In what area can we help?

Pick an area where you seek improvement and we will show you possible solutions. Each client is unique to us and we always design a custom solution that will meet your expectations.

You do not fit in any category? Is your business unique and you need a specific approach? We will help you make your ideas come true. Reach out to us – initial consultations are free of charge. We look forward to hearing about your project!

Sample projects and solutions

Illustration

Chatbots and Voice Assistants

Chatbots and voicebots can manage thousands of customer interactions daily. We design the architecture and create conversational logic using NLP. We create chatbots and voicebots that are on the same wavelength as your customers.

Advanced helpdesk voice assistant

Helpdesk assistants answer thousands of calls every day. It is a routine task that often leads to employee fluctuation, which causes extra costs related to recruitment and newcomer training. Many helpdesk requests can be resolved simply by looking up a piece of information in the system. A voice assistant can handle non-complicated inquiries on its own and thus alleviate the workload associated with call answering.

Possible solution

12-hour operating helpdesk resolves thousands of phone calls every day. Voice assistant on its own is able to handle 95 % of inquiries. Remaining 5% are complex inquiries that are routed to a senior helpdesk worker. The workload is thus reduced by 90% and the helpdesk provider saves approximately 20 thousand EUR monthly. Customer satisfaction is also improved thanks to zero waiting time for an available helpdesk worker in peak times.

Simple voicebot for e-commerce

Many e-commerce businesses do not receive a vast amount of phone inquiries and therefore there is no dedicated person for call answering. On the other hand, the customer calls are mostly related to order status and present an extra workload for the person who is then less efficient managing their other tasks. The voicebot will handle the customer calls and the employees can then fully attend to their other work.

Possible solution

An e-commerce customer service line may receive approximately 2000 customer calls monthly. Therefore, it is not necessary to hire a dedicated customer support operator. The voicebot will handle 90 % of the calls and only 200 calls will be routed to actual helpdesk workers, which is less than 10 calls a day. The voicebot will save time and effort of the employees and the customers get zero waiting time during peak hours.

Illustration

Text, Image or recording data analysis

We can read invoice data from an image, analyze a voice recording or sort through emails thanks to machine learning . We can process data in all sorts of formats. Leave your routine tasks up to artificial intelligence.

Processing of scanned or electronic invoices

Information systems and databases are built from pieces of information that are structured. For instance, each invoice needs to have a total amount, the invoiced product or service, payment details and due date. The invoice information may be entered manually, however, artificial intelligence can perform the same task just as well and can even be faster. The system may also learn how to post the invoices. Based on the information read from the invoice it can suggest accounts and amounts to be posted and the responsible person can then just review and approve.

Possible solution

Logistic company transports goods all over the globe and receives invoices in very different formats. Therefore, the company needs several assistants to manually enter invoice data into their information system. As a solution, we create a mobile app that sends a photo of the invoice to an internal system that translates the image into text and then finds all the relevant information in the text. In case some piece of information is missing or if there are any discrepancies, an assistant is alerted to correct the invoice or enter any missing data. Then the invoice scan is approved and data are saved automatically into the information system. This solution saves about 10 minutes of time per invoice. As a result, two employees are freed up to perform more demanding tasks than re-typing invoice data. This solution can also be used for tracking employee travel expenses.

Sound analysis of manufactured product for QA

Quality of manufactured products is always checked before the product is shipped out. However, some defects are not detected, which causes additional costs related to warranty claims, making a replacement and logistics. Sound analysis will help to detect the defect in time before the product leaves the manufacturing plant.

Possible solution

A manufacturing plant makes machines that are turned on for a short period of time as a quality check to see if they function properly. Many defects are not detected in this short period of testing. They have created a database of sound recordings capturing the machines in operation. For several weeks they have been recording the sound of the machines during testing and therefore created a database of several hundreds of recordings. A month later they added to this database information whether the machine had any kind of defect. In total, 7 types of defects have been identified and each had 5 to 20 sound recordings. Processing the recordings created a complex neuron network that was able to forecast if the machine will have a defect and which type of defectit will be based on the sound of the machine. A mobile app has been created and is now used for quality assurance in the manufacturing plant.

Sound analysis of machines (machine health)

Manufacturing equipment is required by law to be inspected every 2 years. However, some machines break down during this period and the inspection then reveals the machine is not fit to operate any more. This causes delays in manufacturing. Artificial intelligence can set the inspection time based on input data and provides information on expected machine defect. AI can estimate the time period before the machine breaks down.

Possible solution

The sound of machines is periodically recorded and the recording is sent to a data center. We keep tract of the operating time of the machine, temperature and humidity. We also analyze the frequency of the usage of the machine and extremes of any kind. Other aspects, such as the qualification of the workers using the machine, are taken into account as well. The machine may also be watched using a video camera, which can spot a flashing indicator light or unplugged cables. Based on all this data, current state of the machine is described precisely and thanks to artificial intelligence we can forecast its service life and perform maintenance before the actual defect occurs and prevent production outages this way.

Forwarding of emails from various channels to the responsible department and automatic responses

A company receives thousands of messages via email, website forms or social networks every day. Most messages are handled by customer service employees that forward complex queries to the respective departments. Artificial intelligence can easily handle message forwarding and it can even respond on its own to some of the messages.

Possible solution

We explore the communication channels within the company and see where service requests are directed to and who handles internal and external communication. We focus on all available communication channels, such as email, messenger, CRM and other. The system will then learn message categories that are currently used and identify any additional categories. On top of that, it can assign priorities. As a result, the mail is sorted and forwarded automatically. We can even take it to the next step and implement automatic responses and alert users if new contacts or important documents show up. We will train the system to detect fraudulent messages, such as when the email address does not match the signature.

Illustration

Full-text and semantic search

We have experience with creation of search engines as well as improving search in client's Elasticsearch databases. Speed up your tasks, get underlying data or monitor websites using our smart search.

Full-text search

Overwhelming amount of data in different databases and documents maintained manually by users is the reality we often see at our clients. People tend to make mistakes, may not be trained properly, fill in the same information in an inconsistent way or they are even finding ways around the system. Searching for data then becomes troublesome. We can solve these issues by creating a properly programmed full-text search.

Possible solution

At first we audit all existing databases. Then we suggest suitable automated data fixes and new validation rules for data entry. In addition, we can use smart scripts that prompt the user with correct options and do not allow them to enter incoherent data. For example, when the user enters an inquiry to buy an apartment for 10 000 CZK monthly, the system warns them that they likely meant to rent the place or that they made a mistake in the figure. We can also implement a search system that will find typos, words replaced by a similar word with different meaning or other types of misunderstanding, such as a number spelled as a word. As a result, clean data will make analyses more precise and work faster. Thanks to high quality data you will always be able to find exactly what you need.

Semantic search for administration

Searching by content understands the meaning of words and therefore may be more efficient compared to full-text search. We can implement two search fields to be able to search either by a copy-pasted text or by a document. There is no need to search by key-words only. We can also implement searching by image.

Possible solution

A company has a large amount of documents stored in a document server but there is poor to none categorization the documents (such as tagging). They key-words cannot be relied upon. The documents need to be searched by content rather than their name. For example, we have a financial report from 2015 and would like to find all existing financial reports. When you search by the whole document, the system will find all documents that have a similar structure and content, even when the filenames or headings are completely different. The system is able to apprehend the meaning of the words, therefore when you search for the word invoice, it knows it may also look for a bill or statement. You can also use searching by image, for example in case of scanned invoices it may be useful to search by logo or image of the product.

Illustration

Data management and data analysis

Improve your data quality using artificial intelligence. Users are prone to error when entering data. AI is no only precise but can even recognize faulty products, analyze customers or identify risks and generate alerts. We create solutions that will fit your needs.

Connecting of business data

Companies often use various software tools and they move the data between them in excel or other file format. The systems do not interact with each other. We can help you automate the data transfers. We can also teach you to use data that you have not been using and that may be beneficial for your business.

Possible solution

When we replace manual data transfers with automated communication, we avoid the situation that an employee goes on sick leave and data are missing from the system. When we audit data and systems we often find data that no one noticed before, while the information may be useful for the business. We can build a system for data analysis that will monitor any data anomalies and other interesting facts.

Identification of high-risk documents and access to them

Company has large amounts of documents that are accessed by various users. Avoid documents getting into wrong hands.

Possible solution

Employees have access to data with various sensitivity levels. AI monitors access to the files and its views, edits and downloads. In case that any user suddenly starts to download, delete or view documents they have not been interested in before, an administrator is alerted. The access can also be temporarily denied until the suspicious behavior is resolved. In addition, AI can recognize a potentially dangerous contentof a document based on access privileges, content and its similarity to other documents.

Illustration

Predictive analyses

Artificial intelligence is able to process large amount of data and thanks to that we can predict machine healthiness, cash flow or optimize resources. Let us know what it is you need.

Accident predictions based on historic and geographic data

By collecting relevant data, we can create an AI-based solution that will plan infrastructure inspections so that at every point of time the areas with the highest failure risk are checked.

Possible solution

A company maintains a vast network of underground and above-ground infrastructure. Access to many parts of the network can only be achieved by excavation or the access needs to be planned in advance and approved by authorities. The risk of emergency breakdowns needs to be diminished. How do we approach this?

  1. Analysis of all available data
    • Type and material of the collector
    • Depth
    • Temperature and precipitation levels at the location of the collector and in its proximity (road, mine, factory, highway)
    • Age of the collector
    • Regularity of usage
    • Emergency breakdowns or construction in proximity of the collector
  2. All the collected data enable us to create an AI-based solution that plans inspections in a way that the places where it is most likely that a breakdown will occur are checked.

Inventory optimization

The aim is to plan inventory in a way that you have just the amount of goods that are needed. Inventory optimization can bring substantial savings. The goal is to have just the inventory levels needed to cover the demand at a certain point of time, taking into account lead times, manufacturing times, component availability and prices. Using data analysis we can forecast the demand and therefore optimize purchases and transfers of goods.

Possible solution

Analysis of current inventory levels together with historic data and the plan of manufacturing allows us to create forecasts for purchasing and manufacturing. We factor in aspects such as weather, economic situation based on stock prices or seasonal demand dependent e.g. on holidays. We collect and use supply chain data like lead times, expected delivery dates and component prices. An important input for the forecasts are also sales targets for future periods since any effort put in sales, such as adding a sales channel, will impact the demand.

Financial planning and cashflow management

Cash flow management is crucial for healthy operation of the company. Artificial intelligence is able to plan the incomes based on expected invoice payment dates or periodic expenses from the past. We will help you with sustainable financial management.

Possible solution

Machine learning can forecast a future trend based on large amounts of historic data. We create a system that models future incomes and expenses based on this data. We can also add additional aspects such as payment behavior of the customer or current economic situation.

Prediction of supplier and customer behavior

Every company is dependent on the network of their suppliers and customers and if they hit any issues they need to know in time. It is not possible to rely on being informed by the supplier or customer and also it is impossible to regularly monitor the fitness of all the partners' businesses. Data analysis may help to forecast behavior of your suppliers or customers and you will be able to prepare in time for any troubles.

Possible solution

We create a system that watches stock prices, economic news and other publicly available documents. We analyze the prices of components and finished goods in wholesale markets. The system can also monitor newspaper articles and look for new products, expansion plans but also potential risks, insolvencies and complaints. By collecting all data we are able to forecast probable outages of the suppliers and customers and also increased demand for specific components or goods. The system also generates periodic reports of potential risks and opportunities and sends them to the respective manager.

Illustration

Stylometry and authorship attribution

Stylometry is able to recognize the smallest variations in text. It can be used as a measure of employee mental health, detect fake customer reviews or fraudulent documents. We have been awarded for an authorship attribution system.

Fight against extremism

Stylometry analyzes the text in a way that it is able to identify a behavior that poses a risk by analyzing the person's writing style.

Possible solution

AI check's CTO has been awarded by the Ministry of Interior for security research. He took part in the development of a platform for security forces of the Czech Republic that is able to recognize the authorship of extremist or terrorist texts.

Accuracy check of registration details

Style of writing carries hidden information about the author such as gender or age. Artificial intelligence can thus reveal mistakes or intentional manipulation.

Possible solution

An online dating site publishes posts that are meant seriously. Some users may submit their inquiry to wrong category such as gender by mistake. The administrator does not have the capacity to review every single inquiry and correct mistakes. In many posts it is also not easy to spot the mistake. A plugin using artificial intelligence can identify these cases and categorize them properly. It can also categorize a post where the author does not mention their gender directly.

Detection of fake customer reviews or comments

Agencies that write fake reviews and comments in order to achieve either financial or political profit are the reality. These posts are not only fake positive reviews but also defamations the purpose of which is to ruin the reputation of a product or company. Artificial intelligence can reveal such forgery.

Possible solution

An e-commerce platform allows their customers to submit reviews of purchased products. Regularly, negative customer reviews from several accounts are submitted, which raises suspicion. The provider would like to find out if the reviews are authentic or if they are entered by someone associated with their competitors. How can we solve this? AI check implements a plug-in for the e-commerce platform. This plug-in compares the style of writing of the customer reviews and is able to identify the user accounts with the same style in which case an administrator is alerted. The administrator can then block the user accounts or IP addresses from which the fake reviews were submitted.

Expert reports

Expert opinions attributing document authorship or pointing out plagiarism based on the style of writing of the author have been used in court cases since mid-twentieth century. Artificial intelligence is able to evaluate texts and discover plagiarism or other type of fraud.

Possible solution

A client in the energy industry submits a bid in a public procurement tender. Another company contacts one of the organizers and is able to view the documentation submitted by the client. Subsequently, they create a bid that just slightly changes the formulation of the text in order to avoid suspicion of plagiarism. The client finds out that the bid has been plagiarized, makes a legal claim against the other company and therefore needs evidence of plagiarism. AI check in cooperation with an independent expert processes both bids together with other similar documents created by other authors. The created AI solution will then indicate whether the bid has been plagiarized. We will also present an analysis with a description of the reasoning and graphs to the court.

Contact:

E-mail: aicheck@aicheck.tech

Phone:: +420 605 235 151

At this time, we work 100% remote. Face-to-face meetings can be arranged.

Follow us:

linkedin

© 2021 Copyright: AI check, s.r.o., Pellicova 113/3c, 602 00 Brno, company ID 09830481, VAT CZ09830481 C 120948 administered by the District court in Brno.