Building a data driven supply chain with Cortana Analytics

EXECUTIVE SUMMARY

Dynamics partners have a huge opportunity to reinvent supply chain scenarios and deliver a data-driven application that uses machine learning, optimization solvers and other advanced analytics to embed intelligence and surround the traditional rule-based automated business process enabled by Dynamics. They could do this by dynamically predicting supply chain input. Instead of asking a supply chain planner to enter the ‘temperature’ of a healthy supply chain and then assume that the supply chain runs on that ‘temperature’ all the time, new solutions and services should derive insights and predictions from ambient data, predict what ‘temperature’ is likely to be and provide alternate execution paths to the supply chain planner. An application like this that embeds Cortana analytics suite components into Dynamics business process and is delivered on Azure platform can be built using existing technology - will provide the ecosystem players with most bang for the buck, short time-to-market, will appeal to existing Dynamics customers, will showcase power of Microsoft platform and will provide unmatched differentiation.

SUPPLY CHAINS 2020

Just like service sector is abuzz with talk of how Big data has potential to revolutionize various industries like financial services, healthcare, tourism, similarly in the supply chain field there is no dearth of “experts” claiming how Big data and related technologies will make and break business models. Just like in late 1990s there was a big wave of investments in e-commerce, today coupled with Big data, digital commerce trends are shaping the thinking of supply chain executives who want to build a supply chain that an listen and learn and sense and shape demand. If “every company is a software company” is true, there is truth in “every trader wants to be an Amazon” and “every manufacturer wants to sell an experience and not just a product” . Just like in 1990s financial services firms particularly traders and brokers moved to a high speed digital system reliant on advanced analytics delivered in the back office by data scientists, today supply chain executives are looking to use similar technologies. Supply chain planning will become much more automated and will be dependent on scientific models rather than being based on manual transaction processing systems. Just like in late 1990s supply chain executives in large global firms bought every software that came with a promise of improving supply chain, today again there is an appetite. However, there are key differences and those who understand the differences can capitalize on this opportunity:

  1. Supply chain executives want solutions that help their supply chains become resilient and/or agile, not just reliable. And in some cases for supply chain leaders they are looking for demand-driven approaches.
  2. Larger companies are still digesting all the software from 1990s, real appetite is with SMB companies who are looking towards cloud computing vendors to provide them such services at costs they can afford.
  3. SMB executives want point solutions so they can quickly try them and they also want end-to-end solutions in the cloud
  4. SMB executives are more happy to try such solutions in a subscription model
  5. CIO maybe a Chief Integration Officer in larger enterprises but SMB segment is mindful of integration cost and measured about what IT or LOB executives buy

INTELLIGENT BUSINESS PROCESS

It is true that cloud offers more choice to customers to switch between vendors, however it is also true that this new found convenience will become the norm and IT will find the task of moving to another cloud as onerous as in on premise world. Microsoft provides one of the best and easiest to use full lifecycle data management tools in Azure and some will consider this enough incentive for customers to stick to Azure. However, economics will force large platform vendors to move beyond tools. Just like with the advent of power utilities Edison Electricity Company moved to become GE to provide not just electricity grid platforms and components but also industrial and household appliances, Microsoft together with its partners will reinvent and provide higher value adding applications on its’ data platform. More data-driven applications will get more data which will spur demand for more applications. With MS strength in the enterprise, and GTM geared towards enterprise buyers’, business applications are a natural area to showcase such advanced analytics applications. Within business applications - sales, marketing and supply chain applications are hottest from a VC investment point of view and Microsoft Dynamics applications provide the necessary beachhead in the growing SMB market – and therefore this is the time for Dynamics partners to lead the way.

Microsoft partners should reinvent scenarios in their vertical and deliver a data-driven e2e application that uses machine learning, optimization solvers and other advanced analytics to embed intelligence and surround the traditional rule-based automated business process enabled by Dynamics.

Supply chain specific Discussion

Let’s look a level deeper by considering an example of what could be done. Economic order quantity (EOQ) is the number of widgets you order so ordering costs and inventory holing costs are in balance. We should ask ourselves how someone can really know EOQ when purchasing costs are so hard to nail-down and inventory carrying costs vary depending on geography, distribution center, maintenance, insurance, taxation and change with time. Could it be that EOQ is just one of those highly-averaged broad-brush measures that was good enough to give a scientific “feel” to punch-card era accountants of 60s and 70s? Isn’t it obvious that measures like these were invented by statisticians for accountants who both lived in some data dearth decade? We should ask ourselves – is an individual click on internet really more value-adding than an average product stored in a warehouse. My opinion is perhaps not. Then why is it that it is possible to tag, track & monitor clickstream data and use it to offer a very personalized advertisement for each one of us say, valued for hundredth of a penny but we are happy to use one gross overall EOQ average and keep ordering same quantity of product for years to come with complete disregard to actual demand at the time, cost of product being ordered etc. and destroy millions of dollars of value in the process? Is that because of some grand plan somewhere or is it simply because data science hasn’t reached the supply chain domain yet? No one really knows EOQ but it has been the centrepiece of inventory management ever since such solutions were first implemented in 70s. Millions of products are ordered, stocked, sold based on such unscientific static measures and millions of dollars of value is lost in the process. Many best in class companies know this and are reinventing such systems like Amazon.

Same applies to reorder level. Every company who uses reorder levels sets them statically once for a fairly long time. If they plan to revise this number they will probably look at ABC classification and do so for select high-value fast-moving A class items. They would never get to do more than that and will never get to long-tail products. It’s not an exaggeration to say that the mathematical ingenuity of the whole supply chain discipline is contained within the bounds of Pareto principle! Most software solutions will only allow possibility to set one value, some more recent solutions do dynamic calculation of this metric. Safety stock as the name applies is used to compensate for forecast error. Again most traditional vendors offer a possibility to setup static value per product-location. Same thing can be said about minimum quantities, maximum quantities and standard order quantities in both distribution and production.

Above are just examples. We can ask same questions for every metric, every process. Every time we make a decision based on space-time bound metric we can ask - will capturing more data make the metric more fluid in space-time, more continuous instead of discrete? Will it improve quality of decisions? Is airline pricing more fluid than concert tickets than the price of that polo you want to buy? Probably, Yes. Does a hotel reservation system know its’ inventory (hence availability) at any time? Probably yes, but does a retailer know? How can we reduce data latency between two nodes in a supply chain?

Delivering point solutions like say predicting reorder level of inventory or predicting willingness to pay for retailers or distributors of consumer goods or simply, showing customer segmentation could be one of the options to show the potential hidden in troves of data. Another option could be to build vertical specific apps focused in a particular domain/problems space that combine many such prediction and optimization services together in a chain of decisions which could answer higher level questions like “which is the best distribution center to service an incoming customer order from?” or “what is the best price for a product given customer and channel”, “how best to price a service contract” or “which are most valuable customers”? Cortana Analytics suite has some apps that illustrate this point. Recommendation service is in instance where the cross-sell and up-sell problem has been rethought and a traditional rule-based system has been upended. Similarly, Dynamics partners could reinvent several other sales, marketing and supply chain problems.

However, it’s important to realize that while advanced analytics provides the most needed deep insight, the number of users it touches is fairly limited in most scenarios. Customers are willing to pay top dollar if such insights are integrated into a business process and create a sort of decision management system – that reduces variability and improves reliability and repeatability of the process that most of the organization is engaged in. In the supply chains of future all benefits will be directly related to the speed of flow of materials and information. Therefore, to derive greatest long-term value and stickiness Dynamics partners should integrate insights deep into a business process so a business user like a (1) customer service rep can automatically get the suggested price from Cortana analytics suite while she is in the process of creating a sales order for a customer order within the context of order management process or (2) so that the system automatically choses the right distribution center to ship from while consumer entered order is being logged into Dynamics.

Since most companies even in SMB and mid-market space have multiple business solutions, to provide customer with maximum value at the outset, Dynamics partners could use Azure data catalogue (part of Cortana analytics suite) to publish the result set of such advanced analytics apps into the company data catalogue and provide custom visualizations in Power BI so that the data latency is kept to minimum and business can quickly utilize the results where they want and IT can quickly build the required integrations to in-house or other 3rdparty solutions which may lie outside of Microsoft ecosystem.

OBSERVATIONS FROM ADVANCED ANALYTICS PoCs

Not all projects mentioned below reach a conclusion, some of those that reach a conclusion are not necessary successful. However there is tremendous learning on both sides on how to craft such solutions.

 is a food service company that distributes food products in UK. They first came to us because they felt that after moving their shop online they have significantly improved order taking process however due to reduced human interaction with sales agents but they have lost the ability to cross-sell and up-sell. They implemented Azure ML recommendation service and derived 5% uplift to revenue. They are using it in live environment since Nov 2014. In the beginning such a large lift is expected in environments where there are rich pickings to be had if such an analysis has not been performed earlier. Even after one year in Nov 2015, the lift is between 1-2%, which proves long term value of the service. Then they implemented a self-developed Azure ML model that predicts what their B2B customer (a restaurant owner, a school canteen manager) is likely to buy when s/he logs online. This has helped improve order taking efficiency even further. Then they implemented a simple algorithm to classify their customers which their call-center agents now use to determine what sort of offers/discounts they can offer to that customer - classification is automated but not offers - something that can be done in future. Then they integrated their recommendation engine to the call-center application running on Dynamics AX platform. Now call-center agents can complete orders even faster and can cross-sell and up-sell on phone. Being in a true Omni-channel environment, they built an iPhone/iPad app so their customers can use the app to place orders into Dynamics AX backend. All the Azure ML services including recommendations are surfaced up in this app. Now we are experimenting Azure stream analytics to automatically send alerts if temperature readings go beyond acceptable limits in any of the delivery trucks on the road. We are also experimenting product pricing with them for their bi-weekly fixed-price catalogues. Customer is really satisfied and is a showcase of how automated business process in Dynamics can be made more intelligent by surrounding it with such Azure based services. Company A has a large IT team which is a key reason they have been able to build out quite quickly. CIO recognizes that there are massive efficiencies to be gained in all parts of their business by using more data-driven solutions. Privately held. US$300m. UK.

A list of other ongoing PoCs that are not public yet.

Company B is an industrial distributor that procures quarter of million different industrial supply SKUs from 16000 suppliers and supplies them to its customers. Company is experimenting with us on demand forecasting and inventory optimization as part of their ongoing Dynamics AX implementation. They would like us to be able to predict demand with high accuracy for non-cyclic, non-seasonal products with sparse demand. Then they want us to build them a system that predicts correct safety stock, re-order level and order quantity for various products. CIO recognizes the lead that his company can get by optimizing and making dynamic decisions to improve efficiencies in a volume spare parts business. Privately held. US$250m. NJ, US.

Company C ,  is a typical distributor that provides financial, IT and administration support to hundreds of companies in the group all across US and Europe. They are implementing a new Intershop based e-commerce portal for B2B sales of their products. They were looking for a revenue lifting technology that will help sweeten their e-commerce portal offering internally for their group companies. They are very excited with the models built for them using the recommendations engine. They are evaluating cost of the service, cost of the integrations to their Dynamics AX based call-center and other internal systems before they go-live with the recommendations offering. Company has an own IT team so once after their decision, they could go live quickly with minimal assistance from Microsoft ecosystem. In future, they will like us to advise them on various inventory management aspects. Public. US$5B. NL.

Company D is the largest food company in Israel. This is still a Dynamics AX prospect but was very impressed with recommendations engine in the sales cycle. Sales team built a model to show them the value and now they have started a separate RFP process for recommendations offering. They already own most of the market they can in Israel and see recommendations as the most cost-effective way to increase revenues by cross-selling other products to existing customers. Public. US$1B. IL

Company E is an electrical parts distributor that distributes 400k electrical SKUs to its customers through its many channels including 80+ physical stores all over Italy. Company is experimenting with us on recommendations which they see as a significant way to grow their revenue given their massive product catalogue and complexity. They are keen to work with us on demand forecasting and inventory optimization as part of their ongoing Dynamics AX implementation. Initial analysis of their data shows about 1000 products heavily used and an extremely long tail of products that no one is adding to their carts. A lot of potential to save costs and focus efforts on high value products. Private. US$500m. IT.

Company F is an automotive parts manufacturer that manufacturers and supplies parts to RV (Recreational vehicle) market in US. Company experimented with us on demand forecasting. We built a model for them that had lower forecast errors than ForecastPro an old PC based solution with a battery of algorithms. However they are not willing to adopt the model unless we build a full data pipeline and Dynamics integration for them so they can readily consume the forecasting results in their Dynamics AX and NAV applications with a click of a button. Company is one of these customers where the supply chain professional, procurement department is very old-school and stuck to their EOQ, re-order type calculation and cannot see the light beyond this type of calculation. Although the director, supply chain is very innovative and has been very impressed by Cortana Analytics capabilities. He wants us to built advanced analytics based supply chain solutions as long as we can build a first-class consumption
experience (e2e solution, not point solutions). US$750m. IL, US

Company G is a computer parts and accessories distributor to private individuals and businesses. They engaged with us last year to implement recommendations service together with their Dynamics AX implementation both online and in the call-center. However, the Dynamics AX project is delayed, IT directors were changed a few times so this engagement is stalled. Although what was clear was that with more than 100k products in their catalogue they could easily see how their existing solutions were not sufficient enough to provide them with any
suitable inventory management, distributed order management and trade/promotions management that they desired. Private. US$1B. US.

Company H is a convenience store chain who was looking for a solution to cross-sell and up-sell to its shoppers at gas stations spread through the country. They experimented with recommendations engine but didn’t come back. The results didn’t look very exciting. It seemed that most their shoppers in the morning bought a croissant or a sausage roll and a coffee or a Coca-Cola. Perhaps they could be offered a newspaper too. And all their evening/night shoppers asked for a pizza slice and a drink. These seemed like very set patterns and a basket of complements for which there didn’t seem much more to do. Whatever there is to be done can be done by the POS agent like offering a chocolate, chewing gum or a bag of chips. Private. US$200m. NO.

Company J is a food science company that develops flavors, aromas and textures for large food manufacturing and food service fast food chains. They are interested in several scenarios like A/R prediction and data quality services. CIO is very innovative and regards predictive analytics as really ground breaking for his business. They were awarded most innovative manufacturing solution award last year by Microsoft. However currently their attention on this area is patchy. If we build something, they will use and will be very happy to get going quickly. Private. US$1B. IL, US.

Hopefully this provides you a flavor of experimentation we are doing currently. If you have a similar need to find patterns in your data and inform your business process more intelligently then please contact us.

Comments

  • Anonymous
    September 25, 2016
    This is really fantastic ! I hope to see more businesses accepting ML Solutions in their procurement process. Excited as what more new Predictive apps or solutions can be built by partners in Dynamics AX with Cortana Analytics.