Upload
others
View
4
Download
0
Embed Size (px)
Citation preview
Daisy Intelligence: Bringing artificial intelligence to
the retail end of the food value chain
T. Tucker
Abstract This paper will explore a common marketing / merchandising problem — namely which items from a
large selection of stock-keeping units (SKUs) to feature in a weekly grocery flyer. While traditional
heuristics and “gut feel” have sustained this business decision for a number of years, the availability of
software tools to leverage the vast quantities of years’ worth of Point-of-Sale (POS) data in order to make
a more holistic decision is explored. Artificial Intelligence (AI) “lets the data speak” and provides
retailers with promotional, pricing, and inventory information and insight into the impact of those product
selection decisions across the entirety of the business. Daisy Intelligence is a leader in this space, and is
the exemplar company for this case study.
For use by the Food and Agriculture Business Seminar, Arrell Food Institute of the University of Guelph, June 5-6, 2016
Daisy Intelligence 2
For use by the Food and Agriculture Business Seminar, Arrell Food Institute at the University of Guelph, June 5-6, 2017
Contact Arrell Food Institute
University of Guelph | Guelph, ON N1G 2W1 | www.arrellfoodinstitute.ca
[email protected] | 519-824-4120 Ext. 58782
Table of Contents Abstract ..........................................................................................................................................................1
Contact ...........................................................................................................................................................2
Introduction ....................................................................................................................................................3
Motivation & Problem Definition ..................................................................................................................3
Combinatorically explosive: ......................................................................................................................5
Big Data / Big Analytics ................................................................................................................................6
Artificial Intelligence .....................................................................................................................................9
Industry Intelligence ....................................................................................................................................11
The Customer Perspective ...........................................................................................................................12
The Future ................................................................................................................................................13
Conclusion ...................................................................................................................................................13
References ....................................................................................................................................................15
The author is grateful for the access provided by Daisy Intelligence and its customers to conduct
interviews for this case study. The case was solely funded by the Arrell Food Institute of the University of
Guelph, but the views expressed are those of the author and do not represent an official position of the
Arrell Food Institute or of the University of Guelph.
The research project has been approved by the University of Guelph Research Ethics Board under REB
approval #16MY011.
Daisy Intelligence 3
For use by the Food and Agriculture Business Seminar, Arrell Food Institute at the University of Guelph, June 5-6, 2017
Introduction The Ontario Agricultural College Review asked this rhetorical question in December of 1904:
Let’s fast forward a little more than a century to ask a new question: “Should leaders in the food and
agribusiness sectors study Artificial Intelligence and related topics like Big Data and Big Analytics?”
Much like the article from 1904, “conditions are changing rapidly” — with the advent of low cost / high
capacity data storage to manage large volumes of data that were unimaginable a few year ago to powerful,
scalable ‘cloud’ software-as-a-service (SaaS) computing platforms to host clever analytics and algorithms
to extract ever finer bits of information from a sea of data, it has never been easier to explore consumer
buying behaviour with such accuracy and resolution for the use of marketing decision making.
This study will provide the reader with an overview of the opportunities that Artificial Intelligence (AI)
brings to the food retail space and provide some insight into the capabilities of one such analytics
provider — Daisy Intelligence. This study was developed through publicly available documents and
interviews conducted with personnel from Daisy Intelligence (Daisy) as well as key Daisy customers.
This study does not intend to illustrate either effective or ineffective handling of managerial situations
involving AI or ‘big data / big analytics’, rather the intent is to educate and provide fodder for future
discussions related to this technology.
Motivation & Problem Definition Consider two recent flyer front pages from Canadian grocery retailers Walmart and Sobeys:
Daisy Intelligence 4
For use by the Food and Agriculture Business Seminar, Arrell Food Institute at the University of Guelph, June 5-6, 2017
Sobeys Flyer May 12-18, 2017
Walmart Flyer May 11-17, 2017
Both flyers cover the same period: a couple of days before Mother’s Day on Sunday the 14th
and a few
days in the following week leading up to the first long weekend of the summer season — Victoria Day on
the 22nd
. The Sobeys’ flyer is more strongly Mother’s Day themed with in-store floral arrangements and
hanging baskets displayed prominently above the fold. Walmart’s flyer includes the obligatory Mother’s
Day flowers too, but below the fold. The Victoria Day long weekend is widely acknowledged as the start
Daisy Intelligence 5
For use by the Food and Agriculture Business Seminar, Arrell Food Institute at the University of Guelph, June 5-6, 2017
of Bar-B-Que Season in Canada, and both flyers present BBQ protein options to the reader. Walmart’s
AAA Angus Beef is prominently displayed, while Sobeys offers a variety of BBQ options: beef, salmon,
and pork or chicken kebabs. Beverages to pick up for that long weekend BBQ? Walmart offers up Pepsi
and Coke, while Sobeys has bottled water, lemonade, and iced tea. Finally, for dessert, Walmart would
like you to savour some watermelon while Sobeys offers ice cream and novelties.
How does a marketing manager choose what items to promote in a weekly flyer? Which items should
appear on which page? Even front page items — which ones should go above the fold and which ones
below the fold? While the Walmart flyer is promoting grilling steak above the fold, will specials on
hamburger buns and ketchup below the fold produce a ripple effect and drive customers to pick up ground
beef or frozen hamburger patties at everyday low prices? This case will consider these kinds of decisions
in light of three operational-level questions that a marketing manager must ask themselves:
• What products should I have on my shelf, and what should I promote this week?
• What prices should I put on those products? Either as everyday prices, or discount prices, or mark-down prices? And,
• How much demand should I expect in each of my stores?
Combinatorically explosive:
Of the 50,000 or so stock keeping units (SKUs) in an average food store (Charlebois, 2017), choosing
twenty items (say) for the front page of the weekly flyer results in 3 905 035 049 527 015 951 007 636
445 078 293 739 902 178 475 650 795 921 972 386 778 352 845 310 000 (or 3.9 X 1075
or 3
quattuorvigintillion) possibilities (Wolfram, 2017). The number of possibilities can be reduced if heuristic
rules are used. For example, a rule of thumb like “the front page must feature one source of protein”
reduces the possibilities, but even this “simplification” still requires answering a large number of
questions: Which protein? At Thanksgiving time, the protein will likely be turkey, but what about BBQ
season? Packaged product or catch weight product? National brand name or in-house brand? Package
size(s)? Flavours or other variants? And so on, and so forth.
In the Walmart flyer, Coke is on special this week. More likely than not it was also promoted this time
last year in the pre-Victoria Day flyer of 2016. As Dean Pacey (VP Business Development at Daisy)
explains: “In answering the question ‘what should I promote this week?’ most retailers have a repetitive
nature of about 90%. Look at same calendar adjusted week from last year and 90% of the items will be
the same. How can businesses get out of being a creature of habit? Are the promotions in the flyers
actually resulting in an incremental change in our business or just bringing forward demand a couple of
weeks?” (Pacey, 2017).
Sylvain Charlebois — Dean of the Faculty of Management at Dalhousie University — has similar
thoughts: “To optimize any food retail store, a typical management team makes between 1,000 to 1,500
decisions on any given day. These decisions influence everything from merchandising to assortment
strategies. If you consider the joint effect of directions from head-office with transient customer data and
insights, instinctive decisions on the part of management just won’t work anymore. There are simply too
many calls to be made manually, especially in an environment where margins rarely exceed 2%. The
industry is only starting to acknowledge it simply cannot efficiently manage all the data points it has
Daisy Intelligence 6
For use by the Food and Agriculture Business Seminar, Arrell Food Institute at the University of Guelph, June 5-6, 2017
access to.” (Charlebois, 2017). As Pacey notes: “Daisy is helping clients break through some of the myths
and habits by letting their data speak” (emphasis added).
While some companies have made investments in predictive analytics software (so-called “query
machine”), this software approach requires trained data scientists on staff and it can only handle simple
queries. You can’t ask a query machine complicated questions like: “which products should I promote to
increase my total incremental revenue of my business” — you can only ask it simple, specific things, one
query at a time. For example, you could ask a query machine: “what price should I put Coke at to increase
sales of Coke?” A query machine will generate a price elasticity curve for Coke for that season for that
week based on historical data — but it couldn’t give you the impact of the Coke price change on ALL of
the other items in the pop category. It is that sort of insight — missing from a query machine — that a
manager needs to know in order to determine if the price change on Coke is actually netting anything
from the market or just stealing revenue from other categories like store-branded pop or simply bringing
forward demand for people who stock up on Coke at discount prices.
Pacey gave another example that is simple to understand but computationally complex: “Consider this. If
I put hamburger on sale, I know I’ll sell more hamburger. But I’ll probably sell less hot dogs. Which
means I should probably forecast down hot dog buns. So — query machine tell me — if I drop by
hamburger price by 10%, how much less hot dog buns and relish should I order next week? Those old
predictive machines would go ‘sorry; you’re getting way to complicated.’ This is a massive computation
to be able to do — multiple dimension product integrations. You’re talking about cross-price elasticity,
cross-promotional elasticity, affinity or halo relationships across multiple products. I made it sound
simple, but if you think of Walmart with 168,000 products, which means every product has a ripple effect
across 10- or 20- thousand products each week.”
This seemingly simple example of the impact of hamburger sales on hot dog buns is further complicated
by supply chain considerations. If we just consider hot dogs, buns, and relish, all three products come
from different suppliers. All three are perishable, and each has a different shelf life. A decrease in the
price of hamburger may necessitate a decrease in the size of the order for hot dog buns, and while that
may impact demand for relish, because of its longer shelf life, it may be fine to ignore the impact on that
product. Continue this line of reasoning across other product combinations, and again, the
computationally explosive nature of making a decision and understanding all impacts across other
products is apparent.
Where query machines require a data scientist to generate queries on the data, AI machines “learn” from
the data and improve their outcomes over time. As Pacey notes, “Daisy works in a world of massive
amounts of data — transactional information, client information, social demographic information, all of
the variables that are constantly and dynamically moving like prices and completive pricing, cost, supply
and so on and translate that into better decisions that our clients can execute on to increase the
profitability of the business.” So — what is big data, big analytics, and artificial intelligence?
Big Data / Big Analytics Even if you haven’t thought about it directly, chances are you’ve experienced the impact of big data / big
analytics already. For example, if you did a search for the book “Data Science for Business” at the
Daisy Intelligence 7
For use by the Food and Agriculture Business Seminar, Arrell Food Institute at the University of Guelph, June 5-6, 2017
Amazon.ca website, below the information related to the title, you would see two sections created through
the analysis of Amazon customers buying data: Frequently Bought Together and Customers Who Bought
This Item Also Bought…
When Apple’s iTunes “Genius” suggests songs based on your existing music catalogue or Netflix gives
you movie recommendations based on what you’ve watched in the past or your credit card company calls
you to verify the use of your card because it falls outside expected patterns—these are examples of big
data in action.
What is meant by the term “big data”?
While there is no singular definition of “big data”, there is a general sense of convergence on the many
ideas wrapped up in the term.
• Google defines big data as: extremely large data sets that may be analyzed computationally to
reveal patterns, trends, and associations, especially relating to human behavior and interactions
(Google, 2017)
• Service provider SAS describes it as: large volume of data – both structured and unstructured –
that inundates a business on a day-to-day basis. Big data can be analyzed for insights that lead to
better decisions and strategic business moves (SAS, 2017)
• Consultants at Gartner Group say Big Data: is high-volume, high-velocity and/or high-variety
information assets that demand cost-effective, innovative forms of information processing that
enable enhanced insight, decision making, and process automation (Gartner, 2017).
Daisy Intelligence 8
For use by the Food and Agriculture Business Seminar, Arrell Food Institute at the University of Guelph, June 5-6, 2017
In IBM’s eBook “Understanding Big Data” they provide this figure to illustrate what characteristics big
data has — namely the “three Vs” of volume, velocity, and variety:
Figure 1-1 from IBM’s Understanding Big Data eBook (Zikopoulos, et al 2012, p. 5)
Volume of data is usually what most people associate with big data when they hear the term. We’re
familiar with a byte (a single character or digit), and groupings of bytes into kilobytes (1,024 bytes = 1
KB), megabytes (1,024 KB = 1 MB) — a typical JPG photo is in the range of 1-4 MB. Gigabytes (1,024
MB = 1 GB) are typically the capacity we work in — a DVD movie is about 4GB in size; SD cards for
digital cameras are in the 16-32 GB range. A terabyte (TB) of storage (1,024 GB = 1 TB) is fairly
common now on a laptop computer. A 2012 article from the Library of Congress in the United States
suggested that all of the data contained in all of the books in the Library of Congress amounted to around
10 TB of data (Johnston, 2012). The diagram from IBM talks about “zettabytes” (ZB) of data. A zettabyte
is 1021
bytes of data (1 000 000 000 000 000 000 000 bytes) — one billion TB of data or one trillion GB
of data. No matter how you slice it, that is a lot of data to deal with!
Variety of data is another key consideration. As noted on the diagram above, data is moving from highly
structured (transactional data stored in a traditional relational database system) to include unstructured
sources of data (Schmarzo 2016, p. 59) — things like customer comments, tweets on social media,
photographs and videos would be considered unstructured. Add to the mix semi-structured data — a
database from a customer call centre (for example) would contain structured data (customer ID #,
Daisy Intelligence 9
For use by the Food and Agriculture Business Seminar, Arrell Food Institute at the University of Guelph, June 5-6, 2017
telephone number, name, etc.) but unstructured data as well (a description of the call, a transcript or
recording of the conversation, photos or files to identify the problem, etc.) (Minelli, et al 2013, p. 10).
Finally, the IBM approach depicted above includes Velocity of data as the third “V”. As Big Data, Big
Analytics describes: “Data velocity is about the speed at which data is created, accumulated, ingested, and
processed. The increasing pace of the world has put demands on businesses to process information in
real-time or with near real-time responses. This may mean that data is processed on the fly or while
“streaming” by to make quick, real-time decisions or it may be that monthly batch processes are run
interday to produce more timely decisions. (Minelli, et al 2013, p. 10)” Again, think about your own
experience. There was a time when banks would run daily batch jobs to update customer account
statements—now balance updates are instantaneous: they are printed on your receipt when you withdraw
cash from an ATM. On the process control side, velocity of data is key to some business processes —
witness real time updates from couriers and delivery companies about the status of a package or parcel.
Some vendors add other V’s to the mix. IBM for example, includes Veracity — ensuring that the data
used for analysis is indeed accurate and correct (IBM, 2017c). You may have a great data warehouse and
powerful reporting tools, but as the old adage goes: “GIGO — Garbage In, Garbage Out”. SAS asks its
customers to consider Variability — for example, does data flow in at roughly the same rate, or as is
often the case in social media, a topic or issue starts “trending” and the volume of data increases
dramatically. SAS refers to this as “event triggered” variability to distinguish it from daily or seasonal
variability (SAS, 2017). Finally, IBM adds Value to the list of Vs. As they note: “Value refers to our
ability turn our data into value. It is important that businesses make a case for any attempt to collect and
leverage big data” (IBM, 2017b). This is echoed by Johns Hopkins University data science course which
implores its students to take a scientific approach to the problem at hand and begin with a specific
question to answer using the data versus starting from the data and seeing what the analysis will give
(JHU, 2017).
That describes the “Big Data / Big Analytics” pieces of the puzzle. What about “Artificial Intelligence”?
Artificial Intelligence Most of us are using devices with AI capabilities every day. Facial recognition software built in to a
camera. The ability for Siri (on Apple’s iPhone) or Cortana (Microsoft Windows) to understand your
voice and execute commands are other common examples of AI tools. A tool like Siri simulates human
behaviour — listening to user’s voice, parsing that voice data into an action, performing that action, and
returning results back to the user. This behavioural aspect is captured in a dictionary definition below:
Artificial Intelligence (AI) is defined by Merriam-Webster in two ways:
1. a branch of computer science dealing with the simulation of intelligent behavior in computers, 2. the capability of a machine to imitate intelligent human behavior.
As discussed above, our focus is in the second definition — a machine that imitates human behaviour.
The behaviour of interest to our discussion is decision making. In the case of the grocery marketer —
which items to promote and at what price?
Daisy Intelligence 10
For use by the Food and Agriculture Business Seminar, Arrell Food Institute at the University of Guelph, June 5-6, 2017
Artificial Intelligence has been talked about for decades; some authors date its origin to 1956 (Russel &
Norvig, 1995). One 1987 article heralded: “Artificial intelligence—retail technology’s next frontier”, but
the promise of AI did not become a reality until powerful computing capability and low cost / high
volume storage entered the main (Baker, 1987). As the article alluded to above noted: “Put simply,
artificial intelligence is using a computer to manipulate whole ideas, or ‘rules,’ rather than blocks of
data—the discrete units of information used in standard computer programming, such as stock keeping
units in a merchandise planning module, numbers in a financial package or words in a word processing
program” (Baker 1987, p.11 — emphasis added). Again, rather than looking just at the impact of
promoting Coke this week, what is the impact across the entire carbonated beverage category — or, in
broader terms, the impact of the promotion across the entire business?
This notion of “whole ideas” is further reinforced by Gary Saarenvirta, Founder and CEO of Daisy, who
described what Daisy does in this way: “We take in grocery store point-of-sale receipts. That’s the
information we analyze. Everything that was ever bought in every transaction. We take in several years’
worth of data — a minimum of two, but we like to get more than five years of data. Our AI engine
translates data into operational decisions — by analyzing that data — we help the retailer make smarter
decisions around what products should I promote every week, what are the promotional prices or offers
that I should put forth, and how much inventory I should allocate to fulfil the demand on those
promotional offers. All retailers do those three things every single week. Our objective is to help retailers
significantly grow that 2% net margin by 50% or more by using math and science” (Saarenvirta, 2017).
“We bring to the table the ability to measure all of the ripple effects. Measuring the true impact of a
promotion — rather than just the promoted item sales, we can measure the total impact of a promotion on
the entire business; the ability to optimize the total profit growth. What is the baseline sale of every single
SKU when it is not on promotion — so we have a good idea of what a product does when it is not
promoted. We look at long term baseline, seasonal baseline, etc. Then we look at promotional impact.
The product you promote gets a sales lift due to price elasticity or promotional elasticity. Then we
measure the positive benefit of ALL of the other products that go with that the product you promoted. If
you promote ground beef, then people will buy pasta and tomato sauce if people are making an Italian
dinner or people will buy salsa and taco shells if they’re making a Mexican dinner — so there’s a halo
effect of your promotion and we can measure that. Then we take off the cannibalization — because you
were going to make hamburgers, you bought hamburger buns but didn’t buy hot dog buns. There’s that
cannibalization effect we measure as well as forward buying effects. From that we determine the net
promotional effect. That’s how we measure the impact of our service and report on that on a weekly
basis.”
A recent article by Forbes notes: “Artificial Intelligence (AI) and Machine Learning (ML) are two very
hot buzzwords right now, and often seem to be used interchangeably. They are not quite the same thing,
but the perception that they are can sometimes lead to some confusion” (Marr, 2016). The article goes on
to define these terms as follows:
• Artificial Intelligence is the broader concept of machines being able to carry out tasks in a way
that we would consider “smart”.
• Machine Learning is a current application of AI based around the idea that we should really just
be able to give machines access to data and let them learn for themselves.
Daisy Intelligence 11
For use by the Food and Agriculture Business Seminar, Arrell Food Institute at the University of Guelph, June 5-6, 2017
As Daisy notes in their eBook “Artificial intelligence refers to computer programs that can learn new
actions without human intervention. These programs utilize 100% of the data to exploit historical
patterns, simulate an astronomical number of alternative scenarios, and iteratively learn to optimize a
series of operational events (i.e. promotions, pricing and inventory allocations)” (Daisy 2017, p.7). This
ability to ‘learn new actions’ is really the key — the more data, the more patterns and information that can
be teased out of it over time.
Industry Intelligence Artificial Intelligence and Machine Learning also get top billing in consultant reports. The Gartner Group
sees AI & ML as the number one strategic trend in 2017 to substantially impact consumers and industries
over the next five years (Galer, 2016). In the World Economic Forum (WEF)’s “Shaping the Future of
Retail for Consumer Industries” report from January, they noted that the rapid adoption of “game-
changing technologies” like AI & ML will be a key driver of success. The WEF notes: “Industry
participants will only succeed if they have a relentless focus on using technology to increase the value
added to consumers” (WEF 2017, p.4) Furthermore, “Over the next 10 years, all of these technologies
will come of age in the retail and CPG industries, creating an unprecedented level of disruption” (ibid).
They estimate a “time to full readiness” of the technology in 2-5 years. It is important to note that
“Advances in data security to ensure protection of consumer information” is one of the key enablers for
this technology (WEF 2017, p.12).
Artificial Intelligence/machine learning can increase revenues through a deeper understanding
of consumer behaviour, while saving costs associated with supply chain optimization. In
store pricing and assortments will be optimized and, when combined with predictive
recommendations, will clearly benefit consumers and the organization itself (WEF 2017, p.14).
A January 2017 survey by consulting firm Blue Yonder of 750 grocery retailers worldwide yielded some
interesting results regarding AI & ML and the use of these tools in grocery retail:
• Retailers using machine learning have seen a reduction of up to 80% in out-of-stock rates without
increasing waste or inventory.
• In UK-specific results, 77% of grocery retail directors said these technologies (AI & ML) will
make an impact, with 23% saying they already have.
• Still in the UK, nearly 85% of directors said automation would help them to make faster
decisions, yet in spite of this, half of grocery retail managers said that ‘gut feeling’ still plays a
major part in their decision-making process.
• 31% of directors in the UK feel there are currently too many decisions to be made manually, with
the same number stating that relying on gut feel is slowing them down (Feindt, 2017)
Blue Yonder also reported on one German company that added ML to their decision-making become
much more efficient. With respect to automating the replenishment cycle, this company had:
• €20m less waste from accurate replenishment projections
Daisy Intelligence 12
For use by the Food and Agriculture Business Seminar, Arrell Food Institute at the University of Guelph, June 5-6, 2017
• €5m capital savings due to pricing optimization
• €5m efficiency savings related to streamlining processes
• 20% less out of stock because the right products were in the right quantity at the right time
• Out of stock issues were reduced from six percent to 0.5 percent because of how predictive tools
allowed them to balance between overstocking and understocking (ibid).
In addition to industry advocates, the Government of Canada has been promoting AI as well. In this
year’s 2017 budget, the feds committed $125-million to create the Pan-Canadian Artificial Intelligence
Strategy. While largely focused on research agenda goals, the AI Strategy has pragmatic goals of
increasing the number of AI skilled graduates and developing frameworks for ethics and policies
regarding the use of AI (Shecter, 2017).
There has been a lot of hype in the media about Amazon Go — a futuristic grocery experience when
customers walk in, computers track their every movement and what they put into their bags, and then
leave. Upon leaving — without going through a checkout — their Amazon account is automatically billed
for what they’ve taken. The concept is still in beta and some media outlets are reporting that the
technology tracking Amazon Go customers gets confused with more than 20 people, people who run, or
people who put products back in the wrong places, but those bugs are being sorted out. We’ll stick to
using AI/ML techniques to transform POS data into marketing insight as one proven application of these
modern analytical techniques to the retail grocery space (Hartmans, 2017).
The Customer Perspective As part of the research for this project, the author interviewed clients of Daisy Intelligence. One client’s
experience is presented here anonymously, as indicated in the Research Ethics Board application for this
project. The interviewee is a senior executive with a large Canadian retailer. This company has been a
customer of Daisy’s for the past five years. Daisy started solving segmentation problems for this
company, and then they found opportunities for Daisy in other areas based on the strength of their
analytics capabilities. The company uses Daisy’s solution to optimize the items they put into their flyer
“so we get the most reach and the best return on what we put into a flyer and what we invest in a flyer.
Keep in mind that the flyer is the most read item in Canada.” (Anonymous, 2017)
As the client notes: “There are literally millions of variables. It considers item associations,
cannibalization, weather, loading up, pantry loading, what items go with other items, items that don’t go
with other items — in simple terms that is what they [Daisy] do.” He further noted that flyers can be laid
out on a macro level for the entire year and then optimized at a micro-level at the week level on a just-in-
time basis before the flyers go to print. There is typically a 3-4 week lead time between flyer decision and
delivery.
The main questions from the client that Daisy provides answers for are: “how many items should be in the
flyer? Which items are going to sell the most and have the most associated sales? And importantly, what
items do we not need to put into the flyer?”. The client also noted that Daisy’s solution can address
“what-if” type questions like “If we put this new product in, how do we think it will sell?” Promotional
effectiveness is measured through ROI on media — investment in flyers, stripping out other variables,
Daisy Intelligence 13
For use by the Food and Agriculture Business Seminar, Arrell Food Institute at the University of Guelph, June 5-6, 2017
measuring the lift of the products in the flyer. Other indicators like flyer reach and flyer interest are also
measured, but generally speaking flyer return on investment and sales are the main measures.
One of the biggest initial challenges for this client was getting the right information to Daisy. The usual
“big data” acronym for this process is ETL — Extract / Transform / Load — whereby data is extracted
from a production system, is transformed into another file format that the downstream system can use,
and then loaded into that downstream system. The ETL process is vital to ensuring that downstream
systems (in this case, Daisy), get complete, clean data. Complete in the sense that there are no missing
data items and clean in the sense that a particular data item represents one thing; that we don’t have
multiple data items concatenated together. The ETL process includes all of the security and permissions
associated with the transfer of important proprietary data between company and vendor. However, once
the ETL processes are established and codified, subsequent data transfers become routine.
The main corporate users of Daisy’s system from the client’s perspective are the equivalents of a “Chief
Merchandising Officer” (and their team of Merchants) and a “Chief Marketing Officer” — buy-in for the
decisions Daisy’s solutions prepares are required at the highest levels. The net benefit to the client is the
picking of better items for the flyer. What defines “better items”? It depends — the solution can be
tweaked to drive more traffic, drive more sales, or a combination of both. In all of those cases however, it
boils down to ROI by putting the right items in the flyer and on the front page.
The Future
Looking to the future, the client notes that “I think we’re at an inflection point. There was a lot of promise
around AI; we see it as one of the pillars going forward. We’re past the bluster and hype stage — we’ve
tested it and it works so we’re on board. We’re leaving the old ways and moving into AI. Computing
power now gives us the ability to do that.” For the client, the rationale to use a solution like Daisy’s was
“always about the business and business decisions and improving the ROI. Doing what we do — the
fundamentals — better.”
Beyond picking items for flyers, this client sees wider opportunities for AI/ML in retail with respect to
store layout and space allocation, merchandising, and overall operations back through the supply chain.
He noted that eCommerce especially is awash in data, but there is very little insight yet. Tying bricks ‘n’
mortar to eCommerce with similar tools is the next opportunity.
Conclusion According to an article from the Financial Post in April: “consumers are more likely to go shopping when
they receive a paper flyer”; moreover “91 per cent of Canadians still read a grocery flyer each month”
(Shaw, 2017). Thus, the old-fashioned flyer continues to play an important role in the Canadian retail
scene. In that light, the decisions that merchandisers make about what goes into a flyer and what goes
where in the flyer are important, but as described above, many retailers are still going with gut instinct or
sticking to what has worked in the past vs. using data and analytics to influence decision making.
An idea that germinated in the 1950s, artificial intelligence has gone from the chalk boards of academic
researchers and the computers of early innovators into mainstream use. Complimented by techniques like
machine learning and enabled by cheap, fast, and powerful computing and massive data storage
capabilities, AI solutions to support decision makers are becoming more prevalent. In the case of the
Daisy Intelligence 14
For use by the Food and Agriculture Business Seminar, Arrell Food Institute at the University of Guelph, June 5-6, 2017
seemingly mundane weekly process of producing a weekly flyer, AI makes the computationally
expensive and combinatorically explosive number of choices easier for decision makers at leading
retailers.
Daisy Intelligence and their “Theory of Retail™” solution analyze multiple years’ worth of point-of-sale
transactions to tease out the affinities and cannibalization effects of billions of combinations and
permutations of product baskets. From this analysis, a weekly list of “key products” is produced along
with the price and inventory recommendations for the products. Products are chosen for the flyers, and the
sales lift and overall ROI are measured. Better decisions based on actual past transactions! (Daisy, 2017b)
In conclusion and to answer the opening question “why should a farmer [food & agribusiness
professional] study economics [artificial intelligence]?” the answer is: “the technology has been in
development for decades, the computing infrastructure is capable, the business imperative is real, and AI
has been shown to deliver increased ROI and sales.”
Daisy Intelligence 15
For use by the Food and Agriculture Business Seminar, Arrell Food Institute at the University of Guelph, June 5-6, 2017
References Anonymous. (2017-Apr-17). Personal phone interview.
Baker, R. (1987). Artificial intelligence; retail technology’s next frontier. WWD — Women’s Wear
Daily, 154(86):11.
Brown, W. J. (1904). Should a Farmer Study Economics? Ontario Agricultural College Review,
XVII(3):121–123.
Charlebois, S. (2017-May-02). The end of sticker shock? Canadian Grocer.
http://www.canadiangrocer.com/blog/the-end-of-sticker-shock-72495
Daisy Intelligence (2017a). Daisy’s Artificial Intelligence Series. Whitepaper, Daisy Intelligence.
Daisy Intelligence (2017b). Theory of Retail™. Whitepaper, Daisy Intelligence.
Feindt, M. (2017). Adopting AI is the intelligent move for retail. IT ProPortal.
http://www.itproportal.com/features/adopting-ai-is-the-intelligent-move-for-retail/
Galer, S. (2016). Gartner’s Top 10 Tech Trends You Can’t Afford to Ignore in 2017. Forbes, 2017-DEC-
14. https://www.forbes.com/sites/sap/2016/12/14/gartners-top-tech-trends-you-cant-afford-to-
ignore-in-2017/
Gartner (2017). IT Glossary: Big Data. http://www.gartner.com/it-glossary/big-data/
Google (2017). Big Data. https://www.google.ca/search?q=big+data
Hartmans, A. (2017-Mar-27). Technical issues are forcing Amazon to delay the public launch of its
cashier-less grocery store. Business Insider. http://www.businessinsider.com/amazon-go-grocery-
store-opening-delayed-due-to-technical-issues-2017-3
IBM (2017a). Extracting business value from the 4 V’s of big data.
http://www.ibmbigdatahub.com/infographic/extracting-business-value-4-vs-big-data
IBM (2017b). The Four 4 V’s of big data. http://www.ibmbigdatahub.com/infographic/four-vs-big-data
Johns Hopkins University (2017). The Data Scientist’s Toolbox. https://www.coursera.org/learn/data-
scientists-tools/
Johnston, L. (2012). How many Libraries of Congress does it take? The Signal.
http://blogs.loc.gov/thesignal/2012/03/how-many-libraries-of-congress-does-it-take/
Kamal, I. (2012). Metrics Are Easy; Insight Is Hard. Harvard Business Review.
https://hbr.org/2012/09/metrics-are-easy-insights-are-hard
Marr, B. (2016). What is the Difference Between Artificial Intelligence and Machine Learning? Forbes,
2016-DEC-06. https://www.forbes.com/sites/bernardmarr/2016/12/06/what-is-the-difference-
between-artificial-intelligence-and-machine-learning/
Daisy Intelligence 16
For use by the Food and Agriculture Business Seminar, Arrell Food Institute at the University of Guelph, June 5-6, 2017
Minelli, M., Chambers, M., and Dhiraj, A. (2013). Big Data, Big Analytics: Emerging Business
Intelligence and Analytic Trends for Today’s Businesses. Wiley.
Pacey, D. (2017-Apr-13). Personal phone interview.
Russel, S. J. and Norvig, P. (1995). Artificial Intelligence: A Modern Approach. Prentice Hall.
Saarenvirta, G. (2017-Apr-17). Personal phone interview.
SAS (2017). Big Data. https://www.sas.com/en_th/insights/big-data/what-is-big-data.html
Schmarzo, B. (2016). Big Data MBA: Driving Businss Strategies with Data Science. Wiley.
Shaw, H. (2017-Apr-04). Print media may be struggling but the retail store flyer’s popularity is soaring.
Financial Post. http://business.financialpost.com/news/retail-marketing/print-media-may-be-
dying-but-the-retail-store-flyers-popularity-is-soaring
Shecter, B. (2017-Mar-22). Federal budget 2017: Ottawa aims to make Canada digital leader with AI
strategy, support for fintech. Financial Post. http://business.financialpost.com/fp-tech-desk/federal-
budget-2017-ottawa-aims-to-make-canada-digital-leader-with-ai-strategy-support-for-fintech-
entrepreneurs
Sobeys (2017). Sobeys Flyer. Weekly advertising circular (May 12-18, 2017).
Walmart (2017). Walmart Flyer. Weekly advertising circular (May 11-17, 2017).
Wolfram—Alpha (2017). 50000 choose 20. https://www.wolframalpha.com/input/?i=50000+choose+20
World Economic Forum (2017). Shaping the Future of Retail for Consumer Industries. Whitepaper,
World Economic Forum.
http://www3.weforum.org/docs/IP/2016/CO/WEF_AM17_FutureofRetailInsightReport.pdf
Zikopoulos, P. C., Eaton, C., deRoos, D., Deutsch, T., and Lapis, G. (2012). Understanding Big Data:
Analytics for Enterprise Class Hadoop and Streaming Data. McGraw-Hill.