Member’s Corner: Accuracy Challenge in Receipt Scanning: A three-part series Part 1 Receipt Scanning Fraud and Misredemption

Steven Marcus, Mobisave

In a recent conference call sponsored by ACP I raised the question of manufacturer’s expectations for accuracy in receipt scanning. I heard one voice say 95%.  Even though it’s only one voice I’ll take that as a consensus because while it sounds like a high hurdle when compared to traditional paper coupons (whisper number is 85%), it is a worthy goal.  So to help reach that goal I’m going to outline some of ways we work to improve accuracy in the hope that our receipt scanning colleagues will do the same. And the end result will be better for all our clients. Or said another way, help grow the credibility of receipt scanning as an alternative to traditional paper…a rising tide lifts all boats.


Defining Accuracy

First, we should define accuracy and hopefully use it as a standard in the industry.  It is a given that none of us in receipt scanning can examine in depth 100% of the receipts we receive.  And if we are willing to process receipts from any retailer that scans and itemizes there will always be text that is either generic or so cryptic that that even the most sophisticated algorithms cannot decode it with any reasonable level of confidence.


Accuracy is also dependent on the requirements for the offer.  The more exacting or complex, the greater the chance for error.  On the extreme end, if an offer requires a multiple purchase of one variety of many and an exact size and a minimum price paid and cannot be used (stacked) with other coupons, it will take an army of manual processors to examine these receipts and even then they could be guessing at the meaning of some of the short product descriptions or the list of generic coupons.


At MobiSave we work with clients to find the right balance of exactness to assure we meet their accuracy goals. Often we may be asked to require the purchase of sub-brand but they also will accept the purchase of a brand. As such, accuracy in this case is measured not by the terms of the offer but by what is acceptable within the latitude provided by the manufacturer at a high confidence level.


The basic reasons why receipt scanning is more secure than paper coupons

The core pillar of receipt scanning is the direct connection to the actual redeemer of the offer and if they use PayPal or something similar, having the redeemer’s identification verified by linking it to a banking relationship.  A second core pillar is knowing the unique ID of the smartphone being used to submit the offer.  If there are two typical characteristics of fraudsters, they desire to be anonymous and are looking for the opportunity to make a big score worthy of the effort and the risk.  Because receipt scanning has neither, the sophisticated perps that scam paper coupons most likely stay away from receipt scanning.  Where we’ve seen attempts at fraud, it appears to be done by amateurs rather than organized fraud we see in paper. Here are some examples.


Counterfeiting or altering receipts

Our system is highly automated with most receipts machine read without manual intervention. While others might crowdsource reading receipts, we all have the same issue with professionally done counterfeits. We’ve caught them several ways. First, we look for outlier behavior where the number of redemptions on the receipt or the frequency of submissions is way outside of norms. We recently examined in detail the receipts of 55 of our highest outliers.  We only found one in this group where we saw evidence of fraud. The evidence was in the form of the matching of the font where the rendering of the number “9” did not match known authentic receipts for that retailer.


The second way is we look at the format of the receipt including the spacing and the alignment of standard text. Even though these receipts look official to the untrained eye, when reviewed side-by-side or overlaid on an authentic receipt, the counterfeit become obvious.  We have a few other checks such as the nomenclature and the format of the date/time that also provide telltale signs of counterfeiting.


The third telltale clue is receipts are too perfect.  Even the most careful of our members unintentionally roll or fold or wrinkle their receipts.  If we see a perfect image of a perfect receipt, chances are it is counterfeit.  We confirm it by contacting the retailer if it says Bob’s Grocery (actual example) or checking Google maps for the actual location which often does not exist.


If we see evidence of counterfeiting or alteration, this is a case of “one and done.” The member’s account is blocked and their smartphone ID cannot be registered under a new name. In other words, we’ve blocked the user and their phone.


Submitting old receipts (aka dumpster diving for receipts or buying old receipts)

MobiSave requires that users select offers prior to checkout. We do this to make certain the incentive drives the sale rather than the reverse.  We consider noncompliance with this requirement as misredemption.


To make this happen, our system must read date/time on the receipt and compare it to the date/time of selection. We machine read date/time on receipts successfully in about 80% of the cases. We are working to push this higher but we know we’ll never be 100% because of the limits of OCR machine reading. Here’s where admittedly crowdsourcing seems to have an edge over machine reading. The question is whether the added expense and delayed reward by using crowdsourcing is worth it in terms of costs/fees charged to manufacturers and shopper acceptance of receipt scanning. We think not and here’s why.


As part of our QC process we look at a sample of the 80% of the receipts that pass the date/time machine check and look for false positives.  If we see we are making errors, we tighten the criteria and raise the confidence level required to accept the date/time. We reject noncompliant receipts where our system has a high degree of confidence in the date/time read.


For the 20% of receipts where we are not confident in our machine read we generally accept these if the member has provided a good image.  Our goal is to reduce the chances of false negatives to near zero. But, we also manually review a sample of these and have found the date/time compliance rates are near 100% for experienced members.  Non-compliance typically occurs with new users who have used other receipt scanning systems that do not require pre-selection. In those cases we give them slack with a reminder for future submissions if they are off by less than a day.


So some can slip through but we catch most of them and members who are chronically noncompliant are warned and unless they change their behavior are blocked. To help restore your faith of the honesty of shoppers, we have only had to block .0007% of our members. Unfortunately, the hand few of bad apples have taken to giving us negative reviews in the app store.  We feel that’s a small price to pay to assure the quality of redemptions and keep our promise to shoppers to pay immediately.


Submitting duplicate receipts

Regardless of whether we use machine or manual processing, we all have the opportunity to capture sufficient information on the receipt to detect duplication submissions. Our initial pass looks for duplicates by matching the date/time, the name of the retailer and total till.   We also do a secondary check by looking for the same claimed rebates. So even if the member uses a different smartphone, we can spot duplicates.


Submitting the same receipt on different phones to us is a capital offense punishable by banishment from MobiSave.  But, submitting the same receipt on the same phone could be an honest error where they forgot to claim a rebate. And since our system does not allow reusing offers, they can’t claim previously used rebates. So here’s where we show some flexibility.


The next installment in this series will discuss image processing, OCR and matching accuracy.  The third and final installment will recommend steps that the receipt scanning industry, manufacturers and the ACP can take together to get us all to the desired 95% level. ACP members will find the second and third installment in the member area of the ACP website.  Non-members can contact John Morgan, for access.


About the author

Steven Marcus, founder and president of MobiSave, has spent a lifetime in consumer marketing in consumer packaged goods, marketing services and financial services.  He is an entrepreneur and an Internet pioneer having digitized a multiple rebate system in the mid-1990’s that became the basis for the receipt scanning applications.  He served on the USPS anti-fraud rebate task force.  You can contact Steve at


Leave a Reply

Your email address will not be published. Required fields are marked *