Customer Request Learning Team-Suite (CRLT)

Procreation a NEXT Step in manufacturing
Customer Request Learning for quickening service & adding understanding in Planning and manufacturing

1. Inspiration

We at AOEC find that "procreation" will help manufacturers mitigate the crisis in the demand/supply equation.

We at AOEC have hosted a proof of concept URL to conceptualize how this can be done. As a part of this ... The Customer Request Learning Team-Suite is a framework of Customer Request Learning Tools that use machine learning for different sectors of the manufacturing industry.

The project showcases Customer Request Learning for an Automotive Parts Manufacturer and Service Centre framework.

About AOEC AOEC stands for Akaash Open Enterprise Centre (a Gap analysis and problem solving consultancy) with a team comprising of myself (K.S.Venkatram) and Aakkash K V (Final year, BTECH Automotive Engineering, MSRUAS).

2. Problem solving (background)

An Automotive Parts Manufacturer and related Service Centre framework needs to group similar service or remedy requests about parts into a new classification, so that the classification can be used to send out standard or pre-defined resolution.

The issue being that the Automotive Parts Manufacturer focuses on manufacturing multiple categories of parts, where the SMART Manufacturing concept with sensor enabled control systems improve production volume and achieve high levels of quality.

However the staff allocated to receiving customer requests need to have sufficient product knowledge and also need to remain trend-savvy to resolve queries, requests or issues. Some of this is time consuming and number intensive.

The current economic downturn has affected the company and it needs to control costs.

The cost cutting drive is on, where the calling is for a reduction in the money spent on acquiring markets, ensuring sales & enabling of associated customer services.

Reduction in manpower is a part of this cost cutting strategy.

The financial reports highlight there is a decrease in market share, so the company needs to manage production volume and also respond effectively for all service or remedy requests but with new reduced time-spent-per-request-or-call “rules & regulations”.

The multiple parts being manufactured

The Automotive Parts Manufacturer focuses on the following part manufacturing: Focus 1: Tyres Focus 2: Mechanical parts Focus 3: Body Shell Assembly Focus 4: Interior Trim Focus 5: Seats Focus 6: Small parts like rotors in electric motors for injection pumps Focus 7: Modules like the top and bottom parts of the fuel tank Focus 8: Exhaust systems and parts like the muffler and exhaust pipe Focus 9: Electronic parts like populated printed circuit boards and their components The problem on hand is to reduce the time spent on each request or call for the parts, where customers could need customer service or remedial steps.

3. What it does (Solution and Approach)

The Customer Request Learning Tool (CRLT) reviews customer requests to find that service or remedy requests coming to the Centre

Can have repetitive text in the description or problem details
Can have some standard steps that the customer can take to find resolution for some questions or problems
Can have responses / resolutions that do not need human intervention

4. Inference

The CRLT experts infer that buckets of service or remedy requests can be created through a machine learning algorithm based on the past service requests.
For each bucket, a set of corresponding standard resolutions can be associated to reduce time spent per request or call.
Whenever a new service or remedy request comes to the Service Centre, a machine learning algorithm would identify the bucket to which the new request should be tagged to, where this should help come up with a standard resolution and then help report the same across to the customer.

5. Methodology

In the solution,

The service or remedy requests in the repository are clustered using a combination of (a) Text-analytics of “text fields” with select descriptions, (b) the time estimated or taken to resolve each request and (c) a categorization variable that categorizes the nature of request
The Text-analytics technique is based on Word2Vector
The clustering technique is based on DBSCAN
The Cosine similarity algorithm is used to classify service or remedy requests to fit within one of the buckets created (where this is based on text categorization)

6. How we built it

We at AOEC are developing the idea using the Python & Anaconda framework and different libraries for data analysis, array processing, Natural language processing, Text-analytics & clustering, visualizing of clusters, request or remedy description similarity

The details of the libraries follow:

Specific libraries to load data, perform computation and display output are (a) Pandas – Data acquisition library (b) numpy – Array processing library (c) nltk.data and nltk.corpus – Natural language processing library (d) gensim and gensim.models – for text analytics and clustering, where the Word2Vector function is used (e) gensim.models.keyedvectors – to import keyed vectors (f) matplotlib – for visualizing clusters (g) sklearn.cluster – to import DBSCAN for clustering (h) sklearn.metrics.pairwise – to import cosine-similarity to find out request description similarity

Code snippets in the proof of concept (step wise)

(1) To import libraries and functions (2) To load data (3) For filtering of requests based on groups for “parts and requests categorization” (where there are 9 Focus groups and one Transaction Hub category, it is noted that the Transaction Hub category can be exploded further when the proof of concept is developed into a complete application) (4) Text analytics to create the training data for the machine learning algorithm (5) Running of the clustering function (6) Assigning of a new service or remedy request to a correct bucket based on the cosine-similarity function

7. Challenges we ran into

In the Automotive industry, there are many parts manufacturers. So we needed to categorize parts manufacturing based on an emergent automation and control systems technique, where we found a real-world example for the same.

It is called SICK (a business concept for Factory, Process and Logistics Automation) where sensor control and its intelligence is used to make the complete Plant to Service life cycle more operations intelligent, intrinsically safe-and-resource-effective

8. Accomplishments that we're proud of

Application of real-world illustrations in a Procreation Team Suite that we intend to design further.

9. What we learned (Conclusion)

Machine Learning Algorithms help us use past understanding or today's details to ideate and enable solutions for corresponding or standardized resolution, where machine learning can quicken customer service and help re-engineering at the planning & manufacturing level,

10. Future Scope

Building more scope, intelligence and functionality in the Transaction Hub analytics to design more service intelligence and ensure continual improvement in stages of planning, operations and reworking of multi-nature customer service solutions via automation and machine learning for (1) Product or Part support (2) Solutions Training and Education (3) Value added functions for customized-innovation to suit the customer and to enable end-experience-safety

11. What's next for Customer Request Learning Team-Suite (CRLT)

We know manufacturers are in a time frame where human resources are not available as earlier and money spend & cost dynamics are high. We will take the next steps in designing a more multi-purpose Customer Request Learning Team Suite. We will use and elevate this fundamental concept in the Procreation Centre, in a solution deployment level, that helps manufacturers and their customer service networks mitigate the current “RoI number crunching” demand/supply crisis.