SAS: we analyzed data and trained models long before it became fashionable

Our analytical platform operates in WalMart, Bank of America, Bank of China, Sberbank, MTS. SAS as a subject is taught at Moscow State University, HSE, MEPI, MSTU. Bauman, MEI, MIIT and other universities. And under the cut - our brief history, acquaintance, with which we want to open our blog on Habré.

Who we are

SAS has existed since 1976. We grew out of a small project by a young mathematics professor from the University of North Carolina. It all started with small contracts for the statistical processing of data of the Ministry of Agriculture, which he carried out together with his students.

Of course, standard automation solutions did not exist at that time; therefore, the professor himself wrote most of the statistical functions in C and COBOL.
')
At some point, the quantity turned into quality: instead of just doing certain calculations for its clients, the professor decided to sell his own calculations and other clients who needed to work with analytics and statistics, and build mathematical models. This is how SAS appeared.

Today we work around the world in almost all traditional industries where statistical analysis is needed. Our partners include banks and major insurance companies, retailers and manufacturing companies, energy and oil and gas, restaurant and hotel chains, as well as a variety of government agencies. Today, we serve more than 83,000 customers worldwide. Our company has 14,000 employees, more than 4,000 of whom are directly involved in software development.

We have a rich history of work in the Russian market. Although the official history of SAS in Russia began in 1996, the first major implementations of our software belong to the very beginning of the 90s, and individual solutions worked even during the existence of the USSR

Alfa-Bank became one of the first major clients on the Russian market, the history of work with which goes to the very beginning of the 90s. Among the largest Russian clients, we can mention the company MTS, where we created a data warehouse and management and analytical reporting processing system. The total storage capacity was 30 TB, which at that time (more than 10 years ago) was the largest data repository in the Russian Federation, and possibly even in Europe. Also, our technologies and solutions are actively used in the field of railway transport and some other sectors of the Russian industry.

Why are we here

One of the goals of our presence at Habré is to meet young people, dispel myths about us, including the main one, about the unavailability of SAS due to high costs and other restrictions. No, we do not have everything paid - you can always find options for free use, including for research and study; No, we do not have everything inaccessible - there are plenty of resources in the public domain; No, we do not have everything in English - and we will definitely work on increasing the Russian-language material.

What markets do we work in?

We are actively working in the banking market, our clients include almost all the largest banks. Our solutions are used in client analytics, targeted marketing, help ensure data management and prepare management and analytical reporting. One of the key areas is risk management, which is also used to combat fraud.

Despite the initial skepticism in the industry, we are successfully working with the insurance business. Their main problem is a very small amount of transactional data on clients. A normal person buys a policy and comes only after a year after the next. Therefore, insurers doubted whether it would be possible to extract some benefit from machine learning technologies. But a drop wears away a stone. A couple of years ago, the dam broke through, and we started to make the first projects. The most promising areas for us are the fight against fraud and the assessment of the potential loss of customers.



In recent years, we have been actively entering the retail market. In this industry, our solutions are used both in commodity analytics (optimization of prices, stocks, placement on shelves, etc.) and in client analytics (everything related to the personalization of customer relations). Also, analytics provide real and quick effect in sectors such as logistics, medicine and agriculture.

Today, the needs of customers and the market in analytics are developing extremely rapidly, so we estimate our prospects with great optimism.

Key requirements

However, not all so simple. The efficiency of working with analytics depends on several key factors, the absence of which can ruin the whole effect.

First, for analytics, first of all, it is necessary to have adequate and ordered data. In some industries, this is not difficult (banks, telecom), but in some, structuredness, and simply having the right data is a big problem (for example, insurance, agribusiness).



Secondly, an important role is played by the extent to which management is ready to introduce analytics and use it when making decisions, and the main problem lies not even in costs, but in the readiness to change the schemes of work. There are many places where management believes that if their schemes have been working for twenty years, and experts used to make the necessary decisions without any analytics, then there is no cost to change. In the modern world, this strategy works either until the first major incident (fraud, crop failure, etc.), or until the company begins to noticeably lose on the market to competitors using modern technologies and management schemes.

Thirdly, the company should have a team of analysts who will work with the obtained data. Such teams, too, are not everywhere, and in telecom analytics, mathematics, data scientists will be faster than, for example, in agriculture. However, everything changes here: on the one hand, more and more enterprises realize that it is necessary to have their own analytical unit, on the other hand, the use of artificial intelligence allows analytic systems to work more accurately and better adapt to existing conditions, which increases efficiency in specific tasks. Plus systems have become easier for the user.



Finally, outsourcing is now widely spread, when the actual work of analysts is assumed by the partners, and the customer receives analytic data ready for understanding and use. For example, often at the beginning of cooperation we work as external analysts (according to the RaaS model), and the client gets an effect and an understanding of what works and how, which areas should be developed and which ones should not. This will help him in the formation of his team, or he will be prompted to make a decision to continue working at outsourcing.

What are we doing

Our analytical platform has a very wide functionality. Its capabilities include exploration analysis, data preparation, classical predictive modeling and machine learning, time series prediction, optimization, and much more.

All the blocks and solutions we develop ourselves. Therefore, we understand what and how it works and how it interacts with each other - we have no difficulty in coordinating the work of the various components and solutions among themselves. In this case, the blocks operate on a single metadata management platform and are based on a common programming language SAS Base.

One of our priorities is to integrate our solutions into business processes and decision making schemes in the company. The fact is that if analytics and data exist separately, “in a vacuum,” the effectiveness of their use drops significantly. In addition, without normal seamless integration, the risk of operational errors is very high. And such errors are very much hitting the confidence in analytics and models.


To embed analytics in business processes in SAS, a whole stack of technologies is provided. Data Level Integration (SAS Data Integration), event stream level integration (SAS Event Stream Processing), solution level query integration (SAS Decision Manager), model life cycle management level integration (SAS Model Manager), heterogeneous integration analytical tools such as R, Python, Scala (SAS Viya). In a large organization, whether it is a bank, retailer, telecom or something else, the main difficulty is a huge number of heterogeneous data sources on different platforms and DBMS, and a large number of processes that require the use of analytics (consumers of analytics) implemented in different systems .

Of the new priorities worth mentioning speed. Today, many of our customers are not enough to receive data and analytics with a delay. Decision time is shrinking, and in many cases data is required in real time.

Naturally, today SAS works not only on the traditional model of providing software, but also provides cloud services. In a variety of formats: SaaS (software as a service), BaaS (business as a service - outsourcing analytical processes), RaaS (result as a service - the implementation of a finished product for a turnkey client, from the development of a predictive model to the formation of a statistically sound strategy development of a network of retail outlets, for example).

Cloud services are no longer in demand in banks, where everything is good with data, with money, with analysts, and in other industries - insurance, retail, agricultural sector, for example. Clients in these industries are ready to attract external experts not only at the stage of adjustment, but also for permanent work. Another plus of cloud services is the absence of the need for large investments at the start of work, which gives a faster return on payback and reduces the risk of losses.

How to choose

SAS solutions are chosen by large enterprises for serious work, so choosing a solution takes a lot of time and includes an assessment of the wide range of possibilities and parameters of the proposed solutions. At the same time, the key factor for them is an increase in efficiency, which, like the accuracy of decisions made, largely depends on the accuracy of the model used. Sometimes the customer even arranges a kind of competition - he offers several vendors to build the analytical model he needs and looks at who will better cope with the task.



However, the accuracy of the model itself is not a constant factor. It depends on many factors, both related to the model and external. For example, the accuracy of the model depends entirely on the correctness and adequacy of the data provided in the framework of testing. Sometimes they are there, sometimes the necessary data is in a scattered form and they need to be searched for, collected and brought to a single format, or it is generally necessary to independently organize their collection.

Also at the first stage it is not always clear what is what, what are the features and most importantly what exactly the client needs. Therefore, in the early stages of work, the averaged models, which are already out of the box and provide relatively high accuracy of work, often get the advantage of accuracy. However, such a model with time will most likely lose its relevance due to the fact that the working conditions of the enterprise and the market itself are changing. This must be taken into account and either manually or automatically adapt the model to changing conditions.

With proper organization of work (when the model adjusts to the characteristics of the client and the market), the model’s accuracy at first will increase as the “fine tuning”, data accumulation, etc. In one of the implementations, our model initially performed worse than its competitors, however In the shortest possible time, having gathered the necessary information and having adapted its work, we took the lead.

We are confident in our decisions, and the statistics that we have shows that clients also trust us. The outflow of clients in our company does not exceed 1%; very often managers and clients, when they transfer to a new company, also seek to transfer it to the use of our solutions. In Russia, we are growing by 40% per year, which gives us reason to believe that our solutions are in demand.

What attracts customers

Most of all, customers appreciate what they get from SAS turnkey solutions. Our solutions allow us not only to develop models - we can integrate them into the company's business processes. Very often this possibility becomes decisive when deciding on the use of our product. Because you can make money on any models or analytics only when they really provide an opportunity to make profitable business decisions and increase efficiency. Therefore, we pay great attention to ensuring that our analytics is closely integrated into the company's business processes. But this is not our only dignity.

We ourselves develop all our products and solutions, which ensures uniformity and coordinated work, in the future our solutions can be easily scaled or modified. Our solutions are tightly integrated with each other: we offer a common platform on which you can “hang” different modules. Due to this, the degree of risk in their implementation is much lower - the client does not need to assemble into a single system from other people's technologies and tools with an unknown result. You can immediately (and often before a full-fledged implementation) evaluate the work of the solution, make a preliminary impression.

Our solutions incorporate the knowledge and experience we have accumulated in the field of solving specific analytical problems, typical business processes, forms of monitoring and reporting, etc. Plus, we do not have “just analysts”, our specialists have subject data in a specific area, which allows them to better understand the situation. In principle, even technical specialists have the knowledge and understanding of the processes taking place in the industry, which allows them to work more adequately with the data.

Another point - the speed of data processing. Everyone speaks about high speed, but more often we are talking about some special benchmarks. We have the experience of real implementations of complex solutions with huge processing speeds for companies such as WalMart, Bank of America, Bank of China, Sberbank. In addition, we have a special service: a special data center with 114 high-performance processor cores, 3 TB of RAM, 24 TB SSD capacity, which is mounted in a mobile box weighing 120 kg. We can deliver it to the customer for a couple of weeks so that he can check and evaluate his work. Yes, and often customers do not want to give it back.



Finally, it is worth noting that, although SAS products cost a lot, for our customers the prices of specific solutions, as a rule, do not play a decisive role - they look more at what benefits they will receive from implementation. This can be an increase in profitability, a reduction in costs, or an increase in response speed, which also increases efficiency. However, if the price seems unnecessary or the customer does not yet understand whether it will be beneficial for him to use our solutions, you can start with cloud services on the SAS platform - the threshold for entry is much lower for them, but you can fully appreciate our capabilities.

---

Already now, leading enterprises in many industries are working with SAS solutions. SAS teach in the largest universities - special courses, where anyone can come, are at Moscow State University, the Higher School of Economics, MEPI, Moscow State Technical University. Bauman, MEI, MIIT, and others. We regularly recruit for the trainee program, and it is extended to our customers - everyone needs people with SAS knowledge.

We will be happy to hear your wishes about topics that we should better share. Write questions in the comments, we will be happy to answer, including in future materials.