A Solution to the procurement and management of low-value items in large-scale organizations such as Indian Railways (IR), National Water and Sewerage Corporation (NWSC), Breweries, Industries such as Mukwano, Cement factories etc
What are the procurement problems that IR and other similar large organizations have when procuring low-value items? (4 marks)
- There is a lack of standardization for low-value items. This implies also that their procurement is not centrally coordinated resulting in variations in ordering descriptions by the different zones. There are incomplete and ambiguous item descriptions which gives rise to different interpretations of materials.
- It’s not easy to track their availability in the different depots and stores. This is because field officers use different nomenclature during the search resulting into broad results (hard to work through) or restrictive results (missing out on availability).
- Owing to the complexities around search results retrieval, it’s also not easy to do a price comparison resulting in lost productivity savings that would accrue from such analysis.
- There’s lots of time and effort wasted by field officers during the purchasing process, first during the search for the item, and during the onboarding of new vendors in the case where an item is not found present in the store.
- There’s also lack of consistency against price and quality for different vendors of these low-value items. It’s not possible to track these based on search results. As different field officers return different search results.
Why is the AI approach the ideal and perhaps the only true solution to the problem in this case? Justify your answer.
In Agreement:
The AI approach is the ideal and only true solution to the problem in this case because it utilizes a different search mechanism. The current system utilizes a text-based search that matches words in a field officer’s description to words in the database. This is currently resulting in mismatch problems. First, it’s returning extremely broad results in some cases making it extremely difficult for field officers to swift through those results. In other cases, it’s too restrictive thus missing out on items that are available in the system. Finally, it’s also returning no results in other cases despite the presence of such materials in the stores.
The AI approach offers a combinative solution that utilizes both the lexical similarity and semantic similarity in the search processing. This makes it easy to find past procurements of similar non-stock items but also to find stock items with identical text descriptions.
The AI approach has so far demonstrated better accuracy in terms of matching the search strings with the data base. As such, the AI approach promises better efficiency and effectiveness which will enhance the productivity of the field officers and the store personnel.
The AI approach enables IR to overcome the redundant and unnecessary vendor onboardings which implies better vendor management and tracking as now the Indian Railways must deal with fewer vendors compared to the growing list.
With the ever-growing passenger and freight numbers of IR, the AI approach can be scaled to deal with large datasets and increasing complexity. The advantage with the AI approach, is that the search results get better with more datasets, thus compounded accuracy and advantages.
The AI approach could further be enhanced with a multi-modal generative AI search. For example, an image-based search, where field officers upload an image and text around that image is generated or vice versa. This would enhance search results beyond just text.
The Downsides:
The downside to the AI approach is that it overlooks the people problem, that’s largely around standardization of language and item descriptions. The AI approach also requires large datasets to be trained on to improve its accuracy. First, the existing dataset is full of erroneous descriptions for some of the items meaning the Algorithm is being trained on data that requires clean-up. Secondly, because field officers are always in the process of changing their item descriptions, in addition to some new ones being hired, it poses a possibility of the AI being scored on limited data narrow-sets thus not solving the root cause of this problem.
The AI approach also won’t solve the vendor on-boarding process that’s often unnecessary and long.
The better approach here would be to clean up the process workflow for the procurement of non-stock item. This could start with first, a standardization of the Asset register and ensuring that all non-stock procurements are attached to a particular asset class. Indian Railways also needs to work on a standardization approach for non-stock items to ensure consistency in ordering. For example, when someone in one zone orders for a grinder, another zone should also order by same description (and not cutting tool). Upon this process smoothening and standardization, then the AI solution can build on top of this.
Standard nomenclature lists should be generated, and field officers should be encouraged to adhere to these lists. This includes even the use of material codes (PL codes), once these are well-defined, then an AI solution should be built over.
IR could also consider a pre-qualification of vendors for the common non-stock items thus solving for the long onboarding process. If the common non-stock items are listed and identified in advance, procurement teams can have a vendor selection and onboarding for this so that they’re locked in the system.
If you were Reddy, how would you create a (simple) decision support system using the data from Exhibit 2 to show the most relevant item – based on the search string – as the first item and subsequent items ranked on the basis of the similarity score?
Based on the Exhibit 2, the following challenges were identified in the dataset:
- Different materials are sharing the same PL/Material Code. For example, for PL 001, there are about four materials, one is a printer, scanner, chairs and photocopiers.
- Purchase Order (PO) number does not follow a standardized nomenclature
- PL nomenclature not standardized.
- Only a long text description exists in the data set.
- No standardized nomenclature for the Consignee. For some, they have numbers, others it’s letters
- No specific column/category exists for the Vendors/Suppliers
- Duplication in PL codes, same material have different PL codes
- Incomplete data strings, missing division names for some materials. You can’t identify the department the zone or depot that ordered
- Insufficient descriptions for some of the materials.
- Supplier’s details mixed up with brand name and original manufacturer’s name
- No information on quantity and price in the dataset
- Services are being ordered as store items
Decision Support System Solution
- Step One: Data Clean-up. Flag items with incomplete descriptions, missing details of Zone or Depot, similar items but with different PL codes, and all items with challenges listed above. Differentiate Services from Materials with a unique identifier. For example, all Materials should start with M, and services S. Services shouldn’t have any material/PL codes. Services will always have a quantity of 1 in the new dataset.
- Step Two: Data Preparation and Standardization: Introduction of standardized PL/Material codes nomenclature. New Nomenclature to feature:
Material/Service|Type of Material |Store Location | Asset Number | Item Number
For example, a chair in Central Railway Zone could be identified as below:
M|BR|20|10|02
Where:
M- Material
BR- Bearing
20- Depot/Store Location
10- Asset Number
02- Item Number
- Step Three: Introduce new categories in the dataset and delete redundant categories. First category is that of the Vendor with a unique numbering for every vendor. Second category is for current Price (price can be based on a moving average or based on purchase price for current item in depot), third category is for the store locations, fourth category is for Asset Numbers, there should be an asset register in Indian Railways and all material purchases should be attached to an asset. Introduce a category for Quantity
- Step Four: Introduce both a Short Text Description (standardized across Indian Railways featuring common descriptions such as Fan, Pump, Chair) and a Long Text Description where specifications, and other details can be attached
- Step Five: Expand the search Options. One can search by Material/PL code, one can search by the short description, one can also search by vendor. The search at all times should return the item with the price, quantity and vendor.
- Step Six: Lock up the search by long text, it should only be available after the initial search by the short text.
- Step Seven: Integrate a semantic/relational based artificial intelligence algorithm over the newly cleaned dataset.
- Illustrated New Search Dashboard based on Search string as first item
Based on this new decision support system, the field officers can quickly search for the required item if they already have a material code, the stores teams can always share material code list every month, the procurement teams can analyse inventories and see which materials are driving it, or which vendor is the biggest contributor, analysis can also be made by consignee. It’s also easy to tell services and materials apart.
What are the expected business benefits of AI-based problem solving in the context of this case?
- Increased search efficiency which results in increased productivity. First, field officers don’t have to waste time and effort trying to find an item in the database. This time previously wasted on purchasing can now be re-dedicated to the field operations thus improving service operations. Currently, field officers fail to find the required non-stock item on 15,000 occasions. With the AI accuracy of 85%, these will be reduced to just 2250 occasions with more improvement based on the larger datasets.
- Cost-savings in terms of reduced inventory as IR doesn’t have to purchase already existent non-stock items. Furthermore, because of this, procurement teams are saved from the strenuous and long vendor-onboarding process, as many vendor additions are currently unwarranted, and simply a result of the inefficient search solution.
- Cost-efficiency and effectiveness arising from the ability to compare materials and vendors against past prices and quality thus giving better negotiation advantages to Indian Railways and driving more effective procurements.
- Accelerated Decision-Making, this is arising from the speed at which the AI solution returns results, and the accuracy of these results. This enables more data-driven decision making at Indian Railways resulting in improved managerial competencies and output.
- In the increasing world of BANI (Brittle, Anxious, Nonlinear and Incomprehensible) world, AI powers adaptability and agility for a large organization such as Indian Railways enabling it to cope with increasing complexity, increasing procurement needs, and requirements.
- Strategic and Competitive Advantages accruing from AI’s ability to optimize existing processes and improve automation, information and transformation effects, through the predictive nature based on the clustering and semantic capabilities, AI is in position to also study patterns in the data and enable more future-proofed material management.
- Personal at Scale: By adapting to each field officer’s unique needs and description requirements, AI enables Indian Railways to build a more diverse and inclusive business environment thus spurring the creativity of its human capital through this personalized interaction and search system.
- Improved Business Forecasting and Business Intelligence: By being able to handle complex and large sets of data, the AI approach enables better forecasting of delays, it will show usages by different zones, control inventories by tracking the stocks at different depots thus reducing potential downtimes that arise from stock depletions.
- Fraud Detection: There are lots of non-compliance that arises from field officers intentionally ordering for the same item using different descriptions. With the AI approach, this can now easily be detected and located thus ensuring more efficient and compliant systems with near-zero fraud.
- Coordinated and Centralized Procurements: By being able to cluster these different non-stock items, the procurement of the same can now be centrally coordinated and procured thus enabling better savings from large-scale procurement, but furthermore, it prevents over-stocking by different depots as two depots don’t have to stock the same non-stock item, especially those co-located near each other.