshop preparation:

We asked the client to prepare sample data and identify key stakeholders, including a decision-maker, someone knowledgeable about data flow, and someone familiar with the legacy system. We also prepared a list of questions to prompt internal discussion.

Workshop execution:

We began by explaining the unique aspects of ML projects. We then analyzed the client's sample documents to identify hidden patterns and gather requirements. We mapped their business process and discussed functional requirements, such as outperforming the legacy system in the supported language and achieving at least 50% accuracy on multilingual documents. We also uncovered critical information about their existing quality control mechanism and data volume, which was substantial enough for training a robust model.

Proposal development:

We used the insights from the workshop to create a detailed, tailored proposal. It included an educational overview of ML project planning, validation of their idea, a cost analysis of different solution variants, and a five-phase project plan with precise prerequisites, milestones, and acceptance criteria. The proposal also provided a visual timeline to serve as a deployment-ready roadmap.

Results

The workshop delivered several key results:

Informed decision-making:

The client received a detailed project plan, cost breakdown, and a solution recommendation, allowing them to make a real, informed choice driven by data rather than guesswork.

Risk reduction:

The workshop identified crucial constraints early on, such as the need for multilingual support and the existence of a high-quality feedback mechanism. These constraints allowed us to remove a milestone from the project scope and reduce costs.

Clear success metrics:

We defined a simple, measurable success metric - the number of errors per document - that was directly tied to the business outcome of reducing manual work for users.

Shared understanding:

The workshop built a shared understanding of the problem and the path to a solution, increasing client readiness to make well-informed investment decisions.

Cost and ROI clarity:

The client received a deep analysis of multiple solution variants, including development effort and long-term operational costs. It provided a solid basis for calculating ROI and projecting when the project would break even.