As a highly competitive, dynamic and data rich industry, telco is a perfect setting for data science applications. A mix of socio-demographic, behavioral and technical data (e.g. QoS indicators) makes the telco industry both challenging and exciting.
Although it is much more expensive to acquire new than to retain an existing customer and operators are thus focusing more on customer churn and retention, churn rates on the industry levels are nevertheless rising. The goal is to selectively pick customers or customer segments which are of substantial value to the operator and in line with their portfolio management goals and overall strategy. Apart from the question whom the operator should target for acquisition, there is also the question of how (which channel), by whom (which internal unit or external partner), when and at what cost.
Customer acquisition models enable operators to distinguish between high and low worth customers, customers with different behaviors, lifecycles, etc. Operators which are not able to do so are thus victims of adverse selection and are left with low value and/or risky clients. By acquiring customers with more precision, the operator can align its customer base, strategy and portfolio management goals. Customer acquisition is the first step in transforming from a product oriented to a customer oriented business.
One of the biggest differentiators and sources of competitive advantage in the telecom market today is the operator’s ability to tailor its offerings, price, quality of service, etc. to customers based on their value. The concept of customer value however is much more complex that an ordinary ARPU measure. The goal is to build a model which will provide customer lifetime value on a customer and segment level and thus drive the business towards more profitable customer segments.
Differentiating between customer base on their lifetime value provides huge direct and indirect benefits to the operator. Acquisition is improved by comparing CLV and acquisition costs precisely targeting high value segments in mass marketing campaigns. CLV is also a key input required for effective retention program. Churn models answer the question how likely a given customer is to churn, but CLV has the answer on which clients should we focus and try to save. Knowing true customer value, greatly improves the quality of decisions. Questions like should we spend a particular amount of money on a marketing campaign or invest in customer satisfaction, become much easier to answer.
Highly competitive telecom market requires an individual approach in practically all customer related process: marketing, acquisition, pricing, revenue management, proposition, etc. Customer segmentation allows operators to divide its heterogeneous customer base into distinct sets of smaller homogeneous markets. The goal is to segment customers according to their behavior and lifecycle. Value and churn segmentation are treated separately.
Customer segmentation is the starting point for almost all processes involving customers. It is the foundation for the individual approach. Acquisition for example can be driven by identifying behavioral segments of high worth customers. Product design, pricing and revenue management can be highly optimized by analyzing behavioral and lifecycle segmentation.
Cross and up selling and event-driven marketing (e.g. activation of value added services, new phone purchase, change in usage) are very hard to do properly without behavior segmentation. Proactive offerings can be made much more individual if based on customer segmentation (e.g. new smartphone owners, streaming media offerings, etc.). By analyzing customer migration between segments, behavior segmentation is also useful for strategic planning – operator gets a clear picture of where the market is heading.
There are three main factors for customer profitability for any telecom operator: acquisition, margin and retention. In this area we deal with increasing customer margin. Cross and up selling is becoming harder each day with increasingly complex products, pricing, new value added services, etc. Timing and context are of extreme importance in cross and up selling. The goal is to build a model which will enable the operator to offer the best product, at best time, through the best channel to maximize revenue and customer lifetime value. The ultimate goal is to change from passively offering to proactively monitoring and analyzing customer behavior and upsell and cross-sell by anticipating customer needs.
The obvious benefit of cross and up selling is the increased revenue and profit, but it doesn’t end there. Cross and up selling also increases the dependence of the customer on the operator and reduces churn. By offering products and services that are best suited to a particular customer at particular time, the customer feels more appreciated and less annoyed by offerings that are of no interest to him.
Faced with increasing competitions, declining ARPU, lower margins and other issues, product packaging, pricing and revenue (yield) management becomes one the basic tools telecom operators use to keep and grow revenue and remain profitable. The goal is to build a model which, by determining the optimal product package and price for each customer segment, maximizes revenue and helps the operator achieve other, non financial goals.
Increased revenue and profit are main sources of values, but there are numerous other benefits. Pricing and revenue management also provides value to other customer related processes. Acquisition benefits by offering discounts to high worth leads and by the same token retention benefits by offering perks to existing high worth customers. New markets can be won over with targeted tariffs – for example, by attracting price-sensitive customers without eroding prices and revenue of other users.
It is a well known fact that acquisition cost is much higher than retention cost for an average customer, so operators should focus attention to retention. However, churn rates on the industry level are rising. The goal is to build a churn prediction model, both for forced and unforced churn.
Even small reductions in churn rate result in significant increase in operator’s profitability. By focusing its effort on customers who are both of high value and have a high probability of churn, the churn model serves as a guard of the most profitable customer segments. Apart from obvious financial value, churn prediction also provides strategic benefits. By analyzing churn rates and trends on a segment level, the operator gets a valuable insight into market movements, trends and competitors behavior and strategy.
Telecom operators are exposed to increasing credit risk from consumption of core services, financing mobile devices within a long term contract and usage of value added services. Credit scoring helps in prescreening, approval, pricing, account management and other processes in order to prevent losses and enable a more individual approach in customer management. The goal of the project is to develop an analytical model and information system for credit risk assessment.
The biggest value from such credit risk assessment system is minimization of losses arising from approving customer contracts including device purchase. The system is capable of credit risk assessment online and in real time, without delaying the approval process. Account management also benefits – risk scores are monitored for existing customers and appropriate actions are taken – such as limit setting, value added service offers, etc.
Credit scoring is also beneficial in prescreening and targeting, providing the capability to precisely pick potential customers for acquisition. When specific collections models are not available, general credit risk models can help optimize the collection process (e.g. in selecting which accounts to keep collecting and which to transfer to a collection agency). Having a dependable credit risk assessment for each client enables new products and services (especially high value services) such as payments, credit, etc.
A large amount of revenue is lost in telecom industry due to suboptimal collection systems. Prioritizing accounts based solely on days past due or amount receivable or having the same collection strategy for all accounts are clear signs of a system which would benefit from optimization. More often than not, collection system is also very inefficient. For example, a lot of resources and time is wasted on lazy payers (subscribers who would pay anyway –without any collection action). The goal is to optimize the entire system (process, organization, people, technology and methodology), maximizing the collected amount, reducing collection cost and manual effort.
Increased revenue arising from maximizing amount collected is the basic value. There are additional benefits, such as increased customer satisfaction and churn prevention due to better handling of lazy payers.
Fraud incidents in the telecom industry are on the rise. Technological advancement, constant product innovation, increasingly complex offerings, etc. contribute to the problem of rising fraud cases and associated revenue losses. The most common type of fraud in local market is subscriber fraud, where a subscriber uses services and benefits from long term contracts with no intention of paying. In subscriber fraud we usually distinguish between fraud for profit (e.g. selling long distance calls to third parties) and fraud for personal usage. The goal is to build a system capable of automatic fraud detection, whether in application stage (new customers) or in account management stage (existing customers).
Business value in large part relates to minimization of revenue losses due to fraud, both in device purchase financing, core services usage and value added services. When fraud is detected at application stage revenue loss is completely mitigated, while for existing subscribers it is minimized. Automatic fraud detection speeds up the approval and contracting process and reduces manual work. Customer relationship management also benefits from a reduction in false positive rates.
Due to high volume and velocity of data, financial services are a natural application area for data sciece. Moving beyond usual structured data, we help our clients monetize on data sources such as: payment transactions, social media, emails, call center calls, etc.
IN2data has extensive experience with credit risk modelling for all asset classes, from private individuals, SMEs, corporates to financial institutions and specialized finance. Apart from the usual expert segmentation, application and behavoiral scoring, rating models and risk parameters estimation, we make use of non-conventional data sources, social network analysis, etc.
High performing fraud detection system has to cope with both new customers in application stage and existing customers through account management. It also has to detect as much fraud cases as possible keeping false positives under constraints of available resources in fraud management process. Such high requirements can only be fulfillesd with a combination of different methodological appproaches expert systems, statistical models, behavioral profiling, social network analyis, etc.
Collections in financial services require optimizing the entire system: process, organization, people, technology and methodology. The goal of optimization is no just to maximize the collected amount, but also to reduce collection costs, minimize migration to higher delinquency buckets, etc. Collection accounts are segmented using both expert(e.g. VIPs, employees) and statistical segmentation (e.g. lazy payers). Each segment is associated with an optimal collection strategy, which defines the best collection action to take at a particular day past due.
Customer segmentation is one of the preconditions for effective customer relationship management. We typically use two segmentation layers: behavioral (e.g. overdraft only clients) and lifecycle (e.g. startup). Segmentation is done using both expert rules (e.g. single/double accounting, industry, etc.) and statistical models based on historical socio-demographic, behvioral and other types of data.
Advanced predictive modelling can help financial institutions change from passively offering to proactively monitoring and analyzing customer behavior in order to cross/up sell by anticipating customer needs. Predictive models are built using internal historical data where the outcome is a propensity score for different products.
By determining the optimal product package and price for each customer segment or even for each customer individually, financial institutions can maximize revenue and achieve many other, non-financial goals. Methodology is based on determining price elasticity, optimization models, social network analysis to take cross selling potential into account, dynamic pricing, etc.
Customer lifetime value is a basis for all customer relationship management processes. Differentiating based on customer value enables the financial institution to drive the business towards more profitable customers. Customer lifetime value is much more complex than simple accounting measures: it includes cost of risk, future revenues, time value of money, cross selling potential, loyalty, etc.
Churn prediction models are becoming a neccessity for financial institutions in a competitive market. Our methodological approach involves not just the usual socio-demographic and behavioral data, but also includes social network analysis methods to identify influencers within the customer's network. By focusing its effort on customers who are both of high value and have a high probability of churn, the churn model serves as a guard of the most profitable customer segments. Apart from obvious financial value, churn prediction also provides strategic benefits. By analyzing churn rates and trends on a segment level, the financial institution gets a valuable insight into market movements, trends and competitors behavior and strategy.
Retail industry is one of our main areas of interest and expertise. As a highly competitive, dynamic and data rich industry, it’s a perfect setting for data science applications. We were already engaged in several projects in retail industry including customer profiling and segmentation, category management, fact based negotiation, assortment optimization, pricing strategy, product value & benchmarking, market basket & item analysis, multi variate attribute analysis, consumer behavior analysis, promotion decomposition & optimization, product evaluation, product lifecycle management, market demand forecasting, customer value & credit scoring, churn prediction, customer value scoring, data enrichment, logistic chain & capacity optimization etc. We would like to stress that, apart from methodological and technical competences; we have extensive understanding of retail business domain – its products, processes, market etc.
In retail industry one of most important efficiency driver is finding and creating profitable customers towards determination what drives profitability. This leads to better prospecting and more successful customer relationship management. Retailer can segment and profile their customer base to uncover those profit drivers using the knowledge of their customers, products, and markets; combining those with data-driven techniques to find natural clusters in their customer or prospect base. Whatever the method, the process will lead to knowledge and understanding that is critical to maintaining a competitive edge.
With the goal of profitably delivering customer value at the store shelf, retailers need to have the right products, in the right stores, in the right positions, at the right time. Using different models for category management retailers can effectively plan, execute and analyze demand-based category and assortment strategies, while driving overall profitability.
Fact based negotiation is often called “science of negotiation“. The image of the lone extroverted negotiator skilled in the tricks and tactics of negotiation has been replaced by the teams of negotiators that are empowered by their knowledge, discipline, and preparation. It is no longer a game; it is a process - a process that is learnable and repeatable but based on long-term relationships between the buyer and the seller - not a short-term advantage of one over the other, but in true partnership seeking to reduce cost and improve value and performance.
Getting the product assortment right is difficult and a complex task to achieve. It is also critical to retail success. Unlike inventory management and pricing, where retailers have lots of data and analytical tools to guide decision making, assortment optimization is still much more art than science.
Not all promotions are effective in driving incremental sales, units, and profits for manufacturers and retailers. Two important negative impacts to understand are the effects of time-shifting of consumer purchases and product portfolio cannibalization. Time-shifting can occur when customers adjust their purchase cycle to take advantage of promotions, modifying consumer behavior to only buy the product on-promotion and/or pantry-load during promotions to eliminate the need to buy the product at a normal base price level. High frequency and repetition of promotions and price points can also lead to cannibalization of other products within the portfolio and the category. Consumers purchasing the promoted item instead of a non-promoted product can negatively affect the promoting company as well as the rest of the category.
Demand forecasting and estimation gives businesses valuable information about the markets in which they operate and the markets they plan to pursue. Forecasting and estimation are interchangeable terms that basically mean predicting what will happen in the future. If businesses do not use demand forecasting and estimation, they risk entering markets that have no need for the business's product. The purpose of demand forecasting and estimation is to find a business's potential demand so managers can make accurate decisions about pricing, business growth and market potential. Managers base pricing on demand trends in the market. For example, if the markets demand for pizza is high in a city but there are few competitors, managers know they can price pizzas higher than if the demand was lower. Established businesses use demand forecasting and estimation if they consider entering a new market. If the demand for their product is currently low, but will increase in the future, they will wait to enter the market.
Retailers seek to meet local consumer needs by customizing assortments, store layouts, merchandisable and non-merchandisable space, and unique promotions. They must do so while also maximizing profit and return on investment from every inch of each store’s space. This requires appropriate planning of space allocation; from the floor space assigned for each category, to the shelf capacity for individual items. However, the lack of visibility into each individual store’s actual fixture and space constraints means that prototype store layouts and one-size-fits-all planograms are developed. As each individual store strives to achieve localization, it can be difficult to ensure that what is executed meets headquarters’ overall category goals and corporate objectives. Furthermore, because space planners at headquarters rarely have visibility into and an accurate understanding of the actual modifications performed at each individual store, future space planning becomes cumbersome.
Although it is much more expensive to acquire new than to retain an existing customer and retailers are thus focusing more on customer churn and retention, churn rates on the industry levels are nevertheless rising. The goal is to selectively pick customers or customer segments which are of substantial value to the retailer and in line with their portfolio management goals and overall strategy. Apart from the question whom the retailer should target for acquisition, there is also the question of how (which channel), by whom (which internal unit or external partner), when and at what cost.
There are three main factors for customer profitability for any retailer: acquisition, margin and retention. In this area we deal with increasing customer margin. Cross and up selling is becoming harder each day with increasingly complex products, pricing, new value added services, etc. Timing and context are of extreme importance in cross and up selling. The goal is to build a model which will enable the retailer to offer the best product, at best time, through the best channel to maximize revenue and customer lifetime value. The ultimate goal is to change from passively offering to proactively monitoring and analyzing customer behavior and upsell and cross-sell by anticipating customer needs.
Faced with increasing competitions, declining revenues, lower margins and other issues, product packaging, pricing and revenue (yield) management becomes one the basic tools retailers use to keep and grow revenue and remain profitable. The goal is to build a model which, by determining the optimal product package and price for each customer segment, maximizes revenue and helps the retailer achieve other, non-financial goals.
One of the biggest differentiators and sources of competitive advantage in the retail market today is the retailer's ability to tailor its offerings, price, quality of service, etc. to customers based on their value. The concept of customer value however is much more complex that an ordinary ex post profitability measure. The goal is to build a model which will provide customer lifetime value on a customer and segment level and thus drive the business towards more profitable customer segments.
Increasing the level of productivity of wind turbines by utilizing functional and weather data with predictions regarding the operating conditions. We consider the short term wind prediction (12h) to be a useful asset for a more efficient production management. An extended scope would be deciding upon the optimal position of the wind turbine considering geographical information as well as statistical data on the wind.
Predictive meteorological models usually create predictions for a longer timeframe (one or several times daily). It is possible to construct an additional model which would correct the initial forecasts during the day, for a more efficient utilization of the wind turbine. As the wind farms are located in the zones with a high risk from fast winds, the model can predict an optimal period for stopping and resuming all operations, which would help protect the turbines from damage, and at the same time minimize the unused operational period. The models can be adapted to better accommodate the local specific conditions.
An in-depth understanding of the specifics (weather, wind direction) gives an advantage in planning and capacity utilization. The business value is expected via the location analysis, the feasibility analysis when deciding upon additions to the wind farm, as well as usability analysis. Predictive systems such as these facilitate system maintenance and assessing alternatives.
Increasing the level of productivity for solar cells by utilizing functional and weather data with predictions regarding the operating conditions. The goal is to predict the amount of sunlight in the following period for a more efficient energy management.
For the solar power plants, roughly the same possibilities and conditions are available as with wind farms, but the predictions are based on sunlight instead of wind parameters. Nevertheless, wind does play an additional role in the adjustment of solar panels, and therefore monitoring the wind speed and possible swings are essential for protecting the panels.
An in-depth understanding of the specifics (weather, sunlight) which give an advantage in planning and capacity utilization. The business value is expected via the location analysis, the feasibility analysis when deciding upon additions to the solar plant, as well as usability. Predictive systems such as these facilitate system maintenance and assessing alternatives.
The consumer behavior over time is an important factor for the overall system balance. This behavior can be predicted for the upcoming period (day, week, month, year) based on: last period consumption, weather data, upcoming event data (sport events, concerts). The prediction can be aggregated by location or time.
With a large quantity of input data and constraints it is possible to optimize and plan the production. The optimization model degrades slowly, so it is possible to use it over a longer period.
The execution of the given activities in the appropriate time, as opposed to an ad hoc solution enables significant savings and better resource management. The models we would produce combined with expert knowledge enable a more efficient resource utilization and allocation and better planning capabilities. There are certain specifics regarding Croatia in terms of accumulation lakes and their potential, which help to fight shortages.
The main goal is to calculate, monitor and understand power transfer systems. The systems currently available can plan certain activities with a high precision, with the goal of making the models as precise as possible in the most challenging areas, and shorten the prediction time in under an hour. Since these systems require a continuous tuning, an efficient model is possible only with expert intervention, since the knowledge from a different location is not transferable.
In this case we also consider the imbalance between production and consumption. We use all available sources for the monitoring, and we combine internal and external data for prediction and we adapt them to suit local peculiarities. We aim at building a complete understanding. We take into account the fact that vertical winds can decrease production, and our model takes that into account. The variable with most variance is precipitation, which we also take into account
This model can be used in isolation but also combined with other datasets and proposed models. The business value is aimed at archiving a high level of precision and utilizing the results in the top 5%.
As the difference between planned and actual spending of energy (balancing energy) is monitored according to legal regulation, and errors in the planning of added costs to participants in the network, the objective is primarily to provide a balance but also to allow surplus management, which is an opportunity for additional income. Models that allow reaction in a short time can be particularly important (by combining forecasting consumption and be able to "sell" the excess energy). If a country wants to show preference for "green" energy.
We offer surplus management by monitoring and forecasting resources through provision or trade, focusing on short-term forecasts/predictions. Each energy operator has a department for trading (considering that in this industry is very risky not to use hedging) which usually does not have access to an ideal tool for hedging, but only to simple calculations. Participants who purchase energy must declare it a day in advance (wherein traders are not included) to provide pay compensation for "mistakes". The imbalance of short-term consumption (HEP Supply and HEP ODS) is reported to the institutions HROTE and HOPS. The same process works on a global scale where the HOPS takes care of the balance. Minor corrections are permitted; they are "naturally" offset. Croatia must be neutral by the statistics "of its lines" as well as the system as a whole.
The differences in the form of surpluses and/or deficits are monitored within the given time interval. Surpluses and deficiency are penalized and amounts vary depending on the time. Systems must be able to provide insights on 1d, 1h and 15m predictions. Today traders use 3-4 tools at the same time to analyze alternative approaches, so the understanding of events and quality evaluation of preventive option is of great importance.
The analysis of service intervals through proactive monitoring of working processes of machines. Proactive prediction of machine maintenance. Maintenance of these complex systems consists of: equipment maintenance, maintenance of infrastructure, planning of (seasonal) network load and HR management of people who participate in the maintenance. Equipment maintenance includes high-quality machine tracking, purchase/replacement of machines, control of machine operations, predicting the degree of deterioration of parts of the machine, service on time etc. Infrastructure maintenance includes quality tracking of the infrastructure parts, tracking of network loads, of the deterioration, as well as of the alert systems and emergency planning systems (as opposed to much less efficient ad-hoc controls). Planning of the seasonal network loads includes projections of consumption and supporting resources (e.g. number of support teams). Finally, staff planning and staff training can be efficiently handled with cost optimization.
Modern systems allow different kinds of machine tracking (of their work and components control), as well as tracking of parts of the infrastructure, monitoring of network load and planning of human resources depending on all these parameters. We strive to make a specific examination or service when it is necessary by using the IoT and other parameters. An intelligent alert and early warning system is created by combining these data, which acts proactively, increase the efficiency of prevention and prevent incidents, i.e. optimizes the use of resources at all levels.
Understanding the events and quality evaluation of preventive options. Due to maintenance costs that occur in this sector, intelligent predictive maintenance is an extremely important component of effective business. Although optimizing the costs has the direct commercial value, with additional benefits reflected in better management of all the resources (including human resources), which is particularly manifested in the long term use of these models.
Usage of a variety of data about consumers, their behavior, their consumption (on a daily, weekly, monthly or long-term basis), as well as the network characteristics for the analysis, segmentation and profiling purposes
The system is based on the level of the consumer being connected to the substation. Several analytic methodologies like buyer and consumer clustering, consumption habits etc. are being applied.
The idea is that consumers can themselves participate in the distribution of energy in the network (e.g. the consumer may waive part of the energy). From simple applications in the competitive market it is possible to predict consumers with significant energy needs (and to suggest them models or consumption packages), to provide the market protection in the case of the competition entrance, to manage supply/prices, to plan the renewal of capacity or alternatives.
Political marketing is the application of marketing principles and procedures in political campaigns. Procedures include analysis, development, execution, and management of strategic campaigns. It is about how politicians use marketing tools and concepts to understand, respond to, involve and communicate with their political market in order to achieve their goals. It helps candidates and parties to understand voters, to know how, when and where to react, what messages to send, and to track progress during campaign.
Opinion polls are crucial aspect of political marketing as they give insight into each candidate's policy position in electorate at the moment a survey has been conducted. The goal is to define a process that will continually track political opinion in order to know how, when and where to react, and to predict results of the elections. Political opinion polls represent snapshot of voter’s political preferences in the electorate and it is important to consistently track progress during campaign. At the very beginning, benchmark polls can show candidate’s strengths and weaknesses, and can give an idea of what messages to send in order to strengthen performance in electorate. The value of brushfire polls is that they give insight into candidates name recognition in electorate, can spot issues with current campaign, and can determine voters’ attitude about new problems that may have arisen. Brushfire polls can also determine progress on the ballot and help to rectify mistakes. Usefulness of the tracking polls is that they are not just snapshots of the electorate, they represent a moving picture which shows how are voter’s preferences changing at the end of campaign, and can be used to predict upcoming elections.
Although opinion polls give insight into voters’ political preferences, it is important to describe and quantify why people intend (not) to vote for a particular candidate or party. Political behavior analysis focus on why, rather than just how many, people have such political preferences. In order to truly understand voters, it is important to understand their personality, values, beliefs, what motivates and demotivates them to vote, etc. The goal is to build complex models which describe relationship between dependent variables (political preferences and voting in elections) and independent variables (demographic, socioeconomic and behavioral variables), and to create voter profiles. Political behavior analysis gives insight in characteristics of voters who vote for particular party and who vote in general. Voters who change their voting habits and undecided voters are of particular interest. Knowing your voters in extremely important, but knowing how to motivate swing voters is something that can make difference between winning and losing the elections. With right questions and models it is possible to detect them with high precision. Furthermore, it is possible to determine optimal target groups, with respect to probability of voting and group size, so campaign can focus on them, rather than focusing on unreachable voters or closed voters.
In order to target wider audience, particularly swing voters, it is important to create an optimal system of sending ideas and messages to voters at right time and in right place. The first step is to decide which groups campaign should target. This is usually done after benchmark poll has been conducted and lower support in some voters groups has been spotted. Analyzing behavioral questions, it is possible to obtain which messages and problems interest these groups and what is the easiest way (in the context of media) to send political message to them. In order to find suitable approach, optimization problem is solved, considering campaign budget, size of desired target group in population, and influence of the messages on target group. During campaign, if results from brushfire polls show particular interest for some new topics that were not present at the beginning of campaign, it is necessary define messages that will attract voters’ attention. The biggest value is that choosing
The main idea in political network analysis is to describe relationship and influence voters have with each other and voter dynamics. The goal of this analysis to answer questions like “do voters with same political preferences have stronger tendency to associate with each other than with voters that have different preferences?”, “do parents influence on political preferences of their children?” or “are voters of political party A in general members of some non-political communities?” Using political network analysis it is possible to describe social relationship between voters which can be helpful during campaign to easily reach target groups, especially if members of these groups are strongly connected. Affiliation and social-affiliation network can reveal subtleties in the interactions among both the voters and the organizations. Tracking these networks can explain voter dynamics and can help to understand what the source of influence is and who influential voters are. If influential voters are detected, campaign can focus on them.
Visual analytics provides a visual interface that supports analytical reasoning, bringing visual dimension to the interaction between human analyst and data. People understand information behind creative visualization much better and faster than raw data from letters and numbers in rows and columns, irrespective of industry or the area of analytical challenge.
Visual analytics enable you to more effectively ask and answer important questions such as “Where are sales growing,” “What is driving growth” and “What are the characteristics of my customers using different services?” Combine visual analysis with the power of data science, and transform your data from an underutilized asset to a competitive advantage. Data visualization tools allow anyone to organize and present information intuitively. Data visualization gives key decision makers the ability to see patterns such as sales trends, customer buying habits, or production bottlenecks -- and respond accordingly. Data visualization software puts incredible data analysis capabilities into the hands of people throughout an organization, no matter the amount of data stored. As the data grows, it is becoming ever more imperative for organizations of all types to find ways to sort and analyze the records in their burgeoning data stores. Today, how a business manages its digital files is often the determining factor in its success. The Economist Intelligence Unit says companies that consider themselves ahead of their competitors in their handling of data “are three times more likely to rate themselves as substantially ahead in financial performance.”
For years, organizations have used business intelligence and analytics software to build reports from various repositories. However, this is often a time consuming process controlled by the information technology department. Some companies have recently taken the step of hiring data scientists, but good data scientists can be hard to find and expensive to employ. There is, however, another way. Despite sophisticate, the data visualization software can also be easy-to-use tool that allows anyone who is familiar with an organization’s data to quickly analyze, graphically display and share information. Data visualization software can tie directly into a variety of data sets, files and spreadsheets in real time. Then, with just a few clicks, a user can pull data from just about anywhere and have it rendered in an interactive chart, table or dashboard. Since the software allows users to interact with the data, managers can group, analyze, organize, and summarized data with speed and efficiency. Indeed, data visualization software puts incredible data analysis capabilities into the hands of people throughout an organization, creating a team of knowledge workers able to see problems, spot efficiencies, and seize new opportunities – no matter the amount of data stored.
There is a rich bounty of customer insights hidden in the midst of textual data. But the question always remains, “Where do we start?” We enable companies to incorporate understanding of the textual content, providing critical competitive advantage over their peers, providing key areas of improvement and increased vigil to streamline workflows. Key areas include:
By unlocking the value of content, you are able to deliver smarter solutions for brand monitoring, sentiment analysis, competitive and market intelligence, customer experience management, healthcare surveillance, early warning, eCommerce, eDiscovery and much more. Key applications include:
Clip Counting refers to counting all the editorial coverage (news items, feature stories, guest editorials, reviews, roundup stories, buyer's guides, etc.) that mention a client company, product or service. This coverage includes print media, radio, television, web sites and social media. Clipping services clip literally every article they see that mentions your company name in any way. For example, if an automobile accident occurs near your headquarters building and your company's name is mentioned in a newspaper story, you will receive that clip (and pay for it), even though you probably wouldn't have bothered to clip it yourself.
In addition to Counting Clips method, we calculate Media Impressions (the number of people who may have seen an article, heard something on the radio or in a podcast, watched something on television, or read something on a web page or blog), which can directly help you navigate toward higher profitability.
In addition to the analysis of an individual topic, event, brand or company, a content analysis can also be conducted comparing the performance of companies, brands, topics or events on their media coverage. This can range from comparisons of the total number of clips to the share of discussion to comparisons of the overall prominence one brand or company receives over another.
Stress, emotions, moods, engagement and many other complex mental states can be estimated from person’s acoustic and linguistic speech cues– using data mining methods. In most cases only the prosodic structure is analyzed, i.e. speech intonation, tone, stress, and rhythm. The thing with the acoustic/prosodic speech analysis is that it is not important what is said, which sentence (which words are used). It is only important how it is said. It is much harder to manipulate the prosodic structure than words and therefore acoustic speech measures are more reliable. There are lots of acoustic features that, with more or less efficiency, describe a style of spoken utterance – a typical emotional patterns in speech, e.g. loudness when speaker is angry or slow speech (with longer pauses) when he or she is calm. We are extracting several hundred to thousand (statistical and other) acoustic features from standard speech measures: fundamental frequency, energy, speech rate, pauses, spectral structure measures like harmonics, formants, high-frequency power etc., perturbation measures like jitter and shimmer, voice quality measures etc. Furthermore, a fusion of acoustic and linguistic features can be done for estimation cases like motivational speech efficiency analysis, where word and phrase usage is also relevant and representative. In some cases, only the linguistic cue analysis is possible. E.g. when estimating depression based on person’s social network behavior. Estimation/classification of continuous/discrete emotional and mental states is performed using machine learning methods like support vector machines, random forest, artificial neural networks, etc.
Imagine an angry and unsatisfied customer. Imagine a call center agent – a company’s first and in most cases only target of this angry customer. It would be nice if the agents have as much as possible valuable and quick information about the customers so they can prepare the conversation strategy from the very start. A computational system that analyze emotions and moods like anger, rage, calmness and satisfaction in real-time, from only a few seconds of customer’s speech, can be used to help the call center agents during the conversations. The important thing is to increase awareness, i.e. to understand behavior and resulting customer perception. Discrete emotions like happiness/joy and anger, continuous emotional dimensions: pleasure and arousal, stress level, as well as several moods like calmness, satisfaction and irritability are estimated from acoustic parameters of customer’s voice using machine learning methods (see Speech analytics: A general methodology topic for a more detailed information). Better understanding of customer’s emotional state during the conversation leads to more quality connection. Customer has a better experience talking to the agent, which potentially leads to facilitate the establishment of trust. Customer is consequently more satisfied with the agent (with the company?). Furthermore, more efficient conversation can decrease average call handle time (AHT) – an important productivity measure of a call center. Studies have shown that even a small decrease of AHT can bring large company savings. For example, an AHT reduce by only few percent could literally save the company millions of dollars per year.
Speech analytics can be used to increase sales performance. A system that recognizes human emotions and moods from speech patterns can be used as an assistant to the contact center agents, during the conversation with the customers. This way, agents are more aware and understand better customers’ behavior and mood changes during the product sales. As the result, a general customer’s experience can be improved. Agents are also more aware about the customers’ engagement level during the sale, which can lead to customer’s conversion rates improvement (from unengaged to engaged customers). A general customer’s engagement and satisfaction with company and its products (e.g. Telco or insurance company) can be measured using speech analysis through one or more dialogues with the agent. Speech analysis can therefore be used as one modality in churn prediction. System can also analyze agent’s efficiency and sale skills. Agents can learn how to improve mirroring and motivational skills, how to exude warmth and enthusiasm, and generally, how to be more persuasive – how to increase customer’s engagement. Satisfaction and engagement of a customer are estimated from acoustic parameters of his/her voice using machine learning methods (see Speech analytics: A general methodology topic for a more detailed information). Agent’s efficiency and sale skills are analyzed on the basis of customers’ conversion rates and engagement changes, their calmness and enthusiasm when speaking with the customer (agent’s speech analysis), as well as their ability to control and manage the dialogue. A better customer’s experience during the conversation with the contact center agent is crucial for building trust and satisfaction. These conditions can potentially lead to customer’s engagement increase and consequently: increase of sales performance.
Public speeches are often related with significant speaker’s emotional stress. Stress can largely affect public speech performance, which can be observed throughout various voice quality measures and other speech parameters. E.g. speaker’s mouth gets dry, laryngeal muscles are tenser, breathing is disrupted and laud, etc. Changes in public speech performance entails various acoustic/prosodic speech feature changes, which can be measured using computational (objective) methods. This way, speaker’s stress, as well as speech performance parameters like expressivity, vocal variety, etc. can be estimated from the acoustic speech structure. Speakers can use this computational system as an assistant to enhance public speaking performance. System can be used in off-line mode in a form of detailed speech analysis. In this case, speech samples are acquired and stored during speech (whether it is a practice or the real presentation) and the analysis is performed after the presentation. Stress level, as well as voice quality parameters, expressiveness, prosodic speech parameters like fundamental frequency (pitch), intensity (loudness), speech rate, silence periods, etc. are plotted as trajectories – paired with speech signal during the presentation. Speech can also be paired with presentation data (e.g. PPT slides). This way, it’s easy to observe speech quality and stress in any time during the presentation. E.g. you can see on which slide – which bullet – the stress level is high. Additionally, application can contain educational component with suggestions and guidelines how to improve speech performance. Estimation of various public speech variables is performed using machine learning methods (see Speech analytics: A general methodology topic for a more detailed information). Furthermore, system can optionally be used as a real-time speech quality and stress estimator. This means that a system can run analysis in background during speech and plot, for example, the stress level on speaker’s screen (laptop or his/her personal presentation monitor) in a scale or semaphore format. A system can additionally inform the speaker about the increases/decreases of various speech measures during the presentation like speech rate and intensity, when he or she mumbles – does not speak clearly, etc. It can also “suggest” the speaker when is a right moment to make a longer pause, when he/she can fasten up the speech or when is appropriate to be more expressive. Who can use this computational system… to train and enhance public speaking performance? Managers, consultants, generally all the people who pay attention to every detail of their performance during sales or product presentation. Lawyers, professional speakers, politicians. Public speaking trainers. The academic community: students who want to prepare their presentation better. Such analysis can serve as an objective indicator about the readiness and certainty level of a speaker. Speaker can see on which parts of the presentation he/she has to work more. Stress level and voice quality parameters can be visible to the speaker, but also to the commission during, for example, speech contests – as an assistance during the evaluation process. It can be visible even to the audience. Such objective analysis of speakers’ fluency, voice quality, motivational and other speaking skills can be applied during speech competitions like Toastmasters speech contests, debates (political or any other), TV shows (like Piramida), etc.
Frequent car accident causes are driver’s fatigue, as well as drunkenness. Furthermore, potential problems can occur when driver is surrounded with high-emotional distractions, e.g. when driving kids or during a heated argument with co-driver. The idea is to prevent these risky conditions: to prevent the driver to drive during high fatigue or when drunk, and to redirect the driver’s attention on the road when driver is distracted and pay too much attention to events in the car. Driver’s fatigue is analyzed using an automatic computational system, which estimate a fatigue level from driver’s voice (see Speech analytics: A general methodology topic for a more detailed information). It can also be mixed with video modality, which tracks fatigue from driver’s facial movements. Fusion of these two modalities provides stable use in night driving conditions, the conditions when the driver has a partially covered face (e.g. sunglasses) and when the driver in not talking. A requirement for the use of audio modality is that the driver speaks something. It is thus possible to provide a simple control questions to the driver to ensure a stable usage. An intoxication detection in speech is performed by using driver’s voice (and additionally facial expression) analysis. Acoustic and linguistic structure, as well as facial movements drastically changes when person is drunk. It is important to maintain driver’s calmness and focus on the road. So called anger detector is performed based on driver’s voice (and optionally facial expressions and gestures). Furthermore, driver's excessive occupation with the events in the car can be analyzed through his/her involvement in the discussions as well as dynamic movements and looking away from the road. It is of high importance for safe driving that the driver is sober, not feeling tired, and is calm and focused on the road. New analysis published in the American Journal of Public Health has shown that installing devices in new cars to prevent drunk drivers from starting the engine could prevent 85 percent of alcohol-related deaths on U.S. roads, saving tens of thousands of lives and billions of dollars from injury-related costs.
How to predict mental or physical health problems? How to recognize characteristic cues at the very beginning of upcoming disorder or disease? We have enormous amount of everyday data that “floats around us” – our speech data, text data. We talk a lot during the day. We communicate with others – live and over a phone. We text with our friends, family, complete strangers on internet. Write statuses on social networks. We share a lots of vocal and textual information about ourselves every day. Using these everyday information it is possible to predict a lots of mental and physical problems. For example, a system running in background can detect amplitude changes in voice tremor of Parkinson’s disease patient when speaking. It can thus provide early warning that he/she should visit a doctor. Another case is depression detection based on behavioral characteristics of people on social networks and on the Internet in general. Generally, patients with mental health problems like depression, PTSD, paranoia, schizophrenia etc. can track their condition based on their vocal and textual data in everyday life. A system can give them early alerts when their condition is getting worsen and when is a good time to visit a doctor. Stress is omnipresent – a fact of modern life. Constant exposure to the intensive stress can seriously damage person’s health. It is very important for the person to stop on time, to have a time for relaxation and for rest. With this in mind, stress can be analyzed during the day from person’s voice in order to provide warnings and suggestions when it is necessary to relax, to change environment, to take a break at work, to go to sleep etc. A long-term analysis can also be performed that analyzes stress in a longer period (months/years) and suggests when is a good time to take a break, to go to vacation or similar… In more serious cases to change approach at work or even to change the job if person is constantly exposed to very intensive stress and don’t know how to deal with it. Stress and various mental and physical disorders and diseases can be tracked by analyzing human voice and textual communication (see Speech analytics: A general methodology topic for a more detailed information). Mobile phones can be used for this purpose as well as sophisticated and specially designed microphone sensors for literally all day speech analysis. Reducing stress in everyday life is vital for maintaining overall health as it can improve person’s mood, boost immune function, promote longevity and allow him/her to be more productive. When stress is ignored, the person put himself/herself at risk of developing a range of illnesses – from the common cold to severe heart disease. A quality and reliable stress management system can thus be important accessory through everyday, modern life. Furthermore, early interventions when tracking patients with various mental and physical disorders are very important because potential problems can be prevented before it flares, thus also increasing the healing probability.