Find out how we saved our client huge amounts of time through AI-powered automation

Background

A leading research provider, which runs a huge worldwide B2B project for a banking client, usually spent months managing and manually checking the sample lists, which consisted of millions of businesses operating across a diverse range of business sectors.

They wanted to find a way to take some of the pain out of the checking process.

Approach

We applied machine learning techniques to historic data so we could teach AI models to automatically identify which businesses should be removed from the sample and which were incorrectly categorised.​

It was essential to ensure our models, which were based on both text and categorical data inputs, would generalise to ‘unseen’ data which hadn’t been used for training purposes.​We separated test and training data and used cross-validation to find the best model without overfitting to the training data.

Outcome

Our models, were able to correctly identify which records to remove or re-categorise with extremely high accuracy.

As such, the project delivered huge savings in terms of time and human resource required, even allowing the team to make up lost time for delays due to the Coronavirus.

Furthermore, by using a probabilistic modelling approach, we were able to vastly reduce the amount of  human quality-assurance checks needed because the model flagged up which records it wasn’t sure about, so human checking could be focussed on a targeted subset of records.