Were there really 132 Artificial Intelligence companies at NRF?

NRF AI.png

Were there really 132 Artificial Intelligence companies at NRF?

Spotting the next generation of artificial intelligence


Artificial intelligence and machine learning were a huge buzz at NRF, and it seemed like every other booth had some mention of it. The NRF program listed 132 exhibitors as “AI” providers in their guide. Are AI and machine learning truly that pervasive and are the much-hyped benefits real?

What are Artificial Intelligence and Machine Learning?

Artificial intelligence is most commonly defined as “human intelligence demonstrated by machines”. Yeah, that means just about anything that makes a decision with a computer is AI. While artificial intelligence and machine learning are used interchangeably in common parlance, machine learning is an approach to artificial intelligence. Its definition is also super broad with the most common one being a “system that can learn from experience to find patterns in a set of data.” (1) That includes methods like ages old linear regression. So, it is true that a lot of vendors offer AI and ML, and we’ve seen it for years.

Next generation AI and ML

If we’ve had these things for years, why is it a big deal now? Is it just some marketing ploy by software vendors? There is a very real next generation of capabilities. With the dramatic increase in available computing power, the huge increase in data, and the development of next generation analytical techniques, AI and ML can solve problems it never could before. And, with the digitization of the customer journey, there are entirely new ways to create value for customers from personalized promotions to cross-selling. McKinsey estimates that AI will produce over $600 billion of annual benefit in retail, and results like Adore Me seeing a 15% increase in revenue generated from its promotional campaigns after applying AI validate the potential.

How can you tell old AI from new AI?

So, how can you distinguish next generation AI capabilities from older techniques? Here’s a guide to spotting the differences -

Data volumes

Next generation AI can process huge amounts of data and may make recommendations at microscopic levels. By looking at numerous examples, it can determine the real drivers of a result and then make more accurate forward predictions. For example, older techniques may look at history summarized by location to determine a promotion’s effectiveness. When looking in total, the techniques might reach a conclusion that a type of promotion performs better in suburban areas than urban areas when the promotion resonates with working moms. However, that predictions would be off in urban areas with significant # of working moms. By looking at detailed, customer level responses, the response rate difference by customer type can be spotted, and the prediction could be much more accurate. Not only that, it enables more targeted actions by a retailer like only delivering promotions to customers who would respond and would produce incremental business value.

Data types

Next generation approaches aren’t limited to just simple data sources like sales history. They can take images, written text, customer clickstream data, voice, etc.. and can sift through them to uncover insights. For example, it’s possible to process product images and find commonalities between different products without needing to create an attribute and manually assigning values for each and every product. This can even be used to find common products at other retailers. Similarly, customer reviews can be processed to find the qualities of the product that stand out or could stand to be improved. This leads to a fuller understanding of the product, competitive environment, etc...

The ability to process new data types also enables solving challenges in new ways. The much-hyped Amazon Go cashierless store is a great example of this, but it’s not alone. Walmart is deploying robots to spot empty shelves, and Chick-fil-a is using visual recognition to ensure their food is fresh.

Patterns detected

Next generation AI can find complicated patterns that predict behavior. In the past, AI techniques would use fixed patterns like level, seasonality, and trend and make projections off those. However, we know certain fabric weights only sell well at specific temperatures. Instead of requiring humans to manually assign seasonal profiles that capture these behaviors, next generation techniques can find these differences on their own. And, the patterns it finds aren’t limited to the obvious; Walmart found customers prefer eating berries on days that aren’t windy. Who would have known?


Next generation AI systems don't just try to make the most out of what was done in the past. They look at the past, spot potential actions where they have limited experience, run tests to improve its understanding of how it will perform, and then can automatically decide to stop performing the action or expand it because it performed well. In essence, these techniques can automatically perform the limited tests that were manually executed in the past. B2W, the largest online retailer in Latin America, tested setting their prices using these approaches and were able to increase their gross margin by 30% in one of their product categories. These techniques are also being used in digital marketing to personalize emails and evolve website designs.

What can you do with next generation AI?

The ability to understand the customer (via processing of huge amounts of data and direct interaction with them), competitive environment (via text and image processing of websites), and back-office task (via digitization of transportation, invoicing, and other processes) coupled with advanced methods to find the hidden patterns has opened up a whole new world of opportunity. You can use these AI techniques to improve the effectiveness of your promotions, better allocate product to stores, improve your labor planning. And, the cases aren’t just limited to customer-facing decisions. You can use it to better predict arrival times for shipments, delays at your warehouses, if employees will be absent, or what orders and receipts an invoice matches to. Or, you can make your decision processes more efficient by automatically identifying causes of issues like low stock levels or long wait times in store.


While there are a ton of AI providers, there are a limited number applying next generation capabilities. By finding these next generation software and analytics service providers and the right use cases that are important to your business, you’ll be able to apply AI to truly be more relevant to your customers, more efficient in your operations, and make better business decisions.

About Us

Cognira is a boutique consulting and software analytics company that focuses exclusively on retail merchandising and supply chain. Cognira’s goal is to make science easy for retailers, and to help them get the most value from their advanced science solutions. We bring decades of experience leveraging science, analytics and scalable technology to improve retail decisions.


1) If you’re super interested in the definitions, check out this awesome machine learning intro.

Supply Chain Brief



Inventory forecasting and replenishment in an omni-channel world

Omnni Inventory.png

Inventory forecasting and replenishment in an omni-channel world


Omni-channel retailing has changed the entire nature of inventory demand. The ‘demand’ fulfilled from a location adjusts based on what stock is available in the network. New delivery options and supply methods are constantly being added which makes the previous inventory positioning strategy less than ideal. Customers select from the available delivery and inventory options, but, their final selection may not reflect their first choice, ie. their true ‘demand’. These have all contributed to making inventory management a real challenge, one that over 50% of retailer’s report was their top supply chain business challenge in RSR’s Supply Chain Management Report. So, how can retailers adapt their inventory management processes for this new world?

Aligning on the basics

Like most new things, definitions and terms are used inconsistently between and even within organizations. So, to be sure we’re talking about the same things, here are our definitions -


  • Selling Location = the location where the customer purchased the item

  • Fulfillment Location = the location that supplied the inventory to support the sale

  • Delivery Location = the location where the customer took possession of the item


  • Demand Date = the date the customer decided to purchase the item

  • Sale Date = the date when the fulfillment of the item occurred, and the sales financial transaction was recorded

What’s wrong today

Today, many retailers forecast at sku, location, week/day based upon the fulfillment location and the sales date. This approach ends up repeating the same mistakes of the past. Say a customer purchased an item online and the fulfillment center closest to them was out of stock, so the item was fulfilled from an alternate location. When looking to determine where to place stock in the future, you’d want to place that stock at the closer fulfillment center vs. the less cost effective location, but forecasting based upon the actual fulfillment location for an order will result in it being in the wrong spot. And, the approach becomes even less appropriate as a retailer’s logistic networks evolve, for example, changes like stocking/owning the item instead of drop shipping it from a vendor, or the addition of temporary capacity to support peak periods. Additionally, customers have options to pick up orders and may do that a week after they made the reservation or may even back-order a product that isn’t available. If the forecasting is based upon the sale date, inventory may not be available at the right time to make that commitment next time or better serve customers.

Additionally, returns are frequently not accounted for the inventory plan, and, if they are, they are often forecasted as a % of the demand at that location. With online returns rates often over 30% and the typical time between the customer returning the product and it being available to sell being over 2 weeks, not accounting for the missing inventory in your plan can cause availability issues and excess inventory at the end of life. Additionally, with over 50% of customers preferring to return products in store (1), even if purchased online, forecasting returns only based upon stores sales misses a huge portion of returns and completely misses returns that show up at stores that didn’t stock the product.

Finally, the demand for product at a location is no longer just dependent on what happens at that location. If the customer checks online and sees the product isn’t in stock at their typical store, they may go to another nearby store. This new behavior presents new opportunities to change where stock is located so that you can serve all your potential demand, and it also presents great challenges in determining what customers really want.

Steps on the journey

Here are some proven steps you can take to adjust your inventory management processes for the new world of omni-channel fulfillment and that will improve your customer service, lower your supply chain costs, and increase your inventory productivity.

Step 1 - Capture a complete set of sales and available delivery information

It’s important to put in place the right foundation to understand customers’ desires. A key part of that is being sure you retain transactional data in a spot that is accessible and can be quickly enriched and summarized. The transaction log should capture where the customer purchased the item (i.e. the selling location), when the order was placed, the location that fulfilled the order, the date the order was fulfilled, and the delivery location and requested delivery service. This will then unlock the ability to analyze the data in light of your current and future supply network, better understand customer behavior, and so much more.

Additionally, it’s important to track what delivery options were available and intended so you can develop a true understanding of demand (i.e. could the product be purchased in store, delivered the same day, delivered in two days). This information will enable determining how much demand was lost when a store didn’t have enough stock to satisfy an in store pickup. Or it can be used to identify non-representative sales, such as a store selling a product that was returned there but wasn’t normally stocked there.

Step 2 - Regularly restate your sales history to be at the preferred fulfillment location

Re-stating your sales history based upon your current fulfillment network will enable you to make the most of your existing forecasting and replenishment systems. First, you should double check that the sales date for forecasting is based upon the customer demand date, not the financial sales date. This will ensure you have stock at the right time going forward. Then, a process should be developed to look at all your transactions and map them to the ideal fulfillment location based upon the current network design. Stores sales are the easiest; you just map the demand to that same location. For online sales the delivery zip and service (i.e. 2 day delivery, ground) should be used to find the ideal fulfillment location that stocks the desired product. Since the ideal fulfillment location can change regularly due to changes in network design, fulfillment centers’ stocking assortment , etc., it’s important that the process is able to regularly run and update history.

Additionally, your ‘out-of-stock’ detection logic for your online fulfillment locations may need to be tweaked. The typical approach of looking at historical inventory levels at that location to determine if sales were representative won’t work since sales will be made if inventory is available anywhere in the network. This can easily be adjusted to check if stock was available at any location in the fulfillment network.

With these changes in place, you’ll be able to prevent repeating the same mistakes and be able to place stock in the right spot in your ever changing supply network.

Step 3 - Forecast returns based upon the delivery location and account for them in your inventory plans

The first step in creating a returns forecasting process is moving your ‘demand’ forecast from ‘net’ sales to ‘gross’ sales. Then, it’s time to create the actual returns forecast by each location. Each store’s forecast should utilize the sales that occurred there and the online sales in the area around the store. A simple starter approach (2) is to calculate returns as a function of the % of gross sales from previous weeks. Then, the online returns can be allocated to stores and fulfillment centers by calculating the historical % of online returns received at that location. Finally, these can be combined to calculate the forecasted returns at each location in your supply network.

Then, the returns forecast should then be incorporated into the allocation & replenishment systems based upon how the product will be handled. If the product returned to the store is sold there, the forecasted returns can be treated almost like a receipt at that location. If products returned to a non-stocking store are shipped to a central fulfillment center, that flow should also be represented.

Step 4 - Forecast by delivery location and then determine the fulfillment location in your replenishment solution

When the most preferred fulfillment location for an order runs out of stock, the distributed order management (DOM) solution will start allocating stock from less preferred location to capture all possible sales. That’s great, but it creates ‘extra demand’ at the non ideal fulfillment locations, spillover demand. If that spillover demand is not accounted for, too little inventory will be ordered and your network will be perpetually short of inventory. This can even trigger vicious cycles of out-of-stocks among fulfillment locations; an effect that a recent paper from MIT labeled whiplash.

Example of Whiplash

FC1 starts with plenty of inventory to serve the ideal demand from the location; however, FC2 is short of inventory and so demand spills over to FC1. FC1s ordering process didn’t account for that excess demand, so it doesn’t order enough inventory to meet the demand. That then results in FC1s demand being spilled over to FC2.

FC1 starts with plenty of inventory to serve the ideal demand from the location; however, FC2 is short of inventory and so demand spills over to FC1. FC1s ordering process didn’t account for that excess demand, so it doesn’t order enough inventory to meet the demand. That then results in FC1s demand being spilled over to FC2.

However, it is possible to anticipate spillover demand and account for it in the inventory planning process. The first step is to start forecasting online sales by service (i.e. ground, 2 day) and delivery location. That delivery location can be the carrier’s shipment zone or the zip-3 for the delivery location; this gives enough granularity to understand shipment costs while not being too sparse to forecast. Then the replenishment logic should be adjusted to take that forecast and determine where it will be served based upon the projected inventory availability. This will enable better allocation of stock to the most cost-effective fulfillment locations and ensure that sales aren’t lost.

Step 5 - Adjust your inventory positioning so you can take advantage of the flexibility in your fulfillment capabilities and account for the uncertainty of demand

Moving a unit to store no longer means it’s stuck there since you can now use it to serve online orders as well as demand that walks in the store. This flexibility opens up all new inventory stocking strategies. For example, it may make sense to push more product to store since it increases the chance of selling to a customer who walks and is unlikely to purchase if it’s not there while not impacting the ability to use it to support online sales. Your allocation and replenishment capabilities can be expanded to take advantage of these new possibilities. The solutions should evaluate how the inventory decisions perform across all the ways that demand may occur and then choose the result that delivers the best financial outcomes. This will lead to increased sales and efficiency in your supply chain by letting you truly take advantage of the flexibility that omni-channel fulfillment provides.

Step 6 - Develop a true picture of demand by accounting for trade-offs customers make

With omni-channel retailing, customers are presented with the product and delivery options a retailer can provide at that moment. However, if other options were available, the customer’s selection might be different. To account for this, a true demand model with demand transference impacts (i.e. incremental sales and cannibalization) similar to what is utilized in assortment optimization to understand trade-offs between different products is needed. Additionally, the allocation and replenishment models need to be tied closely to the demand model, so they can understand the impact that replenishment decisions have on demand. If you can tie all this together, you’ll have a more predictable supply chain and a great model to understand other merchandising and supply chain decisions.

What do you think?

While we’ve laid out our view on the journey, we know there are multiple ways of getting to the same place. It would be great to hear your feedback and recommendations. . Here are a few questions that come to mind -

  • What retailers are the most mature in this omni-channel inventory management, and, what are they doing?

  • How frequently are delivery flow paths (i.e. stops on the path to the customer) being adjusted? How frequently should they be adjusted?

  • What steps are need to adjust inventory management processes to account for new vendor or supplier capabilities?

  • How likely is it that retailers will start to flex their supply networks at peak capacity (i.e. utilize more 3PLs during Christmas)?

  • What other steps can be taken to improve existing systems that work at sku, store, week?

  • What novel approaches have you seen to determine the best place to position inventory? Any probabilistic methods such as Markov chains? Any machine learning methods? Any reinforcement learning based methods?

  • How big is the impact of customers’ switching their delivery preference based upon the available options? Should this be a concern or is it too small to worry about?

  • What do you see as the next step on your journey?

About Us

Cognira is a boutique consulting and software analytics company that focuses exclusively on retail merchandising and supply chain. Cognira’s goal is to make science easy for retailers, and to help them get the most value from their advanced science solutions. We bring decades of experience leveraging science, analytics and scalable technology to improve retail decisions.


  1. Per a study by Navar

  2. That starter forecast can be enhanced be enhanced a number of ways, but two important enhancements are to account for impacts of holidays, which increase the return rate and lengthen the return time, and to better identify products with higher than typical return rates.

Supply Chain Brief


Efficient Customer Order Routing

DOM Efficent.png

Efficient Customer Order Routing

Moving past short-sighted customer ordering routing decisions


Efficient and reliable routing of customer orders is critical in today’s retailing environment. Even in one of the largest and most efficient supply networks, Amazon, shipment costs are 14% of retail sales. That’s a big number, and for most retailers who are not at the scale of Amazon, those costs are even higher. With 42% of retailers saying faster shipping is a top priority, these cost are set to go even higher. As part of our series on the cutting edge of omni-channel fulfillment, we’ll take a look at how changing from a myopic view to a more complete picture can increase the efficiency of order routing decisions.

Current State of the World

While there is great variability in order routing techniques, most boil down to applying a set of rules to determine the inventory eligible to use and then selecting the sourcing location with either the lowest shipment cost or shortest delivery time. These decisions are made one order at a time, but the challenge is we don’t just have that one order. We will have a whole set of orders coming and by ignoring these future orders, we’ll make decisions that could increase our cost and/or impact the quality of service for our most important customers.


Assume we have two fulfillment centers - one in Elizabethtown, PA and the other in Cedar Rapids, IA. The Elizabethtown DC only has 1 unit on hand while Cedar Rapids, IA has 10 units. An order for standard 4 day delivery comes in from a customer in Atlanta, GA. The delivery time from both DCs with UPS Ground is 2 days, but Elizabethtown is slightly less expensive at a cost of $10.26 vs. $10.80 at Cedar Rapids. So, the order is routed to the Elizabethtown DC.

US Map with Fullfil - Step 1.png

Then, another order comes in for the same product from a customer in Bangor, ME, but, she wants next day service. Since inventory is only available from Cedar Rapids, it’s shipped from there via Next Day Air from there at a cost of $65.76.

US Map with Fullfil - Step 2.png

However, with perfect foresight, the most cost effective strategy would have been to ship the first order from Cedar Rapids, and the next day delivery from Elizabethtown. The total cost to fulfill both orders, $40.55, would have been 45% less than the short sighted solution.

US Map with Fullfil - Step 3.png

The Cutting Edge

Of course it's not possible to know exactly where every future order will come from and the real life scenarios are much more complicated. However, it is possible to forecast the location of customer demand and incorporate the impact to the fulfillment costs into the decision of how to route an individual order. So, how can we do it?

First, a forecast of the demand by zip code and the customer’s desired service level (i.e. next day, 2 day, 4 day, whenever) is needed. While it sounds easy, it’s really hard to create a good forecast at that level of granularity, and, it often requires grouping zip codes together so that an accurate forecast can be created. Not only do we need the one number forecast, we also need the chances that we sell 0, 1, 2, or more units in each zip code, ie: the distribution of demand. This enables finding the decision that achieves your goal over all the possible demand outcomes, instead of assuming that a specific pattern is going to occur.

Next, the financial impact, including to future orders, of fulfilling a specific order from a specific location can be calculated utilizing linear programing techniques. Alright, that’s nerdy…But, it’s just a way to search through tons of possible answers and find the one that delivers the best result. One of the neat things of doing this is it can produce the expected value of having one more or less unit of inventory at a location. That’s the key value we need to understand the impact of selecting the order from that location!

Finally, these costs can be incorporated into your existing order routing rules in the DOM resulting in more efficient decisions. You’ll be able to quickly see the improvements!

As an added bonus, the forecast used by the DOM solution can also be used by the replenishment solution to ensure that inventory is located in the right place from the get go.

About Us

Cognira is a boutique consulting and software analytics company that focuses exclusively on retail merchandising and supply chain. Cognira’s goal is to make science easy for retailers, and to help them get the most value from their advanced science solutions. We bring decades of experience leveraging science, analytics and scalable technology to improve retail decisions.


This work is inspired by Jason Acimovic and Stephen C. Graves’s Making Better Fulfillment Decisions on the Fly in an Online Retail Environment.

Supply Chain Brief


What’s going on in the last mile?

What’s going on in the last mile?

A look at how Kroger, Target, and Home Depot are adjusting their supply chains to meet customers’ omni-channel fulfillment needs


Free two day delivery and grocery pickup are quickly becoming customers’ ‘standard’ delivery expectations. However, customers’ expectations won’t stop there as more and more retailers figure out better ways to meet customers’ desires for where I want it, exactly when I want it, and cheap delivery. So, how are retailers enabling themselves to provide the next generation of delivery options while still maintaining reasonable profitably? We’ll take a look at how Kroger, Target, and Home Depot are adjusting their supply networks to do it, and, we’d love to hear more from you on what you’re seeing.


While Kroger joined the omni-channel world later than many of the other players, they’ve made huge steps in the last 6 months to meet customers’ new delivery expectations. Today Kroger Pickup has expanded to the majority of their stores, and they’re rapidly expanding home delivery through their partnership with Instacart. They’ve also acquired Home Chef to enter the meal kit delivery service. So, how are they doing it today and what are their plans for the future?

Currently, all of Kroger’s picking and packing for digital orders is done in store, but they’re rapidly building out the capability to do that at dedicated fulfillment centers. They’ve entered a partnership with Ocado Group, the UK grocer and technology company, to build 20 highly automated fulfillment centers over the next 3 years. Rodney McMullen, Kroger CEO, described the overall strategy in their Q1 investor call as “what we're really trying to do is to make sure that we have an overall infrastructure for digital that can support whether it's 5% of share or 30% of share. If [digital] ends up being 30% of share in grocery..., you'll have more [Ocado] sheds and they'll be used to take pressure off of the store, and the store will become more of a distribution point.”

In addition to the investments in picking and packing, they’re working to add more convenient and efficient distribution options. Kroger has entered into a partnership with Walgreens that will enable customers to pick up their Kroger online grocery orders at their local Walgreens store.* And, in more of a moonshot, they’ve partnered with Nuro to pilot the use self-driving cars for same-day grocery deliveries.


Target is leading the pack on fast and free delivery by offering free 2-day shipping without a minimum purchase. They've also expanded their delivery options by adding ‘Drive Up’, their store pickup option, to over 50% of their stores and have expanded the availability of same day delivery through Shipt which they acquired the end of last year.

Unlike Kroger, Target is looking to make the store the center of their fulfillment and distribution network. At their 2018 financial community meeting, Target’s COO, John Mulligan, stated “[b]ecause our stores are the fastest and most efficient fulfillment method, they'll continue to be our preferred shipping point in the long run.” Mulligan also explained that the economics of store based fulfillment trump that of  fulfillment centers due to lower shipping and capital costs. They also see store based fulfillment as a huge enabler for serving peak demand because they don’t have the same physical capacity constraints as fulfillment centers and can easily scale up their volume.

To support the strategy of fulfilling digital orders from stores, Target is adapting how inventory flows into their stores. They are looking to stock stores with exactly what they need and then be able to quickly respond when things change. They’re moving from a typical forward replenishment inventory strategy that ships in packs or pallets to a strategy that replaces exactly what was sold in or fulfilled from a store. And they do mean replacing exactly what sold; they’re adding robotics and other material handling equipment in their store distribution centers to enable them to ship to stores in any quantity from eaches to pallets. Finally, they’re also increasing the frequency that they ship to stores which allows them to respond faster to unexpected changes. They report that their test in the Northeast has cut out of stocks in half and has lowered store backroom stock. They plan to use the newly freed backroom space to add sorting and packing stations to support fulfillment from store.

Home Depot

Home Depot has long offered a wide variety of delivery options from same day in-store pickup to delivery in the next few days to delivery within a specific time window. Most recently, they added same day delivery through their partnerships with Roadie and Deliv. While they offer a nearly complete set of delivery options, they’re focused on making each option more cost effective.

Home Depot describes their “downstream” supply network as fragmented. They service pro customers’ deliveries and all next-day and same-day orders out of their stores. They have specialized facilities dedicated to appliance delivery. They have facilities dedicated to serving their maintenance, repair and operations (MRO) business, which they acquired from Interline Brands. Finally, they have 3 fulfillment centers dedicated to consumers’ 2-day and ground home delivery orders.

With all this fragmentation, they see great opportunity to provide faster delivery at lower cost by consolidating, regionalizing, and specializing their “downstream” supply network. So, over the next 5 years, they plan to invest $1.2 billion and completely revamp their downstream network. Here’s some of the major changes they’ve announced -

  • Increasing the number of large fulfillment centers (moving from 3 to 7)

  • Adding 25 local fulfillment centers to support next day and same day delivery in their top 40 markets. These facilities will support both their consumer, pro, and MRO businesses.

  • Adding 40 flatbed fulfillment centers to support delivery of lumber, drywall, and other bulky goods.

  • Removing all fulfillment from stores in their top 40 markets.

  • Supporting same-day, next-day, and bulky fulfillment for their non top 40 markets from a limited # of “market delivery” stores.

  • Adding 100 cross dock/market delivery operations to support delivery of bulky product such as appliances and patio furniture

What do you think?

Three big players are plotting out very different strategies to meet the demands of customers’ omni-channel order fulfillment. Kroger is focused on centralizing and automating it’s network, Target on making stores the center of its delivery network, and Home Depot on regionalizing and consolidating its specialized network. We’d love to hear what you think about how the supply chain should evolve to support the next generation of fulfillment. A few questions that come to mind -

  • What’s the role of stores in supporting digital orders? Target is charting a very different course than Home Depot and Kroger. Can the economics at stores be better than at fulfillment centers? If so, what will drive that? Does a focus on fulfillment in stores distract from assisting customers?

  • What role does automation and robotics play in warehouses and fulfillment centers of the future? Will they be highly automated, high capital cost options, like Ocado, or more along the lines of traditional sortation systems?

  • Does it make sense to invest in more pickup locations like Kroger is doing with Walgreens? While the costs are significantly lower, this method is still less convenient than home delivery.

  • When will autonomous vehicles or other highly efficient home delivery options become feasible/practical?

  • How should smaller retailers, which don’t have the same scale, look to provide these same delivery options efficiently? Will 3PL’s step in to provide local delivery options?

  • What other changes should retailers be making to their supply network?

About Us

Cognira is a boutique consulting and software analytics company that focuses exclusively on retail merchandising and supply chain. Cognira’s goal is to make science easy for retailers, and to help them get the most value from their advanced science solutions. We bring decades of experience leveraging science, analytics and scalable technology to improve retail decisions.

* Be on the lookout for additional distribution announcements. Kroger’s announced that they want to serve all of America, and they’ve hinted that expanding grocery delivery into the Northeast, where they don’t have stores, would be it’s next step.

Supply Chain Brief


3 Ways Excess Data Can Stunt Your Growth

3 Ways Excess Data Can Stunt Your Growth

We love data. But we also know too much data can be a bad thing.  Here are three ways we’ve seen retailers using data to their detriment.

And some ideas on how to avoid these pitfalls.

1. Analyzing anecdotes

One data challenge we've seen is people manually reviewing too much data anecdotally.

Examining sales volumes for the last time you ran that promotion is valuable. But a single person can only consider a few data points at a time. And that data is open to interpretation, which is often biased.

At one retailer, team members in three functional areas, were each reviewing historical anecdotes, estimating the impact of the same up-coming promotions, and arriving at different conclusions.

Analyzing anecdotes uses valuable time. And bias can lead you in the wrong direction.

Strong analytics will curb reliance on anecdotes.

Your teams will always want to see sales history. Making it efficiently available brings a lot of value. To deliver even more value, provide demand insights based on proven modeling techniques. These will steer teams to better informed, less biased decisions. This investment in data analysis, paired with an explanation of insights will also curb reliance on so much historical data.

2. Divining too many details

Want to know whether a discount with a recipe sent to Jasmine’s mobile account will inspire her to buy that lemon-infused Tunisian olive oil when she walks into your Kansas City store the second week in July - if July 4th falls on a Wednesday – and it’s raining, but just a little bit?

While many systems can mechanically model customer demand at very, very specific levels of detail, taking into account nearly unlimited factors, often the data cannot support reliable results.

Systems with erratic recommendations are burdensome to manage. And these systems don’t get used, leading to less effective decision making.

A balanced approach will help ensure rich and reliable results.

Well-equipped implementers, data analysts and solution providers can balance detailed insights and dependable results in a variety of ways.

However, when one retailer reached out for help with their forecast challenges, they painfully admitted - “Our last implementor was very good – they did everything exactly as we asked them.”

Pushing for very specific approaches to demand modeling may pressure your implementor or data teams to make you happy in the short term. Instead, being clear about where you are looking for value, how you will leverage the forecast, and your bandwidth to manage exceptions will help produce detail-rich forecasts in which you and your teams can have confidence.

3. Bigger is better syndrome

With so much data available to retailers today, some can become eager to leverage more of it than is necessary to address today’s most pressing opportunities. And that data can be difficult and time-consuming to access.

Data-intensive analysis can leave other important, more easily addressed challenges unattended. Valuable insights are left unmined and unseen. Out of tune systems continue to misbehave, getting you into who knows what kind of trouble.

The right data for the job will help give you higher returns on your effort.

Petabytes of data and shiny, new tools can be enchanting – and in many cases, are worthwhile. But the most powerful and valuable approaches to leveraging data in retail are not all new. Being open to a variety of data sources and solutions can help you get the most value for your investment.

Can you think of other ways too much data can stop progress or bring harm to your business?


A little bit about Cognira

We’ve helped retailers get value out of big data and little data. Using the most modern cognitive and machine learning tools. And time-proven, robust statistical techniques.

Supply Chain Brief

What Your Inventory Can Tell You About Your Promotional Strategies

What Your Inventory Can Tell You About Your Promotional Strategies

Many retail challenges, like promotions, impact and require coordination across many areas of the business. We find it helpful to work across functional areas and bring different perspectives to the table. Below are some thoughts about promotions from the POV of an Inventory Manager.

Inventory Managers can see that there is no turning back time on the discount clock.  Promotions are totally ingrained into our customer’s buying behavior, even (or especially) for the most loyal of customers.  What do we see?

Customers will wait for promotions on products they already want to buy.

We’ve seen products where up to 90% of sell-through happens at discounted prices, products which consumers will not buy at regular price.  These items could be fashion items or consumable products that are on sale frequently. 

If customers know significant price drops are coming, they can wait to buy the fashion item.  And the consumable product has a shelf life that extends until the next promotion, why not stock up?   

Other circumstances we've seen are retailers who have one or two major sales per year, during which they do the majority of their business.  They sell expensive items at a good discount.


Why do Inventory Managers care?

As we condense our sales of our items or business into fewer time periods the volatility increases, and demand is more challenging to predict.  Achieving high service levels for our customers is more difficult and expensive. - It’s harder to move inventory through the supply chain quickly in response to sales. Sometimes, to chase demand, we have to use alternate sourcing, which increases costs and strains the supply chain.

Customers are given offers they cannot refuse.

Limited time offers and bundled offers can convince customers to buy. They may purchase items they don't really want or spend more than they planned. 

 Free shipping on orders over $50

$20 off when you buy $50

40% off marked items, today only!

40 percent off.jpeg

This is awesome, right?  From a customer standpoint, the risk is limited, because they can always take the items back, and we do get sales we might not have otherwise.

How does this impact Inventory Management?

It’s not unusual for 25% of online fashion sales to be returns and returns for these types of promotions are even worse.  

This impacts our stocking strategies. Sometimes we cannot track it or transfer it because the items aren't ranged to sell in that location.  We have to mark it down to sell it. Other times we have stores package it to fulfill online demand. 

It’s almost too easy to connect with Customers

The email you send out today impacts what customers buy tomorrow.  Anywhere from single person to all loyalty or potential customers could be targeted.  When customers share promotions on social media and third party sites publish them, the promotion audience can grow amazingly fast.


What burden does this place on Inventory Management?

We may see sales spike and discounts applied, without insight into the promotional types or vehicles.  If we don't know in advance what offers, we can't respond in time to get inventory in the best locations.  Even if when we do know a promotion is coming, the results are difficult to forecast, because we don't know important details.

What can retailers do?

These practices add a lot of uncertainty, causing us to lose sales or increase costs.  It’s a balancing act; we need to work together across functional areas to make the best decisions. 

We can:

Share data and create good analytics

If the data is not centralized, each group may only see a part of the puzzle.  Don’t assume everyone is looking at what you are.  Information is the first step.

Are we able to answer important questions with our data? 

  • Does pantry loading drive traffic or higher spend overall? Under what circumstances?

  • Are we running promotions on competing products at the same time?

  • What promotions are driving cherry pickers?

  • How much inventory is orphaned by returns coming into stores? What is the impact of these un assorted items on markdowns, store labor?

  • What is our promotional service level? How much inventory does it take to support different types of promotions?

Objectively assess our forecasting ability

Promotional forecasting is challenging, but some practices make sales even more difficult to predict:

  • Promotions that recur within a short time frame

  • Promotions with unpredictable audience, e.g. social media for example

  • Promotions that move majority of demand to short time frames

A review of promotional predictability could reveal more.

Reviewing promotions can help categorize the impact on the promoted item's sales, and the impact on other items and locations. Understanding the uncertainty associated with the promotions helps highlight promotional risk. 

These insights can inform stronger decisions upstream and downstream, encourage tighter collaboration.

Understand and agree upon stocking strategies

Imagine being able to have these discussions:

These items are frequently on promotion, have huge pantry loading effects. - 80% of sales are at a discount.  Do we want to continue these promotions? If so, what should our stocking strategy be?

Another group of products are expensive, highly price sensitive, and are rarely promoted.  We are confident in the total sales but to have high service level at the store, we will risk overstocks after the promotion.  We'll hold inventory back at the warehouse, replenish during the promotion and offer customers home delivery if we stock out.

Occasionally, let the tail wag the dog

Listening to your Inventory Managers and Supply Chain can help you see your promotions from a different perspective and may inspire you to tweak and align practices to improve your promotional performance.


About the author

Linda Whitaker has worked with hundreds of retailers applying data science and artificial intelligence to a wide variety of areas including merchandise planning, pricing, promotions, customer insights, demand forecasting, replenishment, allocation and supply chain. Prior to joining us at Cognira, Linda served as Chief Science Officer at Quantum Retail Technology (which she co-founded) and 8th Bridge (now part of Fluid, Inc.). 


Supply Chain Brief