Seven business cases of using AI for retail promotion targeting · 2019-08-08 · 3 out of 100 items on promotion in grocery stores, meaning 97% are not relevant or discovered. Mass

SO1 GMBHAI FOR GROCERY RETAIL

Seven business cases of using AI for retail promotion targeting

PROLOGUE THE CURRENT STATE OF PRACTICEToday, consumers are used to personalized experiences by interacting with brands such as Amazon, eBay or Alibaba. Yet the majority of brick-and-mortar retailers, whether in grocery, drugstores, DIY, electronics or fashion, still focus on general mass promotions and circular offers. For customers, most of the promoted products are not relevant. Our research in Germany revealed that on average, shoppers purchase just 3 out of 100 items on promotion in grocery stores, meaning 97% are not relevant or discovered.

Mass promotions can also be costly. A research done by Nielsen1 reveals that the majority of the discounted sales would have happened anyway. As a result, 59% of global mass promotions in grocery retail do not break-even. For this reason, many retailers end up fighting for the consumer with everyday low prices and mass promotions to the level where profits are barely maintained.

Retail is changing and so should the approach towards promotions. Retailers need a system that will attract new customers and earn their loyalty, while at the same time increase their basket size and optimize discounts towards higher profits.

That’s where AI (Artificial Intelligence) comes into play. Modern machine learning algorithms are able to leverage the incredible amount of customer data retailers have. They can spot even the tiniest clues in customers’ behavior and recommend the right product at the right time for the right price. All of this is calculated on demand, in seconds, individually for each customer.

Our real environment tests with some of the major German and US grocers showed that our tested AI, the SO1 Engine, is able to increase redemption rates multiple times, while growing revenues from loyalty club members by 15.8% and profits by 36.4%.

In this paper, we will take a look at 7 business cases supported by evidence. The purpose is to get the retailers inspired how these technologies can help to grow their business.

2 / SEVEN BUSINESS CASES OF USING AI FOR RETAIL PROMOTION TARGETING

Table of contents

PROLOGUE 02

TABLE OF CONTENTS 03

1 MAXIMIZE OFFER REDEMPTIONS & MINIMIZE DISCOUNTS 04

2 INCREASE BASKET SIZE AND INCREMENTAL REVENUES 05

3 INCREASE CUSTOMER LOYALTY 08

4 REVEAL STRATEGIC CUSTOMER INTELLIGENCE 09

5 OPTIMIZE FOR ANY CHANNEL 10

6 CONSIDER CHANGING BUSINESS GOALS 12

7 ATTRACT ADDITIONAL REVENUE FROM BRANDS 14

SUMMARY 16

ABOUT SO1 17

SEVEN BUSINESS CASES OF USING AI FOR RETAIL PROMOTION TARGETING / 3

MAXIMIZE OFFER REDEMPTIONS & MINIMIZE DISCOUNTS

AI doesn’t need to cluster customers, it can recommend promotions on an individual customer level calculating exactly how much they are willing to pay. In real environment tests, the SO1 Engine delivered 9 times higher redemption rates while lowering the average discount by 60%, compared to general offers.

1

Figure 1: Average Redemption Rate and Average Discount of individual AI-based offers (blue) vs. promotions selected by a retailer expert team (gray). Setup: Offers distributed via printouts with the identical layout at the point-of-sale kiosk located near the entrance to the store

Each basket tells a story since every customer selects and combines products for certain reasons: she wants to cook a special meal, has a food intolerance or loves a sweet breakfast. These signals are weak and tiny, but an advanced AI can reveal them and recommend the most attractive among all available offers.

The AI can also predict how big the discount should be to avoid over or under discounting. It can calculate an individual price sensitivity for each customer–product re-lationship, allowing it to offer products for prices exactly matching the true willingness to pay.

We conducted an A/B test with a big German grocery retailer. One delivering personalized offers using the SO1 Engine, another with teams of CRM experts handpicking the best discounts of the day. In both cases, coupons

were delivered via the same channel: check–in coupon kiosks at the entrance to the stores. The AI delivered 9 times higher redemption rate (8.3% vs. 0.9%) compared to general offers, while on average it needed less than 60% lower discount (-11.8% vs. -29.7%).

Such an increase in performance is possible because the AI reveals hidden patterns that even an expert system is unable to anticipate, especially when taking into account the varying preferences of individual shoppers. This can not be achieved by segmentation, nor by a rule-based system. To learn more about this technology, read the blog post: How does SO1’s AI compare to Spotify’s.

SO1 Engine AI

REDEMPTION RATE AVERAGE DISCOUNTExpert-based system


https://blog.so1.ai/so1-ai-explained-with-spotify/

INCREASE BASKET SIZE AND INCREMENTAL REVENUES

By recommending the right products, AI can increase cross-sales and generate additional revenue from complementary purchases that wouldn’t have happened

otherwise. In 5 different A/B tests with various retailers, the SO1 engine increased incremental revenues from targeted customers by +15.8% and profits by +36.4%

(thanks to cross-sales and lower discounts offered).

2

Once we know which products individual customers like, increasing redemption rates is not too difficult. All it takes is to offer them their favourite products with discount high enough to convert them towards purchase. But an advanced AI goes beyond and increases not just redemp-tion rate, but also the true basket size and incremental revenues, meaning customers will spend more with pro-motions than they would have spent without them.

In cooperation with German retailer Kaiser’s Tengelmann (later acquired by Edeka and Rewe), we conducted an A/B test of our SO1 solution where we compared perfor-

mance via check-in coupon kiosk system. The test was running over an extended period of time in 2 regions – a control region and a region where our AI was automati-cally offering personalized discounts.

We observed an incremental revenue uplift of 2.8% across the stores in the region where AI was placed. Out of this 2,8%, around one fourth was made out of non-discounted products, demonstrating a strong cross-sales effect (when promotions in a certain category lead to higher revenues in other categories).

Figure 2-1: (a) Revenue over time measured at a store region with SO1 promotion personalization vs. control store region with regular weekly campaigns. A 12-week comparative period was measured in the beginning to be sure the performance in the 2 regions overlap. (b) The average uplift effect over the observed period was 2.8%.

Test Stores (SO1 Engine)

Control Stores

TEST PERIODCOMPARATIVE PERIOD

Weeks 2014

Inde

xed

Sale

s

Revenue from SO1 Campaign Items

Total basket size

Baseline Item Sales

+2.8%Confirmed

revenue uplift via SO1

STORE REVENUE UPLIFT

(a) (b)


Figure 2-2: A/B test: Incremental revenue and margin uplift for users targeted by personalized offers. Results vary across retailers because of different size, product portfolio, campaign pool, channels in use).

When zooming in on revenues coming from targeted customers only, the results are even more impressive. Al-together we conducted 5 A/B tests with various retailers.

On average, the SO1 Engine achieved +15.8% increase in incremental revenues and +36.4% in profits (coming from lower discounts).

LOYALTY PROGRAM REVENUE UPLIFT & SAVED DISCOUNTS

These results are possible because the AI understands the relationships between different products that work well together (complements) and which products compete with each other (substitutes). For example, the AI knows that promoting tortilla wraps will likely also make the customer buy beans, jalapenos, salsa sauce or other complementary products, generating additional revenue. It will also limit offering discounts on supplementary products that the customer would have bought anyway

unless there is a higher margin on them.

The system will understand these relationships automati-cally, from the co-occurrence of products in millions of shopping baskets, and considers this knowledge to achieve the business goals of the retailer. By recommend-ing the right products, AI can increase cross-sales and basket size while still maintaining an impression of low prices, thus increasing customer satisfaction.


MEXICAN FOOD

RELATIONSHIP

Complement

Substitute

TYPE OF PRODUCT

Beans

Jalapeños

Sweet Corn

Tortilla Wraps

MAPPING RELATIONSHIPS BETWEEN PRODUCTS (COMPLEMENTS VS. SUBSTITUTES)

Figure 2-3: Excerpt of a two-dimensional Product Map derived from basket data revealing the general affinity of a customer for Mexican food. SO1 Engine promotions favor offers from different categories (‘complement’, green lines connecting different types of products) instead of selecting offers with a high likelihood of replacing regular sales (‘substitutes’, red lined connecting similar types of products).


Figure 3: The usage of exclusive offers as % of Americans who would use an exclusive offer. Source: A survey of 1,023 adults representing the online U.S population prepared in May 2018 by Kelton Research and SheerID.

Customers are increasingly demanding when it comes to experiences. A recent survey2 done with 6,000 consum-ers across 15 countries suggests that 47% of consumers would go back to a company offering a personalized, intuitive CX, even if a rival was cheaper.

The SO1 Engine aims to do both—improve customer ex-perience by offering personalized promotions to channels of customers’ choice, and create the impression of low prices by discounting some of the customer’s favourite products.

Even the idea of couponing itself improves the customer experience. One recent study3 revealed that 94% of

Americans indicated that they would take advantage of an exclusive offer if it weren’t typically offered to the gen-eral public. Survey participants selected this option over a price-match guarantee, and 41% said they would likely seek out something to buy just to use the offer.

Personalization can also lead to more efficient recruit-ment of new loyalty program members. For 67% of consumers, discounts and promotions are among the top three pay-backs they would expect in return for sharing their data, while 39% desire exclusive experiences2. The promise of these benefits can attract more customers to sign up into a loyalty program—an essential first step towards increasing their satisfaction and loyalty.

INCREASE CUSTOMER LOYALTY

AI promotion targeting attracts more customers into a loyalty program and increases their satisfaction by providing exclusive discounts to highly relevant products. Unlike mass promotions, it does not cut down on margins or revenues, in fact, both can be increased in the process.

3

WHICH OF THE FOLLOWING DESCRIBE HOW YOU WOULD USE AN EXCLUSIVE OFFER PROVIDED BY A BRAND?

Make a purchase sooner than I normally would 48%

Be more likely to seek out something to buy so I could use the offer

I treat myself to something I really wanted, but didn’t need

Be more likely to purchase more items than I normally would

Save the offer to make a purchase for a special date

Purchase a more expensive product than I originally intended

Spend more than I normally would

41%

38%

37%

30%

28%

25%


REVEAL STRATEGIC CUSTOMER INTELLIGENCE

Machine learning algorithms can be trained on household panel data, which will allow them to attribute valuable insights to any given customer in a retailer’s CRM. These include socio-demographic, behavioral and even competition-related insights such as spend at top competitors or top categories purchased there.

4 How much do retailers really know about their custom-ers? Asking them questions when registering to a loyalty program or simply conducting panel research are common ways of obtaining customer intelligence. Both can provide an overview of who customers really are, but the only way to be sure is to observe their shopping behavior.

AI can be used to analyze massive amounts of shop-ping data and transform them into valuable insights on customers such as their age, income, household size, and various behavioral labels describing shopping behavior (e.g. Gourmet Shopper, Bargain Hunter or Brand Preference). All of these attributes can be determined for every single customer in the retailer’s CRM separately.

Even more opportunities arise as AI can also reveal customers’ behavior outside of the retailer’s stores. It can predict the shopping potential of the customer (overall monthly grocery/drugstore budget) and even top retail competitors they shop at, together with top categories they tend to buy there. These insights are valuable to tailor better campaigns to both existing and new customers and better optimize the strategy to beat local competitors.

This is possible thanks to a combination of AI and

household panel data. A unique example of such a product is Attribution+ powered by the SO1 Engine and GfK Household Panel data. First, SO1’s algorithms are trained on the robust household panel data from GfK. This data includes years of shopping history across the whole market for thousands of consumers. As a result, SO1’s AI is able to spot even the slightest differences between the shopping behavior of certain shoppers groups (i.e. customers aged between 30 - 39, earning over 2.000, with 66% of their monthly grocery budget spending at other specific retailers).

This knowledge is then applied to the customer’s shopping data of any given retailer on the market. Customer shop-ping data (baskets incl. price, discount, time, location, etc.) are imported into Attribution+ and the algorithms translate this data into customer intelligence and will update the retailer’s CRM database accordingly.

The system doesn’t need years of shopping history to know the customer. With only 5 shopping baskets, Attribu-tion+ is already able to make correct predictions in 68% of cases, and this number increases with every new shopping trip—as the AI learns more about the customer.

Figure 4: With data from 5 shopping trips, Attribution+ was able to predict the attributes in 68% of cases. After 10 trips it was 72%, after 20 trips 74%, and after 40 trips 76%. The attributes with lower accuracy can be simply rejected to avoid the risk of false data.

ACCURACY OF PREDICTIONS VS. THE NUMBER OF SHOPPING BASKETS PER CUSTOMER


http://bit.ly/2OKQfZk

A further advantage of AI is that it understands differenc-es within customer behaviors related to communication channels. Retailers can, for example, target customers when they are planning the shopping trip (through an app, email or social media), when they just entered the store (through check-in coupon kiosk or apps) or after they finished shopping (via a coupon printed on the receipt). The reaction of consumers might be also differ-ent when an offer is valid for only one day as opposed to two weeks, and if the next trip planned is the big weekly shopping trip or a smaller midweek trip.

An established AI solution is able to understand all these nuances, adapt promotional offers accordingly, and influ-ence the customer towards the desired behavior. This could be to get them into the store, to increase their basket size, or to encourage them to purchase certain brands. Depending on availability and individual prefer-

ences, a retailer can use AI-based targeting to address individual customers via multiple channels, therefore increasing the chances of a successful conversion.

Naturally, redemption rates differ from channel to chan-nel. An impulse right before the point of sale will have a very different influence than the one received at home. Measuring the performance of an identical offer, for example, revealed almost 12-times higher average re-demption rate for promotions distributed shortly before the point of sale (via check-in coupon kiosk) than those distributed shortly after (via coupons printed on the re-ceipt). Check-in couponing (a kiosk where customers scan their loyalty card and receive a print with offers) is the most effective way of increasing basket size, but retailer’s app, e-mailing, website and social media are better for bringing customers into the store.

Figure 5: Different channels offer different results. While check-in couponing is great to achieve high offer redemption and increase basket size, check-out prints, emails and apps are good to bring customers into stores (although with lower performance on previously mentioned KPIs).

REDEMPTION RATE—CHECK-IN VS. CHECK-OUT COUPON PRINTOUTS

CHECK-IN COUPONING VIA KIOSKS

CoC CiC

1,80%

21,21%

REDEMPTION RATE

11,8x

CHECK-OUT COUPONS PRINTED ON THE RECEIPTS

OPTIMIZE FOR ANY CHANNEL

Customers can react to promotions differently based on the delivery channel, time, situation, and shopping intent. AI can consider all of these nuances and offer the best mix of promotions for a given goal (e.g. to get customers into the store or increase their basket size).

5


The redemption rate is

12x higher at check-in than at check-out


CONSIDER CHANGING BUSINESS GOALS

AI can be easily switched to adjust promotion delivery with respect to different busi-ness goals, such as maximizing revenue and profit, increasing sales in underper-forming categories or improving customer satisfaction. .

6

Redemption rateIncentivised revenues Margins

Figure 6: A/B testing of 3 different cohorts favoring different business goals—left: Maximize revenue (baseline—all KPIs set to 100), middle: Increase redemptions; and right: Minimize discount. The KPI observed were Revenue (blue), Redemption Rate (light grey) and Savings (dark grey).

In various situations, a retailer’s business goals can change. When sales are good, the focus may shift to prof-its. In another case, market research may show a decline in customer satisfaction, making this a top priority for the management. AI can enable retailers to stay flexible in their business goals, without needing to impact the sales organization, asking marketing teams to develop new plans, debriefing agencies or changing tools and processes.

An advanced AI such as the SO1 Engine can quickly adapt when retailer’s business priorities change. The retailer just needs to change the goal setting in the frontend, and the AI will change the promotion delivery in real time in ac-cordance with the new goals.

Retailers are usually aiming for a maximum increase in revenue. However, this can be easily changed into higher customer satisfaction (by offering products the customer likes a lot), maximizing profits (by selecting high margin products and minimizing discounts), attracting new category users (by offering higher discounts on products underrepresented in a customer’s basket) or reducing write-offs (emptying stocks).

A combination of goals is also possible, e.g. aiming for the most efficient increase in revenues so that the SO1 En-gine will consider high customer satisfaction and profits at the same time. Results show that SO1 Engine can suc-cessfully trigger conversions at a wide parameter range, even with strongly reduced discounts.

CHANGE IN KPIS ACROSS DIFFERENT SETTINGS RELATIVE TO THE BASELINE (AVERAGE PERFORMANCE UPLIFT)



ATTRACT ADDITIONAL REVENUE FROM BRANDS

Brands are increasingly shifting towards digital channels that allow better targeting. By integrating AI promotion targeting into brick-and-mortar retail, retailers can in-crease the ROI of brand promotions and retail media, therefore attracting additional revenues.

7

Figure 7-1: The inefficiency of current promotions is driving brands to allocate budget to channels with a higher return on invest-ment. Source: a worldwide study by Nielsen1

Brand marketing budgets constitute the majority of to-day’s promotion spending. Faced with the inefficiency of current promotions, brands are reevaluating their strate-gies and shifting increasing amounts of their promotion spending to channels with higher efficiency, e.g. targeted digital marketing.

Consequently, brands aiming to increase their market share are reallocating their budgets in favor of channels that offer:

• high efficiency (return on investment) via personal-ization

• high impact in terms of reach and sales quantity • fully-automated execution (programmatic promo-

tions) aligned with brand goals • measurable impact and payment only for successful

conversion

0 10 20 30 40

2012

50

7.1

49.8

201719.9

43.4

FMCG MARKETING EXPENSES BY TYPE, IN PERCENT OF ALL EXPENSES

59% OF PROMOTIONS DON’T BREAK EVEN

59%

Trade Promo Digital


By applying AI to promotion targeting, retailers can easily meet these demands and therefore attract additional revenues from brands. Retailers can, for example, offer brands a self-service solution to launch AI-optimized promotions with individual goals, e.g. initiating a brand switch campaign precisely targeting a specific group of customers based on their purchase history. And thanks to features like brand blacklistings and retail goal prioritiza-tion, retailers remain in full control.

Integrating the SO1 Engine for personalized promotions enables retailers to engage in a win-win collaboration with brands by sharing their highly efficient channel. While brands benefit from an unmatched ROI on their promotion spending, retailers increase their topline revenue and preserve their profits because discounts are fully covered by the manufacturers.

Figure 7-2: Return on investment measured for brand campaigns from different product categories. The chart depicts the performance of distributing each campaign via couponing, circular promotion and using the programmatic promotion channel offered by SO1 Engine.

Couponing Circular Promotion SO1 Promotion

Biscuit Frozen Pizza Liqueur Coke Frozen Dessert Ice Cream

RETURN ON INVESTMENT OF BRAND PROMOTIONS PER PRODUCT CATEGORY



SUMMARY With innovators such as Amazon constantly increasing customer expectations, and discounters such as Lidl and Aldi pushing prices down, things are getting complicated for the majority of retailers stuck in between. However, building meaningful customer relationships is often just about small things done really well, such as recommending the right product for the right price to the right customer.

Mass promotions attract customers to a retailer’s store and generate peak sales, but come at a high cost in terms of the retailer’s profit margin or supply chain costs. Mass promotions or incorrect targeting also fail to build a relationship with the customers. In contrast, artificial intelligence has been proven to effectively reveal customer’s preferences, target them with relevant offers, and increase their long-term satisfaction and loyalty.

EXTERNAL RESOURCES (1) Trade promotion doesn’t have to be a guessing game, Nielsen, 2015(2) What customer experience do consumers REALLY want?, Verizon, 2019(3) New Survey Reveals Consumer Conflict between Privacy and Personalization, Kelton Research and SheerID, 2018

About SO1

SO1 (Segment of One) helps brick-and-mortar retailers to individualize, optimize and automate their promotional offers with smart solutions based on a unique, proprietary AI. This AI creates promotions that influence every single customer’s purchase decision in a most efficient way and builds meaningful one-to-one relationships:

Automatically identifying latent product attributes

Understanding individual consumer preferences

Calculating the true willingness-to-pay per product and consumer and adjusting discounts accordingly

Predicting the individual buying intention to avoid substituting offers

Aligning its recommendations with the retailer’s business goals, like revenue or profit maximization or pushing customer loyalty

The SO1 Engine is already running with some of the major German and US retailers and was successfully tested against many existing market solutions with outstanding results. This technology was created by some of the leading AI and machine learning experts in cooperation with scientists from MIT, ETH Zürich and Humboldt University Berlin.

For more information and case studies, visit www.so1.ai or contact:

SALES US

PATRICIA A. CUCINELLIPhone: +1 917 757 6221E-Mail: [email protected]

SALES EUROPE

STEPHAN VISARIUSPhone: +49 160 93 59 69 95E-Mail: [email protected]


mailto:cucinelli%40so1.ai?subject=Whitepaper%3A%20Nine%20Reasons

mailto:visarius%40so1.ai?subject=Whitepaper%3A%20Nine%20Reasons

Seven business cases of using AI for retail promotion targeting

SALES US

PATRICIA A. CUCINELLIPhone: +1 917 757 6221E-Mail: [email protected]

SALES EUROPE

STEPHAN VISARIUSPhone: +49 160 93 59 69 95E-Mail: [email protected]

GENERAL CONTACT

SO1 GMBHPhone: +49 (0)30 208 987 30Fax: +49 (0) 30 208 987 228E-Mail: [email protected]

ENGINE

mailto:cucinelli%40so1.ai?subject=Whitepaper%3A%20Nine%20Reasons

mailto:visarius%40so1.ai?subject=Whitepaper%3A%20Nine%20Reasons

mailto:info%40so1.ai?subject=Whitepaper%3A%20Nine%20Reasons

Seven business cases of using AI for retail promotion targeting · 2019-08-08 · 3 out of 100 items on promotion in grocery stores, meaning 97% are not relevant or discovered. Mass

Documents