The data science market is highly competitive, fueled by ever-growing automation. The solution to all standard tasks is gradually being packaged into ready-made software solutions and services. And the ultimate business value comes from having seamless, seamless platforms.
In other words, instead of competing with a tractor with a shovel in hand, it is better to use the net for the tractor.
Stages of using analytical tools
Below, for each stage, literally several examples will be considered. But these examples are enough to understand the essence of the offers on the market today.
Stage 1. Choosing a product niche
1) Yandex.Wordstat
Statistics of queries in the Yandex search engine by keywords. Dependence on the region and seasonality.
2) Google Trends Google
search query statistics. Geography and seasonality of requests.
3) similarweb.com, pr-cy.ru
Website traffic of potential competitors in dynamics. Distribution by traffic sources (search engines, social networks, referral links, etc.). Popular queries in search engines leading to the site. Keywords in search engines, which are used for referrals.
4) mpstats.io
Monitoring of Wildberries and Ozon marketplaces: sales volumes, prices, stock balances. Indicators are available over time.
2. ,
1) Mindbox
Platform: personal recommendations through machine learning (best offers, analogues, related products), segmented distribution of offers, the possibility of manual customization of recommendations and advertising campaigns.
2) Yandex recommendation widget
Shows the most relevant articles from your site, thereby increasing the depth of viewing and the time spent by the visitor on the site. A useful tool for increasing conversions and search engine optimization.
3) 1C-Bitrix BigData
Product recommendations for a store visitor based on his behavior on your own site or based on the analysis of user behavior on third-party sites.
4) Yandex.Metrica, Google Analytics
Web analytics, including conversion analysis in various customer acquisition channels.
5) retailrocket.ru
Personalization of the site and mailings in real time.
Stage 3. Optimization of the product range and warehouse stock
1) Navicon S&OP
Business Process Management System. Sales planning based on demand forecasting using machine learning algorithms.
2) Business Scanner
Sales forecasting using predictive analytics. Allows you to optimize inventory management.
3) Visiology
Analytical platform. Sales forecasting.
Where is machine learning hidden?
All machine learning services on the market solve two types of business problems:
- Forecasting sales at the customer level
What is the probability of a specific product being sold, depending on the customer's purchase history on this site or other third-party sites, the context of the request, the color of the button on the site or the channel from which the customer came. Under the hood, we solve the problems of binary classification, collaborative filtering, or calculating the similarity with already purchased goods (content-based recommendations). - Forecasting sales at the product and store level
Forecasting time series.
Thus, either an event with two outcomes (buy, not buy) or a time series is predicted. The prediction of a binary event, in turn, is reduced to either the standard problem of binary classification, or to the problem of collaborative filtering.
The training of predictive models, most likely, takes place on the data of the platform that provides the service. After the model has been trained, the service is embedded in the online store and calculates predictions based on both the platform's data and sales data and user behavior in the online store. Subsequently, additional training or updating of the model on new data is possible.