× STEPS
Case study | Real-time NLP, Afreximbank

Business Sentiment Index, reading the mood of an ecosystem

For Afreximbank, STEPS built a data lake that collects information about the bank’s community from more than 100 web and social sources, and analyses it with real-time NLP to produce a continuously updated sentiment index.

100+ sources
Real-time Kafka
Dashboards and alerts
01 - The context

Knowing what is being said, before it matters

A pan-African bank operates in an ecosystem of countries, companies and partners. What is said in the press and on social media influences its decisions, but nobody can read everything.

!

An impossible volume to readHundreds of articles and posts every day, in several languages.

!

Weak signalsA negative trend on a country or sector is often spotted too late.

!

Manual monitoringPress reviews took time and remained partial.

02 - The flow

From raw text to index, continuously

Each article or post goes through the chain in seconds: collection, cleaning, analysis, aggregation.

01Collection

100+ web and social sources.

02Data lake

Storage and history of all content.

03NLP

Cleaning, enrichment and sentiment analysis on Kafka streams.

04Index

Aggregation by country, sector and theme.

05Alerts

Dashboards and alerts for the teams.

03 - The interfaces

A geographic view and a news watch

Ratings map

Economic stability and sentiment by African region, with a breakdown by category.

Analysed news feed

Each article is summarised, classified and scored, with its impact on the index.

100+web and social sources monitored
24/7continuous collection and analysis
1single index to track the mood of the ecosystem
04 - The benefits

What the index brings to the bank

Monitoring turned into an indicator, readable by teams and management alike.

Main benefitA head start on information

The bank sees sentiment trends on its countries, sectors and partners continuously, and spots weak signals before they become risks.

RiskEarly alerts

A drop in sentiment triggers an alert for the teams concerned.

DecisionAn objective indicator

A numerical index complements the analysis of country and credit teams.

TimeAutomated monitoring

No need to read every article: what matters surfaces on its own.

Coverage100+ sources monitored

Press, institutional sites and social media, continuously.

HistoryMeasurable trends

The data lake keeps the history to compare periods.

ReputationThe bank's image tracked

What is said about the bank and its programmes is measured.

05 - Our work

From data lake to dashboard

PHASE 01Sources

Selection of web and social sources and setup of collection.

PHASE 02Data lake

Storage, history and cleaning of collected content.

PHASE 03NLP

Real-time sentiment analysis pipelines on Kafka streams.

PHASE 04Reporting

Index calculation, dashboards, map and alerts.

06 - Architecture

Under the hood

A real-time data architecture, built to scale.

Data lakeKafkaNLPPythonDashboards
  • CollectionConnectors to 100+ web and social sources
  • StorageData lake with the full content history
  • StreamsKafka for real-time processing
  • AnalysisNLP pipelines: cleaning, enrichment, sentiment
  • ReportingIndex, country map, news feed and alerts
FAQ

Frequently asked questions

Where does the data come from?

From more than 100 public web and social sources: press, institutional sites and social media.

Is the index updated in real time?

Yes. The NLP pipelines run on Kafka streams and the index is updated continuously.

Can a country or sector be tracked?

Yes. The index is broken down by country, sector and theme, with configurable alerts.

Want to measure what is said about your ecosystem? Let's talk about your AI monitoring.

Book a scoping session ↗