‘Demand’ is one of the most vital factor in planning capacities, production schedules and supply chains of manufacturing, retail and services companies. Goods – both finished goods and raw material – worth millions of dollars are in the ‘pipeline’ and any shocks in demand impact the entire supply system resulting in either stock outs, higher inventory costs or wastage of goods and material. A nirvana state is when suppliers are able to predict demand well in advance so the lead times for procurements, production and delivery to points of sale can be planned better. If demand is predicted with fair accuracy, the producers can price their products intelligently – start raising the price when demand is estimated to go up while prepare to offer discounts when demand is estimated to reduce.

So how does one forecast demand? While capacities and lead times of factory production, suppliers and logistics are easier to predict, it is often difficult to predict end customer demand. Because demand depends of many external factors like weather, consumer behavior, competitive forces, single events (e.g. strike, natural disasters). Depending on the business need, demand has to be forecasted at different levels of granularity, e.g. at geography, customer segment, product category or at SKU level.

Normally, demand forecast is done in organizations using simple mathematical computations on excel sheets. We leverage Machine learning techniques by using statistical algorithms to decipher demand patterns (e.g. seasonality, peaks) and establish interrelationship between demand and factors like weather, customer demographics, category growth, events, population growth and health of economy. We use forecasting techniques like time series, decision trees, Support Vector Machines and K-Means clustering. Demand forecasting modelling generally requires past data (e.g. at least 2 years data to capture the seasonal trends) like company’s own sales (e.g. by geography, category, SKU), competition sales, macro factors, demographic changes and other environmental factors.

Predicting demand with reasonable accuracy can save organizations millions of rupees in terms of optimized inventory, reduced wastages, low production losses, intelligent pricing and minimized sales opportunity losses.

A typical organization procures thousands of goods and services from as hundreds of vendors. Procurement activities call for tremendous manual effort for activities like requirements documentation, vendor evaluation, drafting purchase order, invoice processing and contract closure. These tasks require perusing piles of documents and processing high volumes of data on systems like ERP, CRM and Accounting software. We have helped clients in digitizing these processes using intelligent automation technologies thereby reducing costs and improving process quality.

The building blocks of intelligent automation include Robotic Process Automation (RPA), OCR (Optical Character Recognition) and NLP (Natural Language Processing) among other technologies. Intelligent OCR can be trained to read PDF documents or Images like Invoices received in hundreds of different formats and extract required fields like Invoice number and GSTIN. Further, RPA enables operating enterprise systems like ERP, CRM and Accounting in terms of logging in, navigating screens, processing data and entering data fields. For example, RPA and OCR can read documents sent by vendors in PDF, MS Word or MS Excel to register a new vendor in the vendor master of ERP. And intelligent automation can pickup data from proposals sent by vendors in MS Word and MS Excel and create draft Purchase Orders in a standard company format. It can also navigate through the SAP or such ERP systems and check for pending items in a PO and email the summary to all the suppliers.

We bring our process knowledge and experience in OCR, RPA and NLP technologies to automate procure to pay processes. This helps clients in reducing high manual effort on routine tasks allowing officers to spend more time on analyzing vendor performance, evaluating goods quality, optimizing inventory and reducing lead times.

Each producer, be it a steel plant or a car manufacturer, deploys thousands of equipment which are sourced from multiple OEMs and have their own lifecycle and maintenance needs. For producers, maintenance planning of these machines and equipment is critical to ensure minimal downtime. Timely maintenance is even more critical for producers who use high cost equipment because loss incurred from every minute of downtime is very high. Manufacturers use various approaches like Preventive, Reactive or Predictive Maintenance, and Predictive maintenance is the most optimal of these, but it also requires scientific methods.

So how does one plan Predictive maintenance? The central premise of predictive maintenance is to carry out repairs just when it is required to do so i.e. when a part or an equipment is nearing close to failure. It means we put efforts for maintenance activities only when it is really required to do so. But how can we predict when an equipment is close to a point of failure and hence needs a maintenance? We help clients predict this by deploying machine learning models.

When a machine is expected to reach a failure point depends on various factors like age of equipment, events (e.g. change in a spare part), asset usage (e.g. hours of running), triggers (e.g. high temperature alarms), health parameters (e.g. vibrations), history of maintenance activities and failures. Most manufacturers capture these vital data in digital form often using IoT devices, SCADA, PLCs and instruments to capture data like temperature, moisture, oiling, etc. Often these data are captured at frequency of minute or hour and hence results in Big accumulated data over a period of time.

There are various machine learning techniques that can be used for predictive maintenance including Regression model to predict Remaining Useful Life or Time before next failure, SVM or neural networks. Inputs from Original Equipment Manufacturers (OEMs) and maintenance & repair engineers are critical for building a Predictive Maintenance model. The end goals of these models have to be optimizing maintained frequency to an extent that the ratio of ‘cost of conducting maintenance’ to ‘loss due to down times’ are reduced compared to earlier benchmarks.

The journey of Predictive maintenance requires iterations and refinements but can be replicated across equipment of a manufacturing unit and offers huge cost saving opportunities to producers.

Locational intelligence or the ‘science of WHERE’ is a critical aspect in managing business. Visualization of business data like sales, customer segments, distributors, inventories and logistics on a geographical map improves ability to assimilate and analyze business. We bring leading Geographical Information System (GIS) solutions to your strategy teams so they can add locational intelligence to business data and thereby take smarter business decisions for achieving growth. Here are some examples,

• Understanding customers – analyze clusters and density of customer segments across geography for assessing untapped potential and cross-sell opportunities
• Target marketing – identify geographical clusters of potential customers with lifestyles and buying patterns that your products and positioning
• Site selection for customer service centers – identify customer clusters across districts to setup customer service centers at centrally and well-connected locations
• Defining Sales territories – define sales personnel territories on a map based on criteria like business potential, geographical span and target customer segments
• Stock status – visualize stock levels at warehouses and retail outlets across the country at SKU-level to identify spots with potential stock-outs or over-stocking. Plan logistics to move inventory smartly