Accurate takeaway recommendation: not so fast, or even at all

2020-07-31 22:50:15 0 Comment 278 views
abstract

The author of this article will take the particularity of the takeaway product type to realize the impact of accurate takeaway recommendation on the e

The author of this article will try to analyze the impact of accurate recommendations on the existing business model of the platform, the current state of competition in the food delivery market, and the impact on the realization of precise targeting from the particularity of the type of takeaway products: Why can't we achieve accurate takeaway recommendations?

Accurate takeaway recommendation: not so fast, or even at all

After choosing for so long, I don’t know what to order. I believe many people have had this experience like me. The level of desire for accurate recommendations for takeaways can be described by endless calls. Use the recommendation results to reduce the time for takeout selection, and even choose to order takeaways through recommendations instead ofGive up takeout and choose to cook or eat out.

Reducing the time for users to make purchase decisions, increasing the rate of order transactions, improving user experience, increasing user retention, and realizing accurate and personalized recommendations for external sales seem to be really beneficial. However, the food delivery platform has been hot for a long time, and the recommendation system has been around for a long time. However, in such a long time, we have not waited for the food delivery platform to achieve accurate and personalized recommendations.

Takeaway platforms have always had recommendation systems, but they have not yet achieved our ideal "precision and personalization". Recommendations are based more on groups, geographic locations and a small amount of user behavior. . In addition to the completeness of data, the support of algorithm technology, and the input of computing resources to achieve accurate recommendation, the most fundamental thing is to consider the necessity of achieving accuracy from the beginning, or to maintain appropriate personalization, and also face commercial value versusGame between user experience.

This article will try to analyze the impact of precise recommendations on the platform’s current business model from the particularity of the takeaway product types, the current competitive state of the takeaway market, and the impact on the realization of precise targeting: Why is the precise recommendation of the takeaway that we haven't waited for?

Accurate takeaway recommendation: not so fast, or even at all

(Meituan Waimai’s existing takeaway recommendation)

1. For some people, ordering takeaway is a rigid demand scenario

Although since the end of last year, the United States The group began to increase the commissions of takeaway merchants, and the average price of takeaways also rose slightly. But quotientHome and the market share reaction is still relatively benign, there is no major fluctuations, the first quarter of this year's financial report of Meituan shows: the takeaway rate and the number of transactions are still showing an upward trend.

Accurate takeaway recommendation: not so fast, or even at all

(Source: Meituan’s 2019 Q1 financial report)

This phenomenon shows that there are still many users who are not very price sensitive when selling out. In many cases, ordering food has become a rigid demand scenario, especially For some user groups in first- and second-tier cities, ordering food has become a part of their daily life. Compared with price-sensitive users in lower-tier cities, the increase in food prices may shift to self-cooking, but there are someUsers will choose to order takeaway no matter how the situation changes.

Mainly manifested in several aspects:

  • Personal factors: not able to cook, too lazy to cook, tired of eating restaurants around the dining hall, and no time for work.
  • Environmental factors: There are no restaurants nearby or the consumption level of nearby restaurants is high.
  • Weather factors: rain, high temperature, severe cold, etc.
  • ……

If in a certain situation you will eventually order takeaways, it doesn’t matter for the platform to do accurate recommendations for takeaways, you Eventually, a deal will be reached, and even an order that exceeds his "budget" will be completed. But if you have other choices besides ordering takeaways, such as cooking, eating out, and dining, then accurate takeaway recommendations have value for the platform—retaining users, increasing transaction rates, and increasing purchase profits.

For many people, takeaway is a just-needed scenario, but for some users, it is just one of the choices. Which user has the upper hand? The relative advantages of sticky users and lost users affect the takeaway Recommend the necessity of system development.

Second, the impact of accurate takeaway recommendations on the marketing business of existing merchants on the platform

The balance coexistence of merchant activities and precise recommendations that are different from other e-commerce shopping , Accurate recommendations for takeaways will have a big impact on merchant advertising and marketing activities. The general e-commerce purchase decision time may span several hours, days or even longer. The purchased items are not considered urgent, and will pay more attention to shopping around, spend more time browsing, and repeatedly check.

The selection time for takeout purchases is very short. It usually only takes a few minutes to ten minutes from opening the APP to completing the order. The pursuit is fast selection., Order quickly and deliver quickly. Once a precise personalized recommendation for takeaway is formed, the user's purchase process will most likely only focus on the recommendation result and choose to ignore other products from other merchants.

Accurate takeaway recommendation: not so fast, or even at all

(Meituan Takeaway Merchant Activities)

Meituan Takeaway’s existing merchant marketing forms include out-of-store coupons, takeaway food activities, etc. Like other types of products, this set of models of the takeaway platform is also based on Initiated by the user’s price-sensitive nature.

Different from the price sensitivity of some users mentioned above, it means that the acceptable price rises and fluctuates slightly, and the preferential price is still for most users.It is more attractive.

For users who do not have clear needs, these merchants who are exposed by issuing coupons and participating in activities will undoubtedly pay more attention to viewing, or is it because the decision-making time of takeaway is short, and takeaway is a form of marketing promotion and exposure order Conversions should be better than other product types. Once the precise recommendations for takeaways are formed, this marketing model will not be able to maintain the existing high levels of attention, and users will turn their attention to browsing and following the recommendation results.

Customer unit price is an important dimension of recommendation evaluation. The recommendation result is based on the user’s acceptable price range. Once the recommendation result’s price meets user expectations, discounts will be given to those who are carrying out activities Voucher merchants will not pay much attention unless they can't find the result they want in the recommendation. You may have questions: "Taobao will also issue merchant coupons and guess you like it?"

Taobao merchantsCoupons are more to provide users with one more choice, and they will naturally click to view when shopping around. The choice of take-out will not take too much time to compare, don't want too many choices, just make a quick decision, it is unlikely that you will check the recommended results and then compare the merchants with special offers.

Isn’t it enough to add active merchants to the recommended results?

Accurate takeaway recommendation: not so fast, or even at all

(Today’s headline content recommendation page ad interspersed)

The recommendation result is calculated to predict the possible purchase of the user’s preferences, and thenSort in descending order of possibility.

Different from content distribution recommendations such as Toutiao, Douyin, and e-commerce, users are not purposeful, have no request for rapid decision-making, and do not refuse to browse the large recommendation results page, and the decision time for takeaway purchases Short, it is likely that the order selection was completed in the single-digit product results. Even if this advertising content achieves high-precision delivery, fine audience targeting, and outstanding click conversion effects, the number of commodity merchants exposed through this form of advertising will be very limited, the overall exposure traffic will be greatly reduced, and the overall advertising revenue will be large discount.

If it is to increase the advertising fee for each exposure of the business, this approach will increase the head effect of the platform. Advertising can bring huge high conversion traffic to the business, but for the business without advertising In terms of traffic, there will be a lot of pressure, and will be robbed by too much traffic.

Accurate takeaway recommendation: not so fast, or even at all

3. Competition in the takeaway market has not yet reached the time for precise online audience targeting.

Nowadays, all major platforms are admiring precise recommendations. A major benefit is the ability to achieve precise audiences for major advertisers. Targeting, precise placement of advertisements. If you log in to the website where the Meituan takeaway channel advertising is placed, you will see a sentence at the bottom:

(URL: http://www.meituanwaimai.com.cn/product.asp)

Although the hot state of the takeaway market has lasted for a long time, most of theTime is a price subsidy battle between multiple takeaway platforms. The focus is on harvesting user traffic, penetrating and expanding the market size, and relying on platform traffic to attract catering businesses to settle in. It is only since last year that Meituan Food Delivery has gradually increased merchant commissions, but the market has not reacted too violently, verifying that the platform itself has a large number of sticky and loyal users and merchants.

In this long battle for user traffic, Meituan Takeaway is currently in the leading position, but this advantage is not obvious. As a leader in the food delivery market, Dake will continue to play the dominant card of high-frequency scenarios and high-traffic platforms.

This is why Meituan Waimai’s current online advertising resources are still charged based on CPT (settled by display time), and there is no need to initiate targeted advertising to attract advertisers, and once targeted advertising is competed The opponents followed suit, and other platforms to achieve targeted delivery can also serve as takeawaysHome brings high-precision user traffic, and platform advertisers are easily divided.

For the relatively laggards in the market (e.g. Are you hungry), the main goal at this stage should be to try to grab more user resources from the leaders while maintaining their existing market scale. If you are the first to invest human and financial resources in the research and development of the huge project of accurate targeted advertising, your own market will easily be swallowed by the leaders.

Accurate takeaway recommendation: not so fast, or even at all

(Source: Introduction to Meituan Takeaway Advertising Resources)

Four. The particularity of takeaway from other products

For other products, in addition to the short purchase decision time mentioned above, some other unique attributes also make it not easy to achieve accurate recommendations for takeaways.

1. Takeaway pays more attention to novelty and surprises

When I am aimless and do not know what to choose, I tend to choose those popular rankings. Before the commodity. The purposeless situation is more common in takeaway purchases. The ranking recommendation traffic of takeaway platforms has always maintained the largest proportion of traffic.

For each user, the popular top products in the geographic location will basically not fluctuate much within a period of time, that is, the ranking of the product list that the user has seen in a long time The results are basically the same. At this time, the value of the recommendation system lies in recommending relatively less popular products for users. The less popular products are more likely to make users feel novel.Products, the more likely it is to buy other products that are not popular.

However, takeaway is not like other products. Different items have different styles, styles, materials and other characteristics, and there are many varieties. There is little difference between different products of the same type from different takeaway businesses, and it is difficult to tell the obvious difference in ingredients and taste. The types of takeaway are also very few, and the definition of novel products for users is also a big challenge.

Accurate takeaway recommendation: not so fast, or even at all

(Source: Official Account Meituan Dianping Catering Academy)

The embodiment of surprise lies in the recommendation of products that are not similar to the user’s historical interests but satisfy the user.

When ordering the same merchant or the same takeout multiple times, it may not mean that the user prefers this type. In most cases, in order to save time and effort to choose, just continue to use the previous order Selection, taste, price, delivery time, etc. of the takeaway are still within the personal acceptable range. Even if we are a little bored, it is still acceptable. We are eager to have more other options, but we don't want to spend too much time and energy to find them.

This kind of "false" preference feature is a common trap in take-out purchases. It is necessary to be cautious to determine whether the user has completed the same order multiple times because of personal preference or just to solve the meal Repeated mechanized purchase execution completed by the problem. And the more this preference behavior may occur, the greater the weight, the less credible the prediction of user preferences, and the greater the demand for surprise.

2. It is difficult for take-out users to finely define in multiple dimensions

When we buy breakfast bread on Taobao, it may know that you are a user who has the habit of preparing breakfast, and will recommend other products related to breakfast for you. When we buy "Apocalypse" on Amazon, it may recognize that you are a product manager just getting started, and recommend other related books of product managers for you.

But the behavior of eating is something everyone has. It is difficult to identify the user’s behavior habits or roles through the user’s order behavior to clearly or finely define the user type. Probably the only way to construct user portraits can be based on user tag features with few dimensions, such as region, personal category preference, customer unit price, etc. The scarcity of user features will also make the recommendation results calculated based on the similarity of users and items relatively rough, which most people see The results of recommendations are likely to be similar, and recommendations for takeaways may not be highly personalized like other products.

Accurate takeaway recommendation: not so fast, or even at all

(Source: " Meituan Waimai’s 18-year commercial advertising introduction")

3. Time factors in food delivery

Users’ interest in food delivery may gradually change during working days , Basically, there will not be much change, mainly fast food rice. However, during weekends and holidays, user interest may be abrupt, with many choices, and may show completely different user interests from weekdays. Some types of takeaways will not be enjoyed on weekdays, and some types of takeaways will not be ordered on weekends.

The difference between weekdays and weekends and holidays for takeaway recommendations is an issue that needs to be considered.Comprehensive gradual user interest and sudden user interest. On weekends and holidays, the unit price of users will be greatly increased compared to usual. Eat well and reward yourself after a week of hard work and study. At the same time, the time for decision-making and selection of takeout will also increase. I like to try new types of takeout. The diversity of takeaway recommendations can be appropriately increased during weekends and holidays.

Fifth, business value guides business realization

The prerequisites for the generation of the recommendation system are information overload and unclear requirements.

These two characteristics are undoubtedly the most characteristic portrayal of the current Internet industry and Internet users. The industry development trend of the Internet from an incremental market to a stock market has also created the popularity of recommendation systems. Algorithm package products solve problems and retain and convert users. All walks of life hope to advance from the Internet of Tools to the Internet of Algorithms, complete product upgrades, and satisfy users.User needs.

Although the development of the industry has given birth to new business scenarios and new technology upgrades, products are business-oriented at all times, and recommendation systems are no exception. They need to be satisfied with the core Polaris formulated by their own product development. Indicators, to avoid following the craze, under the guise of “upgrades” to do “improvements” that do not actually improve the business or even risk damaging the existing business system.

Recommendation results should be able to allow users to perceive true intelligence and assist decision-making. For the enterprise platform, everything needs to start with commercial value. User experience optimization without commercial value is impractical of. Accurate recommendations for takeaways that have not waited may not come so soon, or even at all.

This article was originally published by @完结. Everyone is a product manager. Reprinting without permission is prohibited.

The title picture comes from Unsplash, based on CC0 protocol

avatar