BitRecommendations — AI-driven decentralized recommendation system
How does loyalty relate to recommendations?
The recommendation system is essentially a part of the loyalty program because product recommendations and remuneration recommendations encourage people to purchase more.
It is impossible to build a good advisory system without a vast amount of data. And this is a problem for small and medium-sized businesses, because they do not have enough traffic and sales in order to independently accumulate the necessary amount of data.
DMP is not enough for e-commerce
Data management platforms, which aggregate data from various sources, are designed to provide the merchants with their users’ profile — social graph, interests, disposable income, etc. This information is useful but insufficient to make the high-quality forecasts in e-commerce, because the user’s habits and consumption patterns are hard to figure out, yet they contain crucial data for decision-making.
For instance, it’s impossible to understand what color of the device a customer will prefer, basing assumptions on the data about their age and attraction to a certain brand of cars.
BitRecommendations — Decentralized Crowdsourced Recommendations for Ecommerce Businesses
As a part of the BitRewards loyalty platform, we are creating the first crowdsourced recommendation system on the decentralized ledger technology. BitRecommendations is a decentralized DMP designed to collect, store and process the retail data using deep learning algorithms. The control of the quality and security of the data is provided for by the blockchain.
It works as follows: Every merchant is a source of huge amounts of data about their shoppers. 90% of this data remains unused. The best that happens with it is it gets sent to Google Analytics, where a marketing specialist will go through its aggregated copy.
BitRewards is rethinking this process. We will give merchants the solution, which allows them to seamlessly and frictionlessly have all the data about their users at their disposal:
- General data types (viewed items, added to order items, order amount);
- Industry-specific data types about the user’s interests (colours of garments, gadgets’ screen sizes, food calories, etc.).
All these data will be stored with a help of BitRecommendations system. There is nothing to configure for this, we will give a completely ready solution that works on any, even the weakest server.
And once a day the store will do the following operation (automatically, of course):
- Every 24 hours the data is encoded and sent into IFPS
- The merchant sends to the BitRecommendations smart-contract the IPFS-address containing the specific data and metadata (number of the unique per period, merchant ID)
- The merchant can send this data to the smart-contract of any other organisation, to which BitRewards DAO entrusted the private keys — it is necessary to avoid the dependence on the decentralized BitRecommendations software.
Upon the receipt of the date into the smart-contract, BitRecommendations runs several processes:
- Checks the merchant’s reputation. If the reputation is unreliable, the weight of the data is lower, or it’s ignored altogether.
- Downloads the data from IPFS on the BitRecommendations servers.
- Starts the lock-up period (2 weeks at the beginning of the operations) and processes the received data via the moderation subsystem. The purpose of the moderation is to determine whether of not the data is accurate and came from the reliable sources, and attach the appropriate weights to the data. To solve this problem, the deep learning and AI is applied.
- After the lock-up period, BitRewards credits the merchant with BitRewards tokens for the valuable data and reduces the rating and repayments in the case of a fake data.
- The accurate data is used for deep learning algorithms.
Machine Learning: product recommendations
To train the recommendation system BitRecommendations will be used:
- Data from third-party DMPs (Oracle Bluekai, Krux, MediaMath, etc.)
- Data on customer activity received from stores, which through the system described above shared with the entire network BitRecommendations.
Based on the specified data, using the methods of machine learning, the BitRecommendations API is created, accessible to all ecommerce business users. Pay for use of this API can only be token BIT.
It allows any store to get high-quality product recommendations from the first day of work, even at very low traffic — due to the use of impersonal data received from thousands of other stores.
Machine Learning: remuneration recommendations
Also BitRecommendations API allows any store to find out the preferences of its customers with regard to rewards. Someone will prefer coffee in Starbucks as a reward, someone will prefer to go to massage, and someone will want to get a ticket for the concert.
The system analyzes the rewards given out in the loyalty programs of different stores, aggregates the received data and predicts what one or another person will like to receive as a gift.
Machine Learning: financial scoring
Due to the fact that the purchasing activity of each visitor becomes available to all BitRecommendations (in anonymous form), the system is able to produce a financial scoring of each visitor.
This is used for the following scenarios:
- Offer goods with postpay — for those customers who always buy out items ordered from the online store
- Offer special loan terms and installments for goods — based on a credit history recorded on the blockchain
- Issue personal discounts to those customers who do not buy without a discount — and do not give out extra discounts to those who buy a lot and often in other stores
The use of the blockchain and crowdsourced-nature of the system make it possible to achieve a level of financial intelligence for retail that was previously unavailable for small and medium-sized businesses.
Using the BitRecommendations API for most stores will be free. The system works on the principle of “take as much as you gave”:
- The store receives BIT tokens for the data provided and gives them for using the BitRecommendations API. The relationship between price and reward will be built in such a way that the store uses the BitRecommendations API for free if it shares information about the activity of all its customers with the system.
- If the store does not want to share customer activity data with the system, it will have to pay BIT tokens for using the BitRecommendations API
Payment for the development and hosting of BitRecommendations is made from the reserve of the BitRewards team, as well as from the commission from all fees paid out within the BitRewards system. This commission will be introduced, according to the plans, not earlier than 1 year after the functioning of the system.
Application of blockchain and risk mitigation
Due to the use of decentralized technologies BitRecommendations system becomes completely independent of BitRewards company, as well as from dishonest participants of the system. At the same time, the system retains the openness properties — any ecommerce business from any country can connect to it and use the accumulated data of thousands of other stores.
- All data received from stores is stored on a decentralized IPFS file system. Metadata, including the rating of data quality, is stored on the Ethereum blockchain. This makes the system trustless — if BitRecommendations ceases to work, with the support of DAO BitRewards a new development team will be hired and a new system using the accumulated data will be created.
- If the store tries to trick the system and sends fake data — this will be detected, its reputation will be lowered, this information will be written to the Ethereum blockchain system and it will be permanently excluded from the users of the system.
- Due to the use of BIT Crypto-Currency, the system can provide high-quality recommendation service BitRecommendations API for bona fide participants of the system completely free of charge.
By virtue of BIT tokens, BitRecommendations can offer the cutting-edge data-driven recommendation service for valuable members of the system free of charge.