<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Anjian Wu on Medium]]></title>
        <description><![CDATA[Stories by Anjian Wu on Medium]]></description>
        <link>https://medium.com/@anjianwu?source=rss-1beca1ea53e1------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*AX4DC6VCk37pxE1kJxzVBQ.jpeg</url>
            <title>Stories by Anjian Wu on Medium</title>
            <link>https://medium.com/@anjianwu?source=rss-1beca1ea53e1------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 10:32:30 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@anjianwu/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Does the Amazon and Whole Foods merger makes sense? A Consumer Cost Optimization Perspective]]></title>
            <link>https://medium.com/@anjianwu/does-the-amazon-and-whole-foods-merger-makes-sense-a-consumer-cost-optimization-perspective-9461b0e6e9b0?source=rss-1beca1ea53e1------2</link>
            <guid isPermaLink="false">https://medium.com/p/9461b0e6e9b0</guid>
            <category><![CDATA[food]]></category>
            <category><![CDATA[amazon]]></category>
            <category><![CDATA[whole-foods]]></category>
            <category><![CDATA[tech]]></category>
            <category><![CDATA[business]]></category>
            <dc:creator><![CDATA[Anjian Wu]]></dc:creator>
            <pubDate>Wed, 03 Jan 2018 20:42:18 GMT</pubDate>
            <atom:updated>2018-01-03T20:42:18.323Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/640/1*T4Se41enDEwy22bIBbqYgQ.png" /></figure><p>In August of 2017, Amazon announced their acquisition of Whole Foods Market for 13.7 Billion dollars. This bold move by Amazon into the brick-and-mortar grocery industry surprised many people and sparked much debate to the merits of such a move.</p><blockquote>We found that both could be well-suited partners in the battle against warehouse club chains such as Costco and Walmart.</blockquote><p>Coincidentally just a few days prior to this public announcement, my team was wrapping up a two-week final class project, where we developed a simple model to calculate the optimal grocery shopping strategy to minimize cost given a customer’s location, cost of transportation, food pricing, and family size. The objective of the project was not necessarily to build and validate the most accurate or complex model, but rather to use <strong>optimization to gain actionable business insights.</strong></p><p>Our findings showed that (at least around the Cambridge/Boston Area) there are signs that AmazonFresh would not necessarily be directly competing against brick-and-mortar store like Whole Foods. <strong>Instead, we found that both could be well-suited partners in the battle against warehouse club chains such as Costco and Walmart.</strong></p><h3>What is Linear Programming?</h3><p>As a student at MIT’s <a href="https://lgo.mit.edu/">Leaders for Global Operations</a> program, I had the pleasure of taking a class at MIT Sloan called 15.066: System Optimization and Analysis for Operations this past summer. Some tools we were introduced to was <a href="https://en.wikipedia.org/wiki/Linear_programming">linear</a>, <a href="https://en.wikipedia.org/wiki/Integer_programming">integer</a>, and <a href="https://en.wikipedia.org/wiki/Nonlinear_programming">nonlinear</a> programming applied to optimization problems in operations and manufacturing. Essentially, these are techniques where you solve for the “best outcome” by:</p><ol><li>Determining what <strong>variables</strong> exist in your business problem (i.e. hours of operation per day, price of a good, or number of people to hire)</li><li>Thinking through what are the <strong>constraints</strong> of those variables (i.e hours must be between 0–24 hours, prices cannot be negative, or we can only afford to pay X people)</li><li>Outlining a single <strong>objective function </strong>of the variables above (i.e profits of the business due to said hours of operation, price of good, and number of people hired)</li><li>Finally, either <strong>maximize</strong> or <strong>minimize</strong> the <strong>objective function </strong>by tweaking the <strong>variables</strong> under the outlined <strong>constraints</strong>.</li></ol><p>If the objective function and constraints are all linear relationships of the variables, then we have a linear programming problem. Otherwise, it is a non-linear programming problem. If the problem is linear and the variables are required to be integer, then it is considered <strong>integer linear programming (or just integer programming).</strong></p><h3>Problem Formulation</h3><p>The following assumptions were made in our model:</p><ol><li><strong>Store Selection</strong>: Only 5 store chains were selected for analysis: Costco, Whole Foods, Star Market, Market Basket and Amazon Fresh. These five were chosen because they represent a wide variety of store types within the Cambridge, MA area: wholesale, premium, mid-budget, budget, and online, respectively.</li><li><strong>Consumer Diets</strong>: We constructed simplified weekly diets based on guidelines and examples published by the US Federal Drug Administration. We then selected 16 food items that normally exist within those diets, and scaled the values according to size of household.</li><li><strong>Item Pricing</strong>: We did in person data collection of food prices at each grocery store. Each store has a variety of items to choose from, so we tried our best to choose the lowest price and keep items consistent across stores. We also assumed that pricing was consistent across all stores in the area from the same chain. Lastly, we also restricted the model from choosing fractions of a package (e.g. can’t buy 3⁄4 of a bag of spinach).</li><li><strong>Transportation Costs</strong>: We model transportation cost by the pricing structure of taking an Uber to the store. If the store was less than 0.5 miles away or if using AmazonFresh, we negated any transportation costs.</li><li><strong>Maximum Number of Stores</strong>: A household could shop at up to 2 different stores every two weeks. We also assumed that the shopper would have to return home after visiting the first store to drop off his or her groceries.</li><li><strong>Other Assumptions</strong>: The extra food that was purchased each cycle has a salvage value of zero (i.e. it is thrown away). Finally, for AmazonFresh and Costco that charge membership dues, we normalized the cost to the length of the shopping cycle (we assumed every two weeks).</li></ol><p>The store considered are shown below:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*5xZFRCdQ2hi1AZyi-RyafA.png" /><figcaption>The area of interest in our analysis (Cambridge, MA). In this case, we assumed the household is located at MIT and the dotted lines show the nearest stores of each chain (Market Basket, Star Market, etc…) and distance from the household. We can then pick ANY location on the map as the “home” and recalculate the nearest stores.</figcaption></figure><h4>Objective Function</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8Sxcg8i-XPxx_6i7Otetqg.png" /><figcaption>The objective function -&gt; minimize cost to needed to get all the groceries you need!</figcaption></figure><p>The objective function contains two summations. The first summation is the total cost of buying X_ij number (an integer value since you cannot buy fractions of an item) of the ith grocery item (16 grocery items) at the jth store (5 chains). The second summation is adding in all the membership (M_j) and transportation costs (T_j) associated with going to the jth store. The S_j is a dummy binary switch variable that is either 0 or 1 (namely if you do buy from the jth store it means S_j = 1 to reflect the required membership fee and transportation cost). <strong>The variables of the objective function are just X_ij and S_j (in other words how many items do you buy from this chain).</strong></p><p>For example, lets say that eggs is i = 5 and Costco is j = 2. If a household requires at LEAST 2 dozen eggs from Costco and Costco only sells eggs in packs of 4 dozens, then X_52 = 1 and S_2 = 1.</p><p>The constraints were then placed around the variables limiting the maximum number of stores, requiring each grocery items to be purchased above a specified amount, etc.</p><h4>Problem Solver and Simulation</h4><p>The model was first prototyped using the <a href="https://opensolver.org/">Microsoft Excel OpenSolver</a> add-in software. We then ported the model to Python using an Optimization Library called <a href="https://pythonhosted.org/PuLP/">PuPL</a>, where we could quickly simulate different optimal solutions across the map, effectively creating a heat-map showing changing competitive pricing advantages of each grocery chain.</p><h3>Findings</h3><p>Some of our most interesting results is shown below. We rerun the optimization problem all over the map with different family household sizes (N), recording the primary and secondary grocery chains used in the optimal price-saving solution.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ierVRqokbMH1ACVxCgk2BQ.png" /><figcaption>Heat map of the dominant grocery chains as a result of the cost-minimization. The top row shows the primary grocery store used (which is dominated by Market Basket across all household sizes). The bottom row shows the secondary grocery store used (which is dominated by Market Basket across all household sizes).</figcaption></figure><p>Not surprisingly, discount chain Market Basket was the dominant chain across all of Cambridge (<a href="https://www.boston.com/news/business/2014/07/28/how-market-basket-keeps-prices-low">how Market Basket keeps prices low</a>). What is more interesting is the secondary stores used, which changes based on region and household size.</p><ul><li>The store closest to the household tend to have a competitive advantage</li><li>As a household gets larger, the region of Costco (wholesale pricing) increases because it becomes more economically viable for families to drive farther away to Costco and save money from buying large wholesale units.</li><li>Premium grocery chains (Whole Foods) quickly lose their price advantage for any household greater than one.</li><li>AmazonFresh, shows price competitiveness for households greater than one in areas away from Costco. Star Market (mid tier grocery chain) also holds up against the Costco domination.</li></ul><h3>Take Aways</h3><p>In some ways, these results validate what many people already know to be true. Of course a large household would probably be better served to drive out to a Costco vs filling up their grocery cart from Whole Foods. It is intuitive that, assuming you live far away from a brick-and-mortar grocery chain, getting your groceries delivery by AmazonFresh is a better choice than to drive an hour away. Seeing these trends reflected upon the model provides a powerful tool by which we can visualize the competitive pricing dynamics and freely experiment with any series of what-ifs.</p><p>So what lessons might Amazon take away from our simple experiment? Well for one, WholeFoods would certainly benefit from lowering their prices and become more like Star Market so that they can create some price competitiveness in the proximity of wholesale chains like Costco. Secondly, AmazonFresh is currently already in a great position to create pricing advantage in households farther away from a grocery store. Perhaps, they should consider doing more targeted advertising, reduce AmazonFresh subscription fees, or aggressively bundling Amazon Prime with AmazonFresh and WholeFoods price matching.</p><p>Hopefully you found this an interesting read and that I have piqued your interest in linear programming. Feedback is always welcome and please reach me <a href="https://wuanjian.wixsite.com/aboutme">here</a>.</p><p>I would also like to acknowledge my teammates in the project: Andy Freiheit, Ken Hampshire, Ashton Imlay, Tom Mangan, and Shalini Singh.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9461b0e6e9b0" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>