The AI team is building models based on raw data coming mainly from digital channels (web sites, apps, …). The IT teams are responsible for preparing the data for the AI team. Data scientists monitor their own results using tools like MLFlow, or alike.

BUT

The IT teams are regularly asked to adapt the model of the ingested data, which often changes the model of the data provided to the AI team. 

However, for IT teams it is not possible to anticipate efficiently those impacts, because of the lack of visibility of downstream users.

Hence, the AI team is either contacted too late, or not at all, provoking failures in their projects ⇒ many wrong decisions taken based on data, and ultimately hitting the customers satisfaction.

WITH KENSU (see how)

Kensu offers a holistic view of all data usages, including the quality information at all control points, such that

• The monitoring of the data in all applications from the ingestion to the AI results allows all teams to communicate efficiently about changes
• MLFlow information about ML metrics are extended by Kensu to connect with the up and downstream lineages
• Alerts can be set on the model accuracy but also on the input data quality.

Therefore, impact analyses and root cause analyses are made easy thanks to the end-to-end visibility on data pipelines.

BENEFITS

Time-saving and re-focus on business demands then increase the market shares thanks to the end-to-end monitoring.

• Increase culture of team cohesion, with a clear set of accountability and a system to communicate
• The adoption of existing EDM solutions  increases overall ROIs of all systems in-place, by reducing the human burden

• Managers: report the work performed and its accuracy.
• Business: explain the results obtained and associated decisions.
• CDO: update data catalogs and demonstrate the respect of governance policies.
• DPO (or other compliance offices): show evidence of the viability and risks tied to any projects in the long run

BUT

In order to respond to such demands, all data intelligence work whether it happens in an ETL, ML/AI notebook, or a BI tool must be reported through documentation. 

The time spent on manual documentation impacts the capacity for an enterprise to grow the data monetization without growing the team linearly, and manage the turnover rate due to the associated boredom of repeatedly updating the document.

WITH KENSU (see how)

Participating in the lifecycle of the applications through the generation of real-time logs about data usage allows:

• the documentation about data pipelines, ML/AI models, … to be generated and maintained automatically, and on the fly.
• your data engineers, scientists, analysts, and alike to focus on generating value on data and not documentation, or learn regulation constraints
• data catalogs to remain useful in the long run by generating and maintaining the information they require for their users to always find the best datasets they look for.

BENEFITS

• Reduce boredom and increase communication quality to strengthen the team cohesion.
• Grow and maintain the trust in results produced by applications monitored (stamped) by Kensu DIM
• Increase the confidence and explainability of decisions taken upon analytics results traced and controlled by Kensu DIM

In the retail market, it is crucial to know the customers better and conduct efficient marketing campaigns.
For this, new sources of data are now used: social network, external newsletters, etc.

CHALLENGES

• Use new external/third-parties data in branded marketing materials (e.g. others emails)

• Use personal information coming from external data and create a too precise targeting (e.g. using marital status)
• Use anonymized data in branded mass communication (e.g. « Hello Mr. John Doe »)

WITH KENSU (see how)

• Wrong categories of data are used (e.g. Marital status)
• Forbidden sources of data are included (e.g. Social network)
• Data coming from different environments are used together (e.g. Lab, Production, etc.)
• There is an unexplainable fluctuation of the accuracy/precision of algorithms (e.g. deep learning algorithms targeting minorities)

BENEFITS

• Being proactive and avoid non-viable decisions to be taken (proposing the wrong product)
• Increase of the confidence in final results (launch mass marketing campaigns)

In the banking sector, new competitors move incredibly fast. Due to new products being introduced in the market some segments are at risk. 

Therefore, a bank company is changing its legacy CRM system to increase the accuracy of their understanding of their customers and their products.

CHALLENGES

• The introduction of the new CRM is in trouble due to a lack of visibility on the migration
• The new CRM aggregates more data and created new churning segments without providing a possible reason and solution

WITH KENSU (see how)

• Trace and validate specific business KPIs of each segment
• Drive the discovery of the reasons for churns - allowing the company to react quickly
• Continually validate KPIs to give the company a proactive advantage if issues raise again

BENEFITS

• Ensure the success of the decommissioning of the old CRM.
• Manage and follow the process of reducing the churn rate.
• Accurately track KPIs on how new customers from competitors are converted.

Based on 360° information of their customer base, the insurance company discovered a new segment (younger car drivers) they can address with a new car policy.
This segment is also exposed to other insurance companies, therefore, the new product needs to be launched and advertised before competitors.

CHALLENGES

However, little visibility was available to understand why the product couldn’t be launched on time

WITH KENSU (see how)

• a data was taking too long to be computed (late arrival) and
• delaying the final results (send the advertisement to the young).

The owner of that data proposed to not use as it is based on old product information (parents‘ past policies) which won’t change any more.

BENEFITS

• An acceleration of time-to-market for new insurance policies benefiting from the suppression of a processing bottleneck and the rationalization of the systems
• A reduction of the operation cost of the database and files in the data warehouse
• A reduction in the cost of maintenance of the processing system