Is email marketing dead for restaurants? Hardly.

With open time personalization of email content, quick-service operators can delight their customers with contextually relevant, hyper-personalized engagement in real-time.

With social media on everyone’s radar, has email as a marketing channel become obsolete? The answer is an emphatic “no.” With hyper personalization, email is still one of the most effective marketing tools that quick-service restaurants can use to acquire and retain customers. Thanks to open time personalization, the quick-service sector has upped its game and the customers are loving it.

However, the mantra to hyper personalization is how you get your act of algorithms right—that is having the right content at the right time and in the right place. For that “right” everything, quick-serves need to know who their customers are, what they are looking for, when they want it, where they want it, and how they want it. Open time personalization is all about 4Ws and 1H.

Individual customer behavior data is critical to open time personalization. The base is a customer data platform (CDP) which has demographic, transactional behavior information of the customer, all sourced from different platforms (point of sale, third party data, CRM systems, website, mobile app, etc). Algorithms take over and curate information about customers to create one comprehensive profile of customers for further downstream analysis from where one can drive ai email personalization.

Open-time hyper-personalized email campaigns are highly dynamic, where the dynamic content personalization in the email is contextually personalized, not at the time it’s pushed to the receiver, but at the time he/she opens the email. Essentially, it means that if there has been any change between the time the email was sent and opened, that impacts what is published in the email, then the change is incorporated to make it relevant to the minute of email opening.

While it is personalized when it is sent to the customer, it gets updated with the latest information based on any transaction that has happened post sending email or there has been a change in the offers or availability of the product. When a recipient opens the email, the email client sends a request in real-time and the personalization platform, using its algorithmic intelligence, pushes the most recent and relevant content directly into the email template.

To illustrate, a quick-service restaurant sends an email offer with a discount on Peri Peri Chicken at 7 a.m. to one of its regular customers, Lily, who during the day has browsed for Crispy Fried Chicken. She opens the email by 2 p.m. and sees an offer on Crispy Fried Chicken. Voila! That is what Lily was looking for and to boot there is a discount offer, exclusively for her. This real-time product recommendation may probably get the quick-serve a happy customer who, again, may look out for such deals going ahead.

Not just that, the real-time inventory updates feature makes changes or updates to the menu at the time the customer opens the email. For instance, the quick-service brand sends Robert an email at 9 a.m. on a combo offer (Veg Extravaganza and Wedges), but Robert only opens the email at 3 p.m. and the offer has changed from Veg Extravaganza to Veg Delight (also Robert’s favourite) and Wedges as the quick-serve ran out of the item. This real-time inventory update is win-win for the restaurant as well as the customer.

Not just that, the real-time inventory updates feature makes changes or updates to the menu at the time the customer opens the email. For instance, the quick-service brand sends Robert an email at 9 a.m. on a combo offer (Veg Extravaganza and Wedges), but Robert only opens the email at 3 p.m. and the offer has changed from Veg Extravaganza to Veg Delight (also Robert’s favourite) and Wedges as the quick-serve ran out of the item. This automatic store replenishment update is win-win for the restaurant as well as the customer.

At 7 a.m. the weather at San Jose was bright and sunny. Grill & Chill sent an email promotion for their coolers and salads to their regular customers. By 9 a.m. it began drizzling and the weather got a bit chill. With open time personalization technology, Grill & Chill could instantly change the offers to Soup & Bread to those who opened the mail at 10 AM.

With open time personalization of email content, quick-service operators can delight their customers with contextually relevant, hyper-personalized engagement in real-time. This enhances customer experience, offer take up, conversion and lifetime value. According to eConsultancy, 64 percent of consumers expect companies to respond and interact with them in real-time and it is happening. AI algorithms make this happen with real time customer data profiles that activates audience and segments and then pushes that to personalization engine for 1-to-1 personalization via omni channel orchestration tools.

The article first published in QSR Magazine

Have you ever wondered how, as you are shopping online, adding items to your cart, the website seems to suggest products that perfectly complement what you have already bought or added to your cart? This is known as cross-sell: recommending additional items that complement the products you are currently considering.

Traditionally, cross-sell has been done using a method called collaborative filtering or statistical “wisdom of the crowd” models. These methods are great in that they are relatively simple and fast. Unfortunately, they do not work well for new products or categories of products that do not get much traffic, i.e. the long tail, because they cannot make recommendations for products without sufficient view or purchase data. Methods using more abundant data (like views or clicks) or content-based unsupervised methods can reliably provide similar product recommendations, but fail to capture the complex associations between products that make good cross-sell.

NLP is a branch of linguistics and artificial intelligence that uses machine learning to analyze and convert natural language into useful representations or insights.

At ADA Global, we use NLP in many ways. One of which is in extracting structured information from unstructured texts. For instance, gathering product features or metadata from its text description (Topic for another blog post). But here we will focus on our deep learning NLP cross-sell models, hereafter shortened to “DeepRecs NLP.”

Rather than using Product IDs as input, like a collaborative filtering or statistics based method would, DeepRecs NLP uses natural language and structured metadata, such as product descriptions, reviews, brand or categories. This allows the model to uncover patterns in the language used to describe a product and therefore inform cross-sell. More importantly, it allows the models to reach products that have rarely been viewed or bought, e.g. new or  long-tail products. These are products with relatively little purchase (or view) data.

We leverage shopper data to build neural networks (a specific type of machine learning model, broadly encompassed by the field of deep learning). These networks consider larger patterns of shopper behavior by learning from pairs of products that co-occur. For example, products that were purchased by the same user, within a given time window, are “connected.” Based on the number of times a product pair co-occurs across all users, each pair is assigned a score between 0 and 1, with a higher score denoting better cross-sell fit. Specifically,

where is the co-occurrence count of product p2 with product p1.

From these product pairs, a seed product and a recommendation (rec) product is defined. Each will have various information associated with it, such as name, brand, description(s) or reviews. We normalize the text data. From there, based on the input type, we might process input in different ways – more on that after we look at the results our clients are seeing.

DeepRecs NLP has consistently shown increases to key metrics our clients care about, such as click-through rate (CTR), average order value (AOV), and revenue per visit (RPV). Of the five clients for which we ran A/B tests comparing slates of recommendation strategies with and without DeepRecs NLP, the treatment side (with DeepRecs NLP) for all five showed a statistically significant increase in CTR. In most cases, DeepRecs NLP also showed an increase to AOV and RPV, with one site obtaining a 7.296% increase to RPV!

Many clients were so convinced with the initial results from DeepRecs NLP that they skipped the A/B test and immediately made it live to all traffic in key places like the Product Detail Page or Add-to-cart pages. In one month alone, DeepRecs NLP provided over 14.5 million viewed recommendations to 20 sites world-wide.

In our DeepRecs NLP system, we use two identical “stacks” of neural networks, one for seed products and another for rec products. In this section, we detail some of the specifics of our model architecture.

Each input feature gets encoded in different ways, based on its type. For instance, for discrete categorical data, e.g. brand, we learn a dense “embedding” vector representation for each unique value, or a “multi-hot” embedded vector representation if there are multiple categories(a vector is basically a series of numbers like [0.52, 0.02, 0.97, 0.08, 0.55]). For unstructured text input we might “embed” each word and run them through an Attention layer [1][2]–a mechanism for combining vectors, focusing on important values to get a feature-level representation. Alternatively, we might run the raw text through a pre-trained BERT model [3], followed by one or more fully-connected neural network layers to capture cross-sell specific relationships from the pre-trained model.

The result is that we have vectors of the same size for each input. A benefit of processing inputs in this way is that we can ingest an arbitrary number of inputs, determined by the client. We then combine them all (usually with a weighted sum or another Attention layer) and optionally add more fully-connected layers to get a final product vector.

To produce a score between seed and rec products, we take the cosine similarity of the normalized final product vectors. For training, this similarity score is compared to the score from Equation 1 using a hinge loss. At prediction time, the similarity score can be used to rank pairs of products for cross-sell.

Each of the “usual” model options mentioned above are automatically determined during hyper-parameter optimization based on the complexity of the site catalog and loss metrics on a validation dataset. This allows each of our clients to have a custom, personalized model that is best suited to their catalog.

BERT has been a breakthrough in the world of NLP for many tasks, including our cross-sell recommendation systems. Because they were trained on a lot of data from the internet, pre-trained BERT models can process just about any text in over 100 languages into a meaningful, dense representation that can be used in downstream tasks without the costs of training from scratch.

Unfortunately, training on internet text also means that the models implicitly encode potentially harmful language bias. Research suggests that BERT introduces language bias into systems that use it ^[4][5][6]. For example, when used in a Natural Language Generation system, it might always associate “Doctor” with male pronouns and “Nurse” with female pronouns.

In the context of cross sell, consider childrenʼs clothing and toys. It is a highly gendered catalog space because that is what society and shoppers reinforce through norms and macro-patterns of purchases. If you add into those interactions the gender bias of BERT, it is not hard to imagine a scenario where NLP cross-sell disproportionately recommends a kitchen play-set or Barbie doll for a unicorn blouse seed, but swim trunks or a backyard science kit for a shark graphic tee seed. Although contrived, this example demonstrates how training data can inject unfair bias into recommendations, ultimately impacting the interests or opportunities of different groups of people. At ADA Global, we are actively researching how large of an issue this is in our models and ways such bias can be mitigated.

Cross sell models that understand industry context (in this case retail) and address domain-specific challenges like catalog long-tail can create high business value. At ADA Global, we leverage deep learning techniques as well as shopper history, patterns and insights that improve relevance of cross-sell, enhance the shopping experience and grow metrics such as AOV and RPS.

References

1. Minh-Thang Luong, Hieu Pham, and Christopher D. Manning. 2015. Effective Approaches to Attention-based Neural Machine Translation. In Empirical Methods in Natural Language Processing (EMNLP).

2. Nikolas Adaloglou. 2020. How Attention works in Deep Learning: understanding the attention mechanism in sequence models.

3. Devlin, Jacob, et al. “Bert: Pre-training of deep bidirectional transformers for language understanding.” arXiv preprint arXiv:1810.04805 (2018).

4. Tan, Y. and L. E. Celis. “Assessing Social and Intersectional Biases in Contextualized Word Representations.” NeurIPS (2019).

5. Kurita, Keita et al. “Measuring Bias in Contextualized Word Representations.” Proceedings of the First Workshop on Gender Bias in Natural Language Processing. Association for Computational Linguistics.

6. Bender, Emily M. et al. “On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? .” Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency. Association for Computing Machinery.

An optimal menu not only drives financial success but also enhances guest experiences. This makes Menu Engineering a key aspect of QSR strategy. With real-time insights, Operations Leaders could engineer their QSR menu to keep costs under check while ensuring customers’ favorites are firmly placed in the menu.

The sudden spike in online ordering through apps presents immense opportunity for QSRs to hyper-personalize customer experience. And menu personalization is a key element of the overall experience. Data on customers’ taste and preferences are collected and analyzed. Using machine learning and AI technology, the app shows users the most relevant menu items, promotions and content based on their individual preferences, past dining history, location, weather and restaurant specific menus and pricing.

With data and intelligence, menu engineering is more science than art, in today’s times. An advanced AI platform helps you engineer your menu across outlets, reduce cost and drive loyalty. The underlying need to design optimized menus is Data – menu items, number sold, food cost, menu price. Clean and structured data on your customers, likes, preferences, transactions, affordability, menu items and their costs are some of the key inputs to the menu formula.

Advanced algorithms analyze the demand and costs of menu items to help you identify your most and least profitable menu items. AI platforms integrate with back-of-house systems to draw this data and throw up in-depth intelligence to help make smart decisions on menu and other aspects of QSR business.

An advanced BI and Restaurant Analytics platform provides real-time insights to re-engineer your menu, forecast sales, reconcile 3rd party sales, optimize delivery time and more.

Manual menu analysis is bereft of inaccuracies and never complete. This leads to loss of revenue opportunities and impacts profitability. As alluded to earlier, data and intelligence could help identify popular menu items across stores and locations in real-time. Advanced algorithms map trends, cull out underlying reasons and recommend next best actions for menu items. For e.g., Algorithms throw up the bestselling pizza by size, day, topping, location and recommend menu combos that would help you leverage its popularity.

By adding data on item cost and profitability to sales of menu items, you can be sure to accurately identify the best and worst performing items, from your long menu list.  Using this, QSRs could develop an action plan to optimize the menu by reducing low profitability items and increasing the popular ones.

Menu engineering is about managing this mix of stars, plow horses, dogs and puzzles that leads to reduced costs, improved profitability without compromising on customers’ wants. Let’s see how algorithms could help in arriving at the right menu mix.

Algorithms identify Star menu items based on factors such as sales volume, costs, price, margins, promotions, etc. They help you analyze the underlying reason and recommend ways to make the most of these popular and profitable items on your menu. AI platforms recommend:

• Other locations where these items can be introduced to enhance revenue opportunity
• Price increase by calculating the likely impact on demand
• Menu combos with other items that are lower in profitability, using association mining

Algorithms are quick at identifying items that are growing in popularity. Juxtapose this data against its costs and you know whether you are heading towards the left of the balance sheet or the right. Trust the AI to do the magic – menu combos that offsets the low profitability of the item or more dishes with high priced ingredients for economies of scale. You may just end up with Star with AI-based recommendations.

With a 360-degree view of sales, marketing & operations data, AI platforms turn into Sherlock Holmes to find that missing piece of the puzzle. Run the analysis and in a matter of seconds you’ll know why they are not flying off the shelf. Turn the recommendation engine on to increase their demand. Reduce price, introduce an offer or run a campaign, these algorithms have a marketing trick up their sleeve.

While the algorithms are quick at identifying Dogs and recommending their expulsion, human intelligence needs to prevail here. For e.g., Kids’ menu is neither high volume nor highly profitable ut you may want it to stay on the menu.

Data and Intelligence are the key ingredients in dishing out a menu with the right mix of Stars, Plow horses, Puzzles and Dogs. Advanced AI tells you what’s popular and why. Likewise, it helps you understand the implications of cost and pricing. And this is a dynamic exercise as customer tastes and preferences, economic and social situation, change. With real-time access to data & insights, AI platforms help you balance high and low food costs to garner a reasonable amount of profit margin yet cater to customers’ preferences.

Your visitors search in different stages of the customer journey – when they know what they are looking for, when they are not sure and need guidance, or they want to solve a problem, but don’t know how.

1. Exact search terms, e.g. model # in electronics, or complex searches such as ‘red midi dress in size 12 sleeveless’.

2. Search with abbreviations e.g. 16”/ 16 in/ 16 inches OR oz/ ounce.

3. Search with spelling errors, or spelling variations? E.g. hair drier instead of hair dryer, or a typo such as Avacado.

4. Alternative words, such as flip flops instead of house slippers or thongs (yes, you – from Australia), or dress shirt instead of a formal shirt, bed sheet or bed linen.

5. Use plural word forms (tomato/ tomatoes), or use slang, shades instead of sunglasses.

6. Problem based search terms, e.g. ‘dry skin’ instead of ‘moisturizer, or ‘turmeric marks’ instead of ‘stain remover’ or ‘headache’ instead of ‘pills’.

There is a wealth of insights in shopper searches, that can be used for showing relevant results for the individual, and for improving overall results, and reducing zero hits.

7. Some terms are more critical than others – can you recognize a 11” sleeve (11” is important), or organic kale (organic is key).

8. Your shoppers’ collective searches can help you form associations between search terms and products/ categories, without changing product data. Self-learning algorithms can improve results without having to create rules, or working on data attributes.

9. This drives your differentiation – A shopper’s affinity for brands, products, categories or custom attributes such as concern type, color, price become known from their current session as well as history. If you can personalize their search experience using their profile (even if they are anonymous), you will be rewarded with higher engagement, conversions and loyalty.

Shoppers are often multi-tasking – they could be distracted or just need assistance with creating a shortlist. This is where search nuances come in.

10. As shoppers type, they may need inspiration. With predictive auto-complete similar to  what Google has trained the world on, you can make sure they don’t feel lost.

11. When products need multiple dimensions for selection, contextual facets and filters can help them narrow down their choices.

12. More than 50% traffic is now from mobile, and the limited screen real estate necessitates personalization. Finding the right result on the first page, what RichRelevance calls Findability, is negatively affected, since the user has to scroll more, use filters, and their journey is longer.

The personalized search box promises simplicity to the user, but is a powerful and sophisticated tool that can help marketers and merchandisers boost conversions, while retaining business controls and commitments around private labels, over-stock situations and trade promotions.