Show everything that is hidden: from what AI costs are built and how to reduce them

Main article: Artificial intelligence

The introduction of artificial intelligence in business processes is one of the main trends in the global technology market over the next few years. According to the IDC^[1], companies' artificial intelligence spending will double and reach $110 billion by 2024.

Interest in artificial intelligence is also shown the Russian IT by giant. Based on the "Roadmap for the development end-to-end digital technology of 'Neurotechnology and artificial intelligence'"^[2]2024 Russian AI Solutions Market will amount to 160 billion, rubles that is, in just a few years it will increase 76 times. By 2030, about a thousand organizations and more than 10 thousand people will work in the area related to AI.

According to forecasts McKinsey^[3]. The vast majority of^[4] organizations plan to invest in artificial^[5] (88%). Business is invested in AI not only in order to achieve breakthrough results in the industry, to get a special technological product, but also to simply not lag behind its competitors.

In this article, we will tell you what the cost of AI solutions is, we will understand non-obvious spending, and together with experts and developers we will find ways to reduce them.

How AI is used. Examples from foreign practice

The world's largest corporations have long used AI solutions to optimize the costs of their business and find effective ways to advance in the market. On average, about half of^[6] spend $51 thousand a year on the introduction of artificial intelligence technologies, another 13% spend from $251 to 500 thousand on this, 5% budget for AI exceeds $5

One of the leaders in investing artificial intelligence the Chinese Alibaba^[7] predictive analytics retailer predicts which next purchase the consumer will come for, and thanks to natural language processing technology () NLP automatically generates a description of goods on sites.

"We put everything on AI" - the slogan of the owner of WeChat - the company Tencent^[8]. An application with an audience of more than 1 billion users analyzes a huge amount of data. Speech recognition, NLP, and computer vision help improve the quality of service and increase the number of loyal customers of the corporation.

Facebook^[9] using a suite of AI technologies to improve the performance of its ecosystem. DeepText allows you to automatically analyze and interpret the content and emotional color of thousands of posts that are published every second on the social network. DeepFace helps recognize people in a photo with an accuracy of 97%, and can also identify and remove unacceptable content from the network.

What AI products Russian companies spend on budgets

In the Russian market, big business spends the most on AI: banks, retail, oil and gas, telecom. During the digital transformation, companies, as a rule, solve three main problems:

• Improving business operational efficiency and transforming production processes.
  What AI solutions are used: predictive analytics and forecasting - for example, hardware failures and breakdowns, load planning, resource requirements, etc.
• Improving customer experience - understanding the needs of consumers, increasing customer satisfaction, increasing revenue, creating and developing client services.
  What AI solutions are used: chat bots, voice assistants, intelligent call centers, advisory support and cross-sales, analysis of customer behavior and preferences, interactive types of communication with the consumer, omnichannel.
• Creating and implementing a culture of continuous improvement and transformation - in other words, creating an environment for the development of the right competencies, managing innovation and digital transformation, forming digital assets, launching new businesses and services, making decisions based on data and intelligent services.
  What AI solutions are used: decision support systems, the introduction of intelligent services into the work of major and auxiliary departments, including analytical and engineering centers, R&D, project offices, legal, financial and HR services. These products are especially in demand by large companies that strive for intellectualization and implement a digital transformation strategy.

How much it costs to implement AI solutions

The price of AI implementation for companies can vary. First of all, the cost depends on whether the company uses the finished software or the product is developed for the tasks of a particular organization. In addition, the price is affected by the functionality of the system - the wider the set of functions, the more expensive the solution for the company.

In general, the costs of developing AI solutions can be divided into three groups:

• Specialist costs. Employees' salaries eat up a significant part of the project budget. AI solutions are always a complex product and the team usually includes a large number of experts. These include project managers, IT architects, business analysts, UI designers, data scientists, testers, DevOps engineers, etc.
• Infrastructure costs. Rent, servers cost, telecommunications purchase of the necessary equipment for testing is another important expense item. At the same time, part of the money can be saved - if, for example, you use servers, not physical ones cloudy.
• Support costs. ON The use of,, API operating systems paid - databases these costs are usually also included in the cost of the product. These items are not overly costly, but also contribute to an increase in the budget for the development of AI solutions.
• Ensuring information security. Where there are large data sets, system security is needed. When implementing solutions of this class, large budgets are spent on adapting the architecture and infrastructure to the requirements of information security - sometimes comparable or even more than the cost of the solution itself.

Hidden AI Costs

In addition to the costs of specialists and infrastructure that the customer understands, the cost of AI solutions includes many other factors, which, as a rule, are known only by the product developers themselves. The peculiarity of creating such solutions lies in working with huge arrays of information - aspects such as the quality and completeness of data can affect the final cost no less than the labor-intensive work to integrate the finished solution into the current business processes of the organization.

So, among the hidden costs may be:

• Receiving and preparing data. The success of the future solution directly depends on the completeness and quality of the data - the more competent the information is collected and processed, the faster and better the product can be implemented. At the same time, data research and preparation can take data science specialists up to 80 percent of the time.
• Data markup is one of the most costly items. Basically, this operation continues to be carried out by human forces and, as a rule, includes several iterations after analyzing the results. The correctness of the AI solution will largely depend on how high the markup was made.
• Search and select the models that will best solve the problem. Developers try different methods until they achieve a better result, and this process can be quite laborious. As of May 2021, there are many ready-made algorithms, approaches and libraries, but the work on their additional training and configuration remains quite painstaking.
• AI interpretability control. Sometimes it is important for the client to see not only the result of the system, but also to understand why it made a particular decision. In such cases, a special analytical module is created to visualize the 'progress of thought' of AI.
• Development of the user side. AI solutions can be embedded both in other systems - then it is necessary to ensure seamless integration of the product (data exchange in the desired formats, quality, periodicity, etc.), and represent a complete solution. In the second case, the costs include the design and implementation of the client part of the system.
• Support for machine learning models after solution implementation.
• Refinement of systems - for example, in terms of functions not provided at the stage of formulating system requirements.

How to reduce AI costs

There are several opportunities to reduce the cost of creating and implementing AI solutions. Here are the most effective ones:

• Hire specialists wisely. As a rule, when developing AI solutions, a large business either forms its own development team, or attracts a third-party contractor. The second option allows you to create the product faster. However, not all development companies have real experience in creating artificial intelligence systems - when choosing a contractor, pay attention to the competence of the team and the portfolio of implemented cases.
• Work with the data correctly. Try to collect and transfer the largest amount of information to the developer. It is important to share the most representative data rather than the ideal sample of well-structured data. This will help to avoid a popular problem when the solution works well on the demo, and when working with real data it gives an unsuitable result. Also be prepared to mark part of the data - only you have expertise in your subject area and can perform this operation most correctly.
• Correctly define the goals and tasks of the created AI solution. Describe in detail what the developed system will be used for, how the processes in which the product is implemented are arranged. Evaluate what pains AI must solve in the first place, as well as what development the system will receive in the future - lay this in the capabilities of the designed solution. The contractor's understanding of your tasks and needs will allow you to achieve the maximum effect of the product being created.
• Use the computing power of machines optimally. Above, we talked about reducing costs by migrating to cloud servers. This practice in Russia is not yet applicable everywhere, while foreign ones actively use distributed data centers located in various countries, which allows you to optimize the cost of IT infrastructure. The ideal option is to hire an experienced DevOps specialist: in the future, this will save more money on infrastructure and circuit security.

Let's summarize

If you decide to consciously make friends with AI and develop evolutionarily in this direction, then it is worth determining for yourself which of the zones of digital transformation will be most susceptible to change or will receive the maximum effect from the introduction of digital solutions - production or operational efficiency, customer experience, transformation office and transformation culture.

Data is a valuable asset. For a reason, digital asset management issues relate to the competencies of top managers - CDO or even CEO. It's great that today companies are learning to work with data: extract, accumulate, process and apply it in their activities. It is important that customers take an active part in AI-related projects, because in the future such tools will definitely allow companies to increase new competencies and bypass competitors in the growing digital race.

The author is Aikanysh Orozbaeva, head of the department for working with partners and customers.

Notes