Signature verification gets smarter by bringing Artificial Intelligence (AI) into the automation process. Today’s newest AI technologies are emerging from the private sector far more often than from government think tanks or exclusive corporate labs. Advanced signature verification software is, for example, capable of matching—and can even be superior to—human skills in verifying signature authenticity.
Your Unique Signature
Handwritten signatures may one day go the way of the Dodo bird to be replaced by other biometric identification methods, technologies and solutions. However, in the foreseeable future, traditional “wet ink” signatures will continue to be relied upon for identification and authorization. Handwritten or wet ink signatures are the most common way to complete legal documents of all types from checks and wire transfers to contracts.
Part of this reliance on signatures is the fact that it is unique to an individual. The challenges are in distinguishing between different peoples’ signatures and also in verifying that multiple signatures belong to the same individual. From one day to the next, an individual’s own signature can vary significantly. Signers and signature verifiers (humans) are both impacted by many factors such as mood, environment, the writing tool, the writing surface and level of fatigue or distraction to name just a few. Therefore, a certain level of variability between genuine signatures is expected and considered when banks, businesses and governments leverage human experts to verify signature authenticity or AI deep learning software.
Preventing Fraudulent Transactions
Relying on human experts to authenticate signatures is neither error-free nor cost effective. Unfortunately, this means that many businesses and financial institutions focus on having their staff verify only the highest risk, high-value transactions, which can leave those organizations vulnerable to fraud. Many examples are available directly from news headlines where fraud has been perpetrated through simply forging signatures. Forgeries can be divided into three basic categories:
- Random Signature Forgery. Random forgeries do not attempt to match the name or the style of the individual’s genuine signature. It is simply a signature (e.g., Tabatha Smith to John Prasad).
- Blind Signature Forgery. Blind forgeries match the signature name (e.g., John Prasad to John Prasad), but do not attempt to match the style of the individual’s genuine signature.
- Skilled Signature Forgery. Skilled forgeries are the most difficult to identify because they try to match the name and style of the individual’s genuine signature (e.g., John Prasad to John Prasad).
Signature verification automation software can play a critical role here in fraud prevention ensuring that all documents (checks and other signed documents) undergo signature verification. This can be extremely high volumes of documents—many 100s of thousands daily if not more. There are many instances where signature verification is necessary in automated fraud management, transaction authorization, absentee ballots, voting registration, timesheets and much more.
The entire automated verification process itself can be compared to the work of a group of highly skilled experts. Each verification process uses a specific approach and set of algorithms, looking at particular characteristics, which is especially efficient in some cases and “good-enough” in others. Once these approaches are combined and optimized, their areas of expertise complement each other, which results in excellent overall performance and accurate fraud detection. Signature verification software is very efficient, accurate and consistent. It has proven to outperform humans on even the most difficult types of forgery. Software can use many different authentic reference images to improve accuracy; humans typically do worse when dealing with more than 2 or 3 reference signatures.
And yet, humans continue to be integral to signature verification and fraud prevention. Human scrutiny is highly analytical and complements the steady, unwavering predictability of software algorithms. To eliminate fraud, the best solutions use software verification automation and human exception-handling that reviews less than 1% of signatures in cases where the software reports low confidence levels.
How Does AI Succeed Over Humans?
Several different techniques and technologies are employed to mimic the best of what the human brain can achieve, which would take a book to describe in sufficient detail. At the highest level, though, there are simply three key areas to highlight that signature verification AI software uses:
- Special descriptive language
- Signature segmentation, and
- Neural networks.
These three key elements, when combined with a stellar “supporting cast” of other complementary techniques, drive a reliable and thorough verification “engine.”
Heading into the Future
As we continue to rely on wet ink signatures, the role and responsibility of signature verification automation will be to eliminate all the negative aspects of human verification. Software doesn’t get tired or have a bad day. It can process dramatically higher volumes of signatures at a fraction of the time and cost, as well as consistently demonstrate higher accuracy.
For skilled forgeries, the most difficult, where the fraudster understands common ways to copy a signature and may even implement variations just to try and fool deeper analysis, these most rare instances that confound even the most advanced AI will continue to require the eye of a highly trained signature verification expert who can analyze the signature based on the individual’s genuine signature and validate whether the signature in question is truly authentic.
Mark Gallagher is the Vice President of Sales at Parascript, with over 25 years of experience in the industry, leading the Parascript Sales team in the US and abroad.