20 Questions to Ask a Vendor Before Implementation
Finding your way around the document recognition technology market is not an easy task. Meanwhile, choosing a system should be approached responsibly: the reliability of the software developer determines whether the solution will function without failures and whether it will work at all.
To ensure that an expensive IT solution does not disappoint in the future, we, Smart Engineshave prepared a universal checklist. It will help you minimize the risks (spoiler: there are many) when choosing an optical character recognition system developer and choose the best solution.
This is what it looks like:
Does the development company officially employ more than 20 people?
Are reviews of the company's products mostly positive?
Does the developer's clients include large banks, telecom operators or other large companies?
The average salary of the company's employees is more than 200 thousand rubles?
Does the development company patent its inventions?
Do the company's employees regularly publish scientific articles and participate in scientific conferences?
Is the developer a Russian legal entity?
Does the developer company belong to Russian citizens?
Are the developer's software products included in the Russian software registry?
Technological aspects of solutions:
Does the software use only the developer's own OCR text recognition technology?
Do the developer's solutions work without GPU? And what OS do they support?
Is document recognition performed within your company's framework?
Images of documents Not are transferred to cloud services for processing?
Capabilities of recognition systems:
Can the developer's software products detect and process tables, stamps and barcodes?
The solution provides recognition of documents from other countries, when, for example, you need recognition not only of a Russian passport, but also of identity cards from the CIS, the EAEU and the whole world?
Do the developer's software products provide recognition of unstructured documents?
Is the solution able to find a document in a photo or scan and recognize it?
Does the solution recognize not only printed text, but also handwritten text?
Is document recognition supported directly on a mobile phone?
Checking the solution “on your own”:
Company Profile
Checking a development company begins with searching for its profile in Rusprofile or other statistics services. It contains verified information about the company's activities, which is necessary for assessing various risks. Among the main aspects that need to be paid attention to:
Number of employees and their average salary. A startup developing recognition systems needs to have at least one or two dozen employees – for continuous development of solutions, software delivery and its subsequent technical support. As a rule, the average salary of IT specialists involved in the creation of recognition systems in Moscow is approximately 200 thousand. If a startup employs only 3-5 people and, for example, for 40 thousand rubles, then this may indicate, in particular, the use of other people's technologies – if you have everything ready, then not so many workers (and not necessarily highly qualified) will be needed to repackage the software into your own software product.
Customer reviews and cases. Usually, IT startups actively inform about their clients: a special section is allocated for them on the websites, and it indicates who integrated the developer's solutions and which ones; press releases and news mention cases of new implementations, etc. The same applies to reviews – they are either posted on the website of the recognition system manufacturer, or can be easily found on various aggregators. If there are no records about clients, or they are available, but among them there are no leading banks, telecom operators or large enterprises, then this is also a reason to choose another developer. It is necessary to pay attention that the logos on the site may be associated with the passage of a competition, hackathon or the conduct of a pilot project.
Among the conditional “red flags” we can also highlight, for example, the presence of arbitration cases related to improper fulfillment of obligations under the contract for the provision of services by the developer; a sharp or continuing decline in profit and revenue indicators; the presence of information on the FTS website about the failure to submit tax reports for more than a year, and so on. All this information may indicate the financial instability of the company, the low quality of its software products and their lack of demand in the recognition systems market.
Scientific activities of the company. We look at the company's website and any open sources. If the employees of an IT company constantly publish articles in scientific journals (especially in those indexed in WoS and Scopus – that is, they are the “elite” of scientific publications), regularly receive patents for inventions and participate in conferences, including international ones, then this is a sign that the company's decisions are based on its own developments, which are also recognized by the scientific community. If nothing is known about the company's articles and patents, this may mean that it uses other people's technologies and is not a developer in the full sense of the word.
Legal risks
After you have reviewed the company profile, you should study the legal aspects of its activities. To do this, pay attention to:
Does the company have a Russian legal entity? It is important that the developer is a fully domestic company. A foreign owner can leave the Russian market at any time, and then users of the product may face either unstable operation of the systems or the shutdown of many business processes. To avoid this, it is recommended to visit the developer's website or legal entity verification services. If the company's structure is hidden (or its form of ownership is indicated, for example, as LLP – Limited Liability Partnership), this is a reason to doubt its long-term prospects for working in the Russian market.
Are the developer's software products included in the Russian software registry?. The situation in 2022, when many foreign IT startups left the Russian market and stopped supporting their corporate products in our country, clearly showed that in the current reality, one should rely on domestic rather than foreign solutions. This is especially true for critical infrastructure facilities, since a ban on the use of foreign software there will be introduced from January 1, 2025. In turn, to be included in the register, it is necessary to meet a number of strict requirements (the majority of the manufacturing company must be owned by Russians; the software must not be forcibly updated and managed from abroad; the software and its components must not have restrictions on use in Russia or certain regions, etc.), and if the software product has successfully passed all the necessary checks, this means that the Ministry of Digital Development, Communications and Mass Media trusts it.
The absence of such verification is fraught with huge fines for the customer company. Thus, in Russia there is a Federal Law “On Personal Data” dated 27.07.2006 N 152-FZ. In particular, it regulates cross-border data transfer – if you work with foreign partners or use a foreign service that processes personal data, you must notify Roskomnadzor about this. In case of failure to comply with this requirement, the operator may be brought to administrative responsibility: for cross-border data transfer without proper legal basis – up to 300,000 rubles (Part 1 of Article 13.11 of the Code of Administrative Offenses of the Russian Federation); for failure to notify Roskomnadzor about TPD – up to 5,000 rubles (Article 19.7 of the Code of Administrative Offenses of the Russian Federation), for failure to comply with Roskomnadzor's order to eliminate violations – up to 20,000 rubles (Part 1, Article 19.5 of the Code of Administrative Offenses of the Russian Federation). So if you use foreign software, you will be forced to regularly report to Roskomnadzor, which also has the right to prohibit data transfer. All this slows down the company's work and in some cases can lead to its partial or complete paralysis.
Technological aspects
Once all the legal details have been settled, the developer needs to be asked about all the technical nuances of his decisions. To do this, first of all, find out:
Is the OCR system used in the solution the company's own development? Optical character recognition (OCR) is a technology responsible for creating a digital copy of printed or handwritten characters. It is a key element of recognition systems, and therefore it is important that the software provider has its own OCR and not use someone else's. The vendor (or community, if we are talking about free or open source software) can limit the functionality of its software at any moment, and then both the developer company that used a third-party OCR system and its customers will be at risk.
What operating systems and architectures is the OCR solution compatible with? For example, with x86 and ARM processors. If you do not specify this, there is a risk that you will purchase a solution that will not run on your computer or tablet. Accordingly, you will waste your money. The same applies to powerful graphics processors (GPU) – developers promise high-quality recognition with no less impressive speed, but this requires powerful equipment, which the organization may not have the funds for. If the solution provider states that the program will provide high quality and speed without a GPU, then this can be considered a good sign – OCR technology is hardware-independent, and therefore it can be safely launched on all your devices.
It is necessary to dwell separately on the fact that where information processing takes place. Recognition solutions are on-premise and on-cloud. In the first case, document processing is carried out in the local RAM on the server, computer, phone, tablet or other device; in the second, images of documents, passports, receipts and the like are sent to cloud services for processing. On-cloud is much less reliable than on-premise, since there is a high risk that when sending data back and forth, no matter how they are protected, they can be accidentally (or intentionally) leaked into the network.
Let us recall that recently the State Duma has been seeking to increase penalties for companies that have committed personal data leaks citizens, and in January 2024, the law on turnover fines was adopted in the first reading. According to it, the amounts collected for legal entities can be from 0.1% to 3% of revenue for a calendar year or for part of the current year – no less than 15 million rubles and no more than 500 million rubles. Moreover, Russia plans to introduce special requirements for the qualifications and business reputation of deputy heads of financial institutions responsible for information security. In particular, with the participation of the Central Bank, a bill was developed that includes a provision prohibiting deputy heads of organizations responsible for information security from holding such a position for 10 years if, during their work in another institution, there were violations of information protection requirements. So, as a result of leaks of clients' personal data (and in fact – the choice of on-cloud solutions), companies may face millions in fines, reputational losses, and even the risk for some employees of losing the opportunity to hold their position for a number of years.
In addition, cloud recognition services are attracting crowdsourcing platformswhose operators check the quality of recognition and, in some cases, type texts from scans or photographs of documents. In this case, literally anyone can become an operator – just register with the service. In other words, complete strangers gain access to the data of your documents, passports or powers of attorney of your clients, who can dispose of such information at their own discretion – since crowdsourcing involves thousands of people, the developer company or cloud provider cannot physically control the actions of each of them. At the same time, some services do not hide the fact that they use the obtained data in training their neural networks. Accordingly, any organization that has used on-cloud solutions risks subsequently facing indignation and lawsuits from clients who have become victims of a pirated AI training method.
Finally, let us mention that the recent outage at Microsoft, which affected airlines, banks, emergency services and even fast food restaurants around the world, clearly demonstrated the unreliability of the on-cloud model. If a company does not want part of its business processes to be stopped due to an error in updating the cloud platform, it is better to choose on-premise.
Capabilities of recognition systems
Having checked the legal status of suppliers, having made sure that the software uses the company's own technologies, and having asked about the technical side of the solutions, the client should then, when choosing a recognition system, focus on what business processes and for what the software will be used.
For example, for banks, insurance companies and retailreceiving thousands and tens of thousands of documents every day, it will be extremely important that the solution works at high speed (for example, recognizes 15 documents per second on the server), has a set of documents whose recognition is already configured (and a designer for setting up forms for recognition, which can be useful in the case of unstructured or rare types of documents) and skillfully detects and correctly processes tables, stamps and barcodes.
For companies that need to verify the identity of customers, visitors, employees, etc. It is desirable that the software product supports passport recognition and other identity documents not only of the Russian Federation, but also of all countries of the world to speed up the service of foreign citizens.
For banks and other companies that use digital service channels, it is necessary for the solution to be able to find a document in photographs and recognize it, since it is often in this form that images of passports and other documents are uploaded by clients. In addition, to apply for loans or carry out other transactions, registration data is required, which in a Russian passport can be printed or handwritten, and therefore the software product must recognize not only typewritten text. It is also important for the banking sector that the recognition system can be launched on mobile devices, including budget ones. This can be useful both for on-site service and in online banking: instead of visiting a branch, clients have the opportunity to quickly send money transfers, update passport data or apply for loans in a mobile application or Internet browser without having to manually enter all the necessary data.
Checking the solution “on your own”
Finally, we would like to mention that before purchasing a recognition system from a developer, it is worth clarifying, is it possible to get the source code of the software solutionThis will allow you to audit the software product, check the operation of the algorithms, assess their vulnerabilities and decide whether the solution is suitable for the company's needs or not.
If such a request is refused, then it should be treated with suspicion: does the reluctance to share the code mean that the solution does not actually have all the functionality promised by the developer? In this case, it is better to contact another vendor who is confident in their software products and is ready to let you test them.
Instead of a conclusion
This is where our methodology, which is based on legal opinions and system analysis, comes to an end (we really hope you were impressed by it!). Of course, these are just friendly recommendations, and if you wish, you can skip all the stages. Of course, you don’t have to bother yourself with long interviews; you can choose the first recognition system you come across. And you might even get lucky – maybe!
Just don't forget: in such a scenario, the client company at its best In this case, it risks not being able to cope with the volume of work assigned to it and wasting its money. And it is not worth reminding what a negative scenario includes – we have heard about personal data leaks, scandals, discontent (and lawsuits) of users and multi-million dollar fines.
Does the company need it? We at Smart Engines don't know, but we are convinced that it is always better to be on the safe side. A thorough check of the recognition system developer will help avoid risks and problems in the future. And the final choice is up to you. We hope it will be the right one! 🙂
How our recognition systems work and our own proprietary OCR can be tried in our demo applications! No GPU, powerful servers, foreign OCR, clouds and especially no crowdsourcing with HITL operators. Only AI!
