Five questions to answer before hiring an AI vendor

Five questions to answer before hiring an AI vendor

/in /by

Many enterprises that consider AI implementation don’t know how to go about it successfully. Companies complain that they don’t have the necessary in-house knowledge or skills, and building a team from scratch is too time-consuming and risky. Because the effective implementation of AI projects requires building interdisciplinary teams – both from business and technology –  executing the process successfully is quite a challenge.

The answer to such a challenge may be to combine in-house and outsourcing models. In one respect, an in-house team provides crucial know-how and knowledge about their company’s operations. An AI vendor, meanwhile, provides cutting-edge knowledge of AI technology and a flexible, innovative approach to realizing the use cases. Ultimately, the vendor’s team becomes an integrated part of the client’s team.  Sounds promising… but before starting cooperation some key questions should be answered.

Where is the business value?

The first element that needs to be decided is what business case the cooperation with the vendor should address. While enterprises are more and more certain about AI’s potential there is still a lot of hype around, making it difficult to decide which use case to implement first. It is not only about not losing money on an AI project, but above all about identifying such business processes where automation will bring visible business values.

The unique approach and work methodology developed by deepsense.ai allows us to quickly identify the most relevant use cases. Our cooperation with a client always begins with an opportunity discovery phase that helps map and understand the business needs. Joint ideation sessions define the main processes and data sources that can become part of the implementation. This approach makes it possible to efficiently determine the scope of cooperation and set up success metrics.

Who from my team will be involved in the project?

Effective implementation of AI projects requires building interdisciplinary teams. Ones that command extensive expertise in available technologies and approaches to data analysis, while at the same time possessing a thorough understanding of the business processes taking place in the enterprise. They must be able to recognize the factors that influence the core operations. As business challenges go, this one certainly qualifies as daunting. However, it is important to consider who – within the organization – will be able to become a partner for cooperation with an AI vendor. This person, or persons, should not only take care of the smooth development & implementation of the project, but also have a full picture of the business situation within the analyzed use case.

What kind of  data can we share with the vendor?

The garbage in/garbage out principle is well known in data science. But, it should also be considered from a business stakeholders perspective. Certainly, most enterprises possess data that can provide valuable insights. However, it is important to review whether the dataset covers all data sources available in the organization as well as assess data quality and consistency. Usually the process of collecting and preparing data for analysis – to the customers’ surprise – takes more time than it was expected. To speed up the process, the deepsense.ai team always begins to review and analyze the data in close cooperation with the customer’s domain experts. At this stage, it is crucial to discuss basic questions, and check features and labels.

An additional element worth paying attention to is the data security. While working with an external vendor the knowledge of core business processes somehow goes beyond the enterprise. That’s why we always implement adequate security measures, such as encryption, network isolation, and strict data access policies.

How much time do I give to see results?

Valuable data analysis requires patience. Often, before reaching the final solution, several models are trained simultaneously using different perspectives and approaches. The duration of the cooperation depends on the specifics of the project. Usually it takes several months from the data audit and development plan to the operational rollout and commercialization of the entire project. The first working proofs-of-concept are usually presented within 4 weeks from the start of the project.

In what model do I want to work with a vendor?

At deepsense.ai we have the flexibility to work along two different models. The first one is focused on team augmentation. We work in cooperation with our customer’s teams to expand their technical skill-set within a broad set of roles. Our data scientists, data, and software engineers are ready to tailor to any specific needs. Our people are equipped with the full tech stack required to help clients reach their milestones quicker, and ready to work hand-in-hand with internal business and technical teams.

The second model provides end-to-end project delivery: from identifying the needs to commercial deployment. We also specialize in handling highly customized challenges for which ready-made solutions don’t exist.

Summary

The dynamic development of AI makes it difficult for companies whose core operations are not related with hi-end technology to keep up with the latest solutions. Close cooperation with an AI vendor allows them to maximize the potential of state-of-the-art technologies and focus on the industry-related aspects of building competitive advantage. We might go so far as to say that truly remarkable results are not possible without combining in-house and outsourcing models. At least not without a great deal of time and much larger financial outlays. Cooperation with a reliable AI vendor ensures flexibility, cost control, and also brings a fresh perspective on the data being analyzed.

deepsense.ai for MIT Sloan Management Review about AI in service of balanced growth

/in /by

The pursuit of sustainable enterprise growth has accompanied the rising interest in optimizing the use of global resources. Companies know that the proper management of global resources today isn’t strictly a question of ethics, but can also have a very real impact on building competitiveness. Sustainable growth is becoming a leading trend in the business world, one that can be supported by AI and advanced data analysis.

In business, basing decisions on reliable, real data can be a key means to achieving sustainable growth. AI has made it possible to monitor and analyze huge amounts of data, from various sources, more or less in real time. This provides a full view of the current situation and enables flexible action.

In the pursuit of sustainable resource use, advanced algorithms can promote better decision-making and process optimization in numerous key areas including not only supply chains, transport, and production processes, but also building value for society and the environment.

Balanced supply chains and transport

Lack of control over the extensive global supply chains takes a toll on the environment and sustainable trade, and also costs companies money. Fortunately, AI has it possible to monitor and optimize, on an ongoing basis, all aspects of the acquisition and transport of resources. For its part, deepsense.ai helps its clients optimize their supply chains in a number of ways, including providing systems to help minimize empty mileage in road transport. Machine learning models analyze the location of transport loads by selecting the most optimal route, maximizing the load capacity of trucks and minimizing the number of kilometers they travel. AI not only increases transport efficiency, but also helps companies reduce their carbon footprint.

Optimizing production resources

The optimal use of production resources is becoming another key aspect of sustainability. It offers measurable benefits in the form of higher operational efficiency. Factories that have adopted a zero-waste policy – completely eliminating waste from production lines or their use in subsequent production phases – meet the demand for resource optimization in its most extensive form. But a sustainable approach to production doesn’t have to be so revolutionary in scope. The deepsense.ai team works with clients on various aspects of production and resource optimization through intelligent quality monitoring. One area of focus is real-time monitoring of production lines using a combination of computer vision technology and advanced data analysis. The systems analyze products for defects and impurities, appropriate shape and distribution and a range of other anomalies. This helps companies identify and minimize losses in the early stages of production while maximizing their resource use.

Boosting employee safety is yet another important aspect of sustainable production. In response to this challenge, deepsense.ai created a system for monitoring the safety of factory employees based on the analysis of camera images. Trained neural networks analyze the situation on an ongoing basis, identifying potential threats such as violations of security rules or the sudden emergence of new threats in high-danger zones. When a threat is detected, the system automatically sends alerts and suggests specific preventive actions.

Building value for society and the environment

Another way companies engage in sustainable development is by also means building a company’s value by contributing to society. Here advanced data analysis can be used in a number of ways. An interesting example is a challenge deepsense.ai took up in 2016 as part of an analytical competition announced by the US National Institute of Justice. The task involved predicting the location of possible crimes in the city of Portland, Oregon, by identifying crime hotspots – small areas with the highest predicted crime rate. Our team’s results were impressive enough to take first place.

AI-based systems can also be employed to help protect the environment. The image recognition-based early warning system deepsense.ai created for detecting potential forest fires is a good example of such a system. The machine learning model automatically recognizes smoke in photos from cameras placed in the forest and provides predictions about the locations of possible fires. The solution not only increases the accuracy of the early warning system for fire hazards, but also enables a much faster reaction from fire departments, potentially eliminating the threat at an early stage.

Summary

The use of advanced data analysis and artificial intelligence can help companies and other organizations maximize their achievement of sustainable development goals, build competitive advantage and create a better, safer world. The COVID-19 pandemic has highlighted how the lack of a sustainable approach can disrupt business continuity. The ideas that define sustainable development have gained in importance and will become a permanent feature of companies’ strategies. However, in the pursuit of optimization through the use of the latest technologies, striking a balance between the outcomes we anticipate and the environmental and economic costs of their implementation is crucial. Ever aware of the importance of this balance, deepsense.ai produces solutions that draw on the latest technologies and the optimal and agile model training process.

The article was published in the Polish edition of the MIT Sloan Management Review, in November 2021.

deepsense.ai for AI Trends about successful enterprise AI implementation

deepsense.ai for AI Trends about successful enterprise AI implementation

/in /by

There is no question that the enterprises that successfully leverage AI’s potential will be the ones to get ahead. While businesses are more and more certain about the goals they want to use AI for, how to execute and deploy it successfully remains a difficult question.

Enterprise AI – evolution not revolution

In today’s world, the dynamics and efficiency of operations are crucial factors, and data is the key to flourishing in the rapidly changing business reality. Data provides insights from various areas of business activity. Solutions based on artificial intelligence or machine learning not only support the analysis of huge amounts of data, but also provide a new approach to optimizing and automating core processes. Enterprises across the full spectrum of industries can build competitive advantage based on AI. Such a strategy need not come down to one-off AI implementations. Rather the most successful will take a holistic approach to treating data and high-end technologies as core organizational assets. From this perspective, enterprise AI can be considered as the next, natural stage in the digital transformation of enterprises.

New approach to AI team augmentation 

Many companies want to implement the enterprise AI approach, but don’t know how to go about it successfully. Companies complain that they don’t have the necessary in-house knowledge or skills, and building a team from scratch is too time-consuming and risky. In fact, the effective implementation of AI projects requires building interdisciplinary teams – ones that command extensive expertise in available technologies and approaches to data analysis, while at the same time possessing a thorough understanding of the business processes taking place in the enterprise. They must be able to recognize the factors that influence the core processes. As business challenges go, this one certainly qualifies as daunting.

The answer to such a challenge may be to combine in-house and outsourcing models. In one respect, an in-house team provides crucial know-how and knowledge about their company’s operations. The outsourcing partner, meanwhile, provides cutting-edge knowledge of AI technology and a flexible, innovative approach to realizing the use cases. Ultimately, the vendor’s team becomes an equal member of the client’s team. Such an approach seems to have a number of disadvantages, as the knowledge of core business processes somehow goes beyond the enterprise. However, the benefits of this approach far outweigh such disadvantages. It might go so far as to say that truly remarkable results are not possible without combining the in-house and outsourcing models. At least not without a great deal of time and much larger financial outlays. Cooperation with a reliable outsourcing partner ensures flexibility, cost control and it also brings a fresh perspective to the data being analyzed.

deepsense.ai’s holistic approach to enterprise AI transformation 

The unique approach and work methodology developed by deepsense.ai maximizes synergy. Our cooperation with the in-house client team always begins with workshops that help map and understand the business needs. Joint ideation sessions define the main processes, data sources and use cases that can become part of the implementation. The deepsense.ai team then begins to review and analyze the data in close cooperation with the customer’s team on a daily basis. This model was implemented in cooperation with our client, the retail research leader Nielsen. Working alongside Nielsen’s experts across the globe maximized the synergy between our two companies. Together we developed an advanced AI solution to automate extraction of content images of varying lighting conditions, viewing angle and quality. After implementing the solution we conducted technical workshops for Nielsen’s team to transfer the know-how and code. We also provided assistance in cloud readiness and scaling.

Summary

The dynamic development of technology makes it difficult for companies whose core operations are not related with hi-end technology to keep up with the latest solutions. Close cooperation with a technology partner allows them to maximize the potential of state-of-the-art technologies and focus on the industry-related aspects of building competitive advantage.

deepsense.ai for MIT Sloan Management Review about AI in platform businesses

/in /by

Businesses based on the platform model are gaining in popularity. Their potential is hidden not only in the territorially unlimited, multi-level network of value exchange. The platforms also provide enormous amounts of data on customer purchasing preferences, vendors’ sales opportunities and new market trends. In-depth analysis of this data allows for the proper implementation of a development strategy and facilitates scaling.

The key to fully exploiting the potential of platform businesses is to build a unique customer experience that translates into a growing base of loyal users.

Advanced analytics based on artificial intelligence and machine learning enables this unique customer experience through hyper-personalization in the most crucial areas including individualized offers and content, dynamic recommendations and service automation.

Hyper-personalization of sales and dynamic recommendations

A key consideration in building a competitive advantage in the platform model is sales personalization. However, customers expect more than traditional personalization based on historical data analysis.

To maintain customers’ interest, they must be provided with the highest quality interactions with the platform. This can be done by adjusting the offer to their individual purchasing preferences “here and now”. The availability of tools based on cognitive technologies and artificial intelligence provides a transition from basic personalization to hyper-personalization, enabling the real-time analysis of customer needs. An example of an innovative approach to personalized recommendations is the intelligent, virtual fitting room deepsense.ai developed. Customers can upload their photo to the application, which simulates their appearance in the garments of their choice. The system can also propose a specific size and adjust complementary styling elements. Multi-stage data analysis based on generative adversarial neural networks takes the customer to a new level of interaction with the offer.

Advanced machine learning models also make it possible to optimize  recommendation engines and transform recommendations into a personalized offer tailored to the needs of a specific customer.

Data analysis facilitates not only tracking customers during their journey to the next stages of the sales funnel, but also predicting a specific moment of conversion. This gives a business the opportunity to immediately respond to the client’s needs and propose a highly personalized offer. An example of this approach is present in an AI solution developed by deepsense.ai for one of the leading european banks, where, a machine learning model creates personalized offers for customers with an efficiency exceeding forty times the baseline approach. The algorithm automatically searches the database in order to identify customers matching a specific profile, individually selects the elements of the offer and analyzes customers’ readiness to buy.

Content hyper-personalization and intelligent search

Highly personalized content makes it possible to build a customer relationship with the platform and maintain customer loyalty, providing each user with a unique experience.

Leading brands like Facebook, Amazon, Spotify and Starbucks have long been taking content personalization to a new, higher level using predictive personalization. AI has made it possible to extract additional information about customer preferences in real time. Taking into account the relevant data, the models are able to predict what type of content a given client will most likely engage with.

Intelligent search also supports content hyper-personalization. Deep learning elevates search from a simple understanding of keywords to an understanding of intent and context.

Understanding the semantics and real meaning of user queries is essential to intelligent search. Neural networks learn to correctly interpret queries and adjust the information users expect. Such solutions facilitate the effective search for content, navigate information in platforms with complex structures and ensure a quick transition of customers to the issues they are interested in.  An example application of the above technology is a system created by deepsense.ai for a platform offering access to scientific literature. To equip users with efficient content search, natural language processing models that can recommend items with similar topics are applied. Thanks to this, the platform’s clients quickly find the materials they are interested in.

Process automation

Intelligent tools that automate the sales process keep the user engaged and support conversion. But it is not only about simple solutions for improving customer service.

Algorithms make it possible to efficiently guide the customer through the purchase process by providing dedicated information obtained in the learning process, based on the service of previous users. Additionally, by eliminating manual operations, they enable real-time customer service. An instance of such automation can be found in the deep neural networks deepsense.ai created to power an intelligent system for reporting and assessing motor insurance claims. Thanks to advanced image recognition algorithms, the system identifies car parts and classifies those that have been damaged. It then automatically assesses the damage and estimates the cost of repair. This solution helps estimate the value of the claim within a few seconds based on photo documentation sent by the customer.

Summary

Platform businesses are a response to the pandemic reality that has forced customers to push deeper into the world of online shopping. The platforms have developed explosively. Going forward, the dynamics of this development will be dictated by the speed with which platforms respond to individual customer needs and advances in data analysis.

The article was published in the Polish edition of the MIT Sloan Management Review, in June 2021.

AI-based applications in the insurance industry

AI-based applications in the insurance industry

/in /by

Innovative leaders from the insurance industry are starting to recognize the benefits of machine learning and deep learning applications. As business approaches to using AI-based solutions evolve, concerns about the transparency and explainability of AI models are also diminishing. While the trust in this type of solutions is growing, AI offers a new look at the product and service portfolio, as well as improve the sales process itself to best suit customer needs. As predicted by McKinsey experts, the technological revolution in the insurance industry has already started, and its progressive development will lead to a massive technological shift over the next decade.1

AI-based customer service automation

One area where insurance leaders look to implement AI solutions is customer service automation. Chatbots and biometric identification are becoming permanent features of the customer service landscape.
Chatbots automate customer service, increasing customer experience and satisfaction. Behind the scenes, automation significantly reduces customer service workloads. As insurance industry leaders have indicated, over 90% of customer service engagements follow basic patterns, leaving excellent potential for automation while retaining high customer satisfaction.
Biometric identification supports explicit or unnoticed identity verification within remote channels. This can include voice identity verification in call centers or typing manner verification in online channels. Similar solutions are poised to gain traction, particularly in the post-covid world.

Back-office optimization

Another area where AI-based applications are highly effective is back-office optimization.  The documents and forms processed in the insurance industry are voluminous, so improvements in this area quickly bring noticeable results. Given that, deepsense.ai developed an insurance claims assessment system for one of the largest financial institutions in Eastern Europe.  Employing advanced computer vision algorithms, the system recognizes car parts and classifies which are broken, assesses the severity of the damage and estimates the cost of repairs. It allowed nearly real-time assessment of claim value based on image documentation sent by the client. With 70% of claims handled automatically, the manual work required in the process is significantly reduced, while the time needed to assess damage falls from days to seconds.

AI-based applications in the insurance industry - deepsense.ai claims assessment system

deepsense.ai claims assessment system

Automating text and image documentation analysis can be widely applied in insurance, including for cash operations, trade finance, insurance application processing as well as classifying incoming emails to go to the appropriate department or customer segmentation.

Customer insights

Applying deep learning to customer analytics makes it easier to combine insights from various data sources (e.g. transactions, online banking logs, call center interactions) as well as a 360o customer view and personalized sales activities. This helps to better understand customers and to build personalized recommendations, making the business more responsive and efficient. For one client in the banking sector, deepsense.ai designed a machine learning model that created personalized recommendations, showing what should be proposed to a given client based on the type of customer they are. The model had a 93% recall rate, which is 60% more effective than traditional product recommendation techniques.
Thanks to accurate AI algorithms, we can also improve customer retention by predicting churn probability. This is important as customers often churn without obvious warning signs. The roots of deepsense.ai’s experience with anti-churn use cases go back to the banking industry. We developed an intelligent anti-churn system based on historical internal data and external databases for a leading financial institution in Central Europe. To identify risk groups that were more likely to churn, we leveraged multiple data sources and gradient boosting trees and also evaluated customer lifetime value. The results were used to prioritize communication with highlighted clients, 50% of which confirmed they were considering leaving the bank.

Insurance risk management

Machine learning enables nearly fully automated fraud detection, adapted to individual patterns and changing behaviors. It can be applied in areas where a large volume of events needs to be analyzed in real time. Our client, a leading CEE insurance company, suspected that some end customers were abusing access to private healthcare. deepsense.ai was tasked with analyzing data in a search for anomalies and spotting the data-marks of fraudulent transactions. With the knowledge it gathered, the team developed algorithms identifying common schemes and techniques of private healthcare abuse. The schemes included excessive medical diagnostics and exploiting flaws in billing systems. The team also identified potential fraud being committed by service providers abusing their agreement with the health insurance company. The model we delivered spots suspicious activities and has enabled the company to reduce losses by over 3M EUR annually.

Summary

A wide range of ML applications is increasingly being used to solve real business problems in insurance. As AI becomes more popular, these applications will become the market standard. Nevertheless successfully leveraging the new opportunities offered by cutting-edge technologies will require insurers to undertake a shift in corporate governance. However, if applied successfully, AI-based solutions will benefit both insurers and clients alike.

1 https://www.mckinsey.com/industries/financial-services/our-insights/insurance-2030-the-impact-of-ai-on-the-future-of-insurance

5 ways how AI can improve your demand forecasting

5 ways how AI can improve your demand forecasting

/in /by

Download the guide and learn how deepsense.ai can leverage the accuracy of your organization’s demand forecasting.

DOWNLOAD GUIDE

 

5 ways how AI can improve your demand forecasting

AI trends for 2021

AI trends for 2021

/in , /by

Interview with Michał Iwanowski, VP of Engineering, about AI development in 2021.

How do you rate 2020’s AI development?

AI is no longer a buzzword. Increasingly, from the R&D phase, it moves towards proven solutions, and nowhere more so than in computer vision and natural language processing.

The computer vision projects deepsense.ai delivered in 2020 prove that numerous industries can benefit from advanced image analysis technology. We have implemented successful projects in modern manufacturing that incorporated machine learning throughout the production process. These included not only intelligent quality control systems, but also safety monitoring and workplace security solutions. Our 2021 projects will develop this potential based on AIoT systems.

NLP-based solutions are likewise becoming more widely applicable. Our portfolio has been enriched by projects ranging from virtual meeting assistants to solutions capable of summarizing large chunks of text and evaluating the quality of student submissions in selected areas of study. With the rise of enormous language models such as GPT-3, we expect to see this area of AI –  including natural language understanding (NLU) – grow more robustly in the coming quarters.

Where can we look for indicators of AI trends for 2021?

Top global research and consulting companies have indicated that 2021 is to be intensive and promising for AI. The emerging trends include the democratization of AI and industrialization of AI platforms. More and more businesses are seeing the real value in analyzing data and are convinced that using AI based on platforms that enable scalability and safety is a wise step forward.

The Covid-19 pandemic appears to have accelerated customers’ openness to AI. Despite the crisis, clients have continued to invest in the technology. They understand that in order to build competitive advantage, AI must become a core element of their business models. With more than 70 data science and machine learning experts on board, we are ready to turn clients’ vision into practical AI-focused solutions.

Do you notice any changes in your customers’ approach to AI solutions?

We work mainly with clients who already have some knowledge of and experience with AI projects. This allows us to act creatively on new AI solutions for various businesses.  When a client retains us for a single project, we frequently uncover several new areas where we can further develop and implement AI.

We highly value clients who trust us enough to jointly take on challenges related to innovative AI applications. One such client is Volkswagen, with whom over the past 18 months we have moved neural network policy from a car simulator to reality. After multiple training sessions in the simulated environment, our algorithm overcame the  sim-to-real gap in the real world. This milestone has opened up the possibility of using distributed computing to train AI models that are easily transferable to physical vehicles and can be tested in real world-environments.

We have also observed increased interest in our R&D projects. Many clients want to get involved in our research work because it leads them to explore new opportunities for their businesses. After a successful reinforcement learning project conducted together with Google Brain, the University of Warsaw and the University of Illinois at Urbana-Champaign, we continue our work in this area. The research conclusions bring tangible benefits to our clients by setting solid algorithmic foundations for tackling business problems that don’t present an obvious objective function, but rather require an AI agent to learn from its mistakes. This paradigm shift opens up brand new areas of application for AI.

Last but not least, we see many clients that are ready to build AI capabilities in-house. With our comprehensive training programs, they get best-in-class know-how that bears long-term results for real-life data challenges.

How do you sum up deepsense.ai’s plans for 2021 ?

Apart from pursuing new projects and challenges in computer vision, natural language processing and predictive analytics, there are two crucial tasks ahead of us in the upcoming year.

The first is to maintain an open-minded and creative approach to the changing reality of remote work and the migration of more and more services to the virtual environment – with a focus on how these transitions can be converted into business opportunities using AI and automation.

The second challenge is to bridge the gap between cutting-edge advancements on the frontier of AI and Machine Learning and reliable, well-tested business solutions. As a company that operates at the crossroads of business and research, we believe it’s on us to take a pioneering role in this challenge.