Image Is Everything

By Tom Jordan, VP-Financial Software Solutions, Digital Check Corp

Tom Jordan, VP-Financial Software Solutions, Digital Check Corp

Scanning ATMs have changed the way banks take deposits–but also introduce some under-the-radar risk. For decades, making a deposit at the ATM typically worked one way: You put your money in an envelope, slid it into a slot in the machine, and that was that. Fast-forward to today, and envelopes are a rarity, as banks have modernized most of their networks with image-enabled ATMs that scan cash or checks on the spot.

"Improving on-the-spot detection capabilities is key to fighting a new wave of fraud, and having a good image is key to doing that"

The change solved a big problem for banks–namely, not knowing what was in the envelope. So-called “empty-envelope deposits” used to rank as one of the leading forms of ATM fraud, while simple mistakes by the customer caused countless next-day balance adjustments. Dumping the envelope has all but eliminated both of those issues.

On the other hand, the envelope’s biggest strength–that the deposit process worked without fail has proven somewhat of a thorn in the side of the bill-scanning ATM. For various reasons, capturing an image of a check is not a foolproof process: It might be folded or torn; it might have faint printing; it might have a vivid background that interferes with the scan. And some documents, such as money orders, intentionally make use of security features designed to thwart attempts to photocopy them – the problem, of course, being that they also thwart attempts to scan them.

However, since the customer’s previous experience was that the envelope went through 100 percent of the time, there’s tremendous pressure to make every scanned transaction go through immediately, whether the image capture went cleanly or not. So when the machine can’t read the key information on a check, it might ask you to verify the dollar amount using the keypad, then force through the transaction. Sometimes with a really bad image, the machine will reject the check outright and tell the customer to bring it inside to a teller; but generally, banks try to avoid that at all costs, so the standards are definitely relaxed.

That’s all well and good for the customer – but a problem for the bank, since it’s left with an image that may or may not be any good. When it comes to exchanging that image with the bank whose name is on the check, there’s a significant chance it will bounce right back as an illegible “non-conforming item” (NCI), racking up fees and requiring a manual investigation. Or, if the dollar amount was actually different from what the customer entered, if you can’t see the real amount, you’re at risk for fraud, balance errors, and unhappy customers.

Here’s where it gets tricky, though: Every check is captured as a high-resolution grayscale (or sometimes color) image, but the official legal format is a 200 dpi black-and-white TIFF file. Really, we should say black-OR-white, because those are the only two colors you’re allowed to have. Each pixel is either “on” or “off.” As you might expect, quite a bit of detail is lost in the conversion process, meaning that even if the grayscale is readable, the black-and-white image might not be.

Without a doubt, it’s better to work with the grayscale image instead of the black-and-white whenever possible, since the original detail is preserved. So if you’ve ever had an ATM ask you to verify your deposit amount, the black-and-white image is what the machine tried to read, but the grayscale image is what it will show you. However, due to the way most ATMs’ “software stack” operates, the grayscale image is thrown out as soon as you walk away from the machine, leaving only the low-quality black-and-white. If there’s a problem later, that’s all there is to go by, short of digging through the pile and finding the original paper document.

Why was it set up this way? Well, back when scanning ATMs were first being developed, self-serve deposits were not a huge portion of the total–maybe a few percent for most banks. That has surged in recent years, though, with some of the financial institutions we work with reporting half or more of all deposits coming in through ATMs.

What was a minor nuisance at 3 percent can turn into a crushing burden at 50 percent, so lots of banks and credit unions were left scrambling after the fact for a way to clean up those difficult images before they start causing problems down the line. Moreover, when half the deposits move to a new channel, you can expect half the fraud to move along with it; and so the ATM has become a prime target for alterations and forgeries designed to fool machines rather than the human eye.

Much of this fraud relies on the fact that–just as with “can’t-read” errors–transactions are likely to be forced through immediately and sorted out later, and by that time, the money is already gone. Improving on-the-spot detection capabilities is key to fighting that wave fraud, and having a good image is key to doing that.

Plenty of equipment manufacturers have ways of enhancing image quality, generally by adjusting light levels and contrast to get the best picture. But ATMs present a special challenge, since all the adjustments 1) need to be made immediately within the brief window when the grayscale exists; and 2) need to happen without any human intervention. So it’s a daunting task to make software “smart” enough to do that cleanup in real time without any help.

Furthermore, ATMs tend to come with their own software stacks included by the manufacturer, then the banks and their systems providers layer their own software on top of it. Then the image-cleanup software has to be added in and interact with that. If that’s not enough, many banks have multiple types of ATMs of all different models and ages in their networks–so any image enhancement has to be more or less universally compatible or it’s a no-go.

But the banks (and the hardware manufacturers) do have one thing going for them here. Most problems with image quality repeat themselves–that is to say, the same few kinds of documents tend to keep causing the same kinds of bad images over and over. Most banks tell us that less than 20 different document types cause 90 percent or more of their errors for non-conforming images.

So, one key tactic employed at the ATM is to read the account and routing information contained in the MICR line at the bottom of a check or money order, and compare it to a database of known “difficult documents.” If the program identifies the document as, say, a postal money order, it can apply a predetermined set of filters to clean up the image and head off problems before they start. Other (at this point still theoretical) options involve flagging any document that required customer re-keying, and preserving the grayscale instead of wiping it clean, so that it could be evaluated or re-adjusted without finding the original paper.

While scanning at the ATM has generally made an overwhelmingly positive impact on banks and consumers–after all, these checks and dollar bills were previously all being handled manually at some point with the old envelope-based process–any time new technology takes hold, it tends to have a surprise or two in store for you.

In this case, fortunately, there’s a way to fill in the gaps. And as banks begin experimenting with “branch of the future” concepts, where the line between self-serve and teller-assisted functions blurs, and employees have some access to the original documents, that’s looking to get easier. Just remember next time you’re making a deposit at an ATM–there’s a lot more going on behind the scenes than you think.

Read Also

Why Insurtech and Insurer Collaborations Often Fail

Why Insurtech and Insurer Collaborations Often Fail

Manjit Rana, Managing Director, Corporate Innovation, InsurTech Business, Startupbootcamp & Rainmaking Innovation Ltd
How Long Before the Door Closes on Open Banking?

How Long Before the Door Closes on Open Banking?

Andrew Cregan, Policy Advisor (Head of Payments Policy), British Retail Consortium
AI and Humans: A Winning Combination

AI and Humans: A Winning Combination

Matthew Key, Head of Customer Innovation, Global Banking and Financial Markets, BT
Robotics Process Automation (RPA) - Experiences from Today, Looking Towards Tomorrow

Robotics Process Automation (RPA) - Experiences from Today, Looking Towards Tomorrow

Jonathan Kidd, Head of Digital Operations & Robotics, Bank of Ireland

Weekly Brief

New Editions