For Afreximbank, the African Export-Import Bank, STEPS designed and delivered in 2025 a system that continuously measures the sentiment expressed about the bank. A data lake centralises content from more than 100 websites and external sources, real-time pipelines clean and structure it, then natural language processing (NLP) models analyse it to produce an overall sentiment index. The project shows how data and AI turn scattered volumes of text into a management indicator.
The need
A large institution is mentioned every day in the press, on social media, at events and in exchanges with its clients. These signals are valuable but scattered, in several languages and in volumes impossible to follow manually. The goal: a consolidated, regular and objective measure of how the bank is perceived.
The solution
- Collection: connectors continuously retrieve content from more than 100 websites and external sources.
- Data lake: raw data is centralised in a platform able to absorb large volumes and varied formats.
- Real-time pipelines: stream processing (Kafka) cleans, deduplicates and structures content as it arrives.
- NLP analysis: natural language processing models detect sentiment and topics.
- Index and reporting: results are aggregated into an overall sentiment index tracked over time.
Why a data lake
The data was mostly unstructured text from heterogeneous sources arriving continuously. A classic data warehouse, designed for structured tables, was not suited. The data lake stores raw content and makes it available to AI. We explain this choice in our article on the difference between a data lake and a data warehouse.
Lessons from the project
- Start from the business question: what management wants to measure, and how often.
- Take care of quality upstream: deduplication and cleaning determine the index’s reliability.
- Favour real time when the topic moves fast.
- Make the result readable: a simple indicator tracked over time beats a mass of data.
These principles apply to any company that wants to monitor its reputation, customer satisfaction or market signals.
STEPS and Afreximbank
Afreximbank is one of STEPS’ main clients, and STEPS is also the integrator of the pan-African payment platform PAPSS. Beyond banking, STEPS supports companies in every sector with data and AI engineering, generative AI and dashboards. For a similar project, book a scoping session.
Frequently asked questions
An indicator that measures, from published text (press, social media, reviews, exchanges), the positive, neutral or negative tone expressed about an organisation, and how it changes over time.
A data lake to centralise data, real-time pipelines based on Kafka to structure it, and natural language processing models to analyse sentiment.
No. Any company that wants to monitor its reputation, customer satisfaction or market signals can use the same approach.