WO2022180230A1

WO2022180230A1 - Claim interrogator

Info

Publication number: WO2022180230A1
Application number: PCT/EP2022/054836
Authority: WO
Inventors: Jan LOVMAND; Lars Torp; Rasmus Bækgård HOLM
Original assignee: Detectsystem Lab A/S
Priority date: 2021-02-26
Filing date: 2022-02-25
Publication date: 2022-09-01

Abstract

There is a need for better identification of fraud in the insurance industry. This is achieved by providing a method comprising the steps: providing a digital channel through which a policy holder can file a claim request for an insurance pay-out, receiving a claim request (110) through said digital channel, the claim request (110) comprising a claim type (111) and at least one digital document (120), retrieve a list of tests (131 ) from a database (130) based on the claim type, analysing the metadata of said digital document using a processor executing the list of tests, the tests at least indicating whether the document metadata meets metadata parameters of the claim type of the claim request, optionally analysing whether document content corresponds with information filed with the claim assess (140) whether the claim request (110) is legitimate based on said executing the list of tests, transmit a signal indicative of the assessment, where the signal is one of a claim verification signal (161 ), a claim rejection signal (162) and a claim review signal (163). Provide a digital interrogation step on a review signal to resolve found anomalies.

Description

CLAIM INTERROGATOR

FIELD OF THE INVENTION

The invention relates to a method of assessing claim request legitimacy, a device instructed to perform the method, and a program and a computer-readable medium with instructions for carrying out the method.

BACKGROUND OF THE INVENTION

Fraud is a global problem that affects not least the insurance industry. It is estimated that 10% of all insurance pay-outs are made to fraudsters. To receive an insurance pay-out, various documents must be presented to validate the insurance claim. Even then, there are loopholes. Fraudsters seek to cheat insurers in a plethora of ways and today, fraud has moved into the digital arena too.

Digital documents introduce a variety of new ways to cheat and commit fraud, not least insurance fraud. Verifying the uniqueness, ownership and authenticity of documents and items is very difficult when documents are digital files. In the past, digital rights management has been used on some file types to ensure that they were not copied, although the inconvenience thereof made it infeasible, and so insecure documents are here to stay. Verifying the uniqueness, ownership and authenticity of documents and items is the job of insurance investigators, who make value-judgments about documents throughout their workday. The more scrutinous they are, the slower and more expensive insurance pay-outs and premiums get. Some insurance companies have decided to solve this by being slack with verification and accepting as high as 20% fraud, since this allows them to have fewer investigators and so retain operative costs low.

Fraud raises the investigative burden on insurance companies across the board, since it is not possible to know the fraudulent claims ahead of time. The more scrutinous the investigators are, the higher premiums get, and the longer pay-out processes become while performing too little investigating attracts fraudsters.

Therefore, there is a need for a better insurance investigation method.

SUMMARY OF THE INVENTION

In an aspect of the invention, there is provided a method comprising the steps: - receiving a claim request 110 through a digital channel, the claim request 110 comprising a claim type 111 and at least one digital document 120,

- retrieve a list of tests 131 from a database 130, at least one of which tests or the combination of tests being selected based on the claim type,

- analysing the metadata and optionally document content according to submitted claim details of said digital document using a processor executing the list of tests, the tests at least indicating whether the document metadata meets metadata parameters of the claim type of the claim request,

- assess 140 whether the claim request 110 is legitimate based on said executing the list of tests,

- transmit a signal indicative of the assessment, where the signal is one of:

- a claim verification signal 191 that is transmitted if the tests determine the claim request to be legitimate,

- a claim rejection signal 192 that is transmitted if the tests determine the claim request to be illegitimate and

- a claim review signal 193 that is transmitted if the tests are inconclusive as to claim request legitimacy, where the claim review signal comprise at least one resolving question specific to a data anomaly that resulted in the claim review signal being transmitted, where for certain answers given to the resolving questions by the policy holder, a claim verification signal or a claim rejection signal is transmitted.

Thereby claim requests can be handled in a much more streamlined manner where each claim request is judged precisely by the same parameters. Therefore, the policy holders can be guaranteed a handling time that is more predictable, also ahead of time since each claim request ahead of the current one can have handling time estimated in an anonymous manner.

Furthermore, the insurance investigator performs fewer mouse clicks / operations to fully handle claim requests, even each type of claim requests.

For all claim requests where everything checks out, i.e. , for claim requests where document metadata and document content match up with expectations for the claim type and filed information, the claim request and pay out process can be automated and fulfilled completely without human intervention or, if desired, a step of human verification can be added on top to further ensure process quality.

For claim requests where there are certain document or data incongruencies / anomalies, a claim review signal is transmitted, and ultimately presented to the policy holder with at least one follow up question specific to the anomaly and claim type in an effort to clarify any outstanding errors, mistakes or mismatches.

By document, digital document and file is meant a digital file that can be opened and perused by a consumer or citizen to visually identify data contents. The data contents are descriptive of a product or person, and indicates ownership/authority and/or evidence of an event such as a burglary or fire. The document may comprise image or text in any shape, such as an image file or a text file. The document may indicate ownership or status of the item or product that the claim is concerned with, such as a stolen watch, a burnt computer or a destroyed door.

The policy holder is the person that has a contract with an insurer to receive compensation for certain damages, losses and given certain situations. The policy holder will typically be a natural person or a group of natural persons, but it may also be a legal person.

Insurance investigator, investigator and claims handler signifies persons who are tasked with handling the claim requests made by policy holders. There may be two types of actual tasks performed, one being a primary, administrative investigation, and the other being an explicitly fraud-oriented investigation, which two tasks are often performed by two different types of employees and in two different departments. The method of the invention may help both types of investigators, and help sort claim requests between the two. Where the administrative investigator is discussed alone, these are termed primary investigators. Where the fraud investigators are explicitly meant, these are termed fraud investigators. In an embodiment, at least one test from the list of tests is a selected based on document type of the digital document. For image files, image manipulation tests can be conducted while for portable document format (PDF) files, fonts and text fields may be analysed, for example. Geolocation of an image file may be compared to a burglary address, and creator program may be compared to a shop invoice generator date for a watch invoice. In an embodiment, the method further comprises tests ensuring that document contents correspond with information filed in the claim. An example would be that an invoice includes the correct phone number of the issuer or that a certificate or health record contains the name and/or address of the policyholder.

In an embodiment, the tests further indicate whether or not at least two metadata signifying overlapping information matches with each other. Thereby, further tests are conducted to assess the integrity of the metadata. It can be determined where relevant whether the file is corrupted or tampered, which may not only identify relevant information to use in follow up questions for the policy holder, but prospective fraud can be identified. Examples of overlapping data is the metadata date and a metadata operating system (OS) of the produced file, which can at least not be older than the release date of the OS in question. There are other such metadata which provide overlapping information in this sense.

In an embodiment, the method further comprise a step of interrogation, wherein the at least one resolving question is part of at least one series of related questions presented to the policy holder through a digital questionnaire through a digital channel, where answers to prior questions of the related series of questions determine subsequent questions posed in a hierarchical manner. Thereby, much more fine-tuned data collection can be achieved before involving the investigator or even ensuring that the pay-out does not need to pass through the hands of the interrogator at all. For some embodiments, the interrogation step converts some claim requests from ‘review’ category to either of rejected or verified. For other claim requests, at least more data is obtained on the important points of interest, i.e. , the anomalous / mismatching data. The investigator normally have to peruse the claim request, identify that there seems to be anomalies in the provided data, and then contact either the policy holder or third parties to ask follow-up questions. This is prone to human error and fraudulent claim requests slips through the cracks of even talented investigators. Furthermore, legitimate claim requests may have certain defects that must be clarified before a pay-out can be made, and these can be remedied by the method as described. Legitimate pay-outs may thus be handled much faster and victims of theft, burglary and other mishaps do thus no longer have to wait in investigation queue for fraudsters to be caught. The digital channel can be any type, such as a telecommunications line, an internet- based connection or any other type. It can take the shape overall of a robotic telephone questionnaire, a website integrated questionnaire, a chain of telecommunication- or internet text messages, such as short message service messages or emails. The principal thing is that the interrogation step comprises a series of related questions, where the answer to a question in the series of questions determines which follow up questions are selected and transmitted to the policy holder.

In an embodiment, if a claim verification signal is transmitted, the policy holder receives said claim request verification signal immediately following filing the claim request. Thereby, an especially fast pay out process is achieved and the number of contacts with the policy holder can be minimized and the turnaround time handling a claim with human intervention brought down to a minimum.

In an embodiment, the tests further comprise claim reporting tests, whereby data specific to the claim request filing is gathered to inform the assessing whether the claim request is legitimate, such as data relating to the flow of the claim reporting or the timing of the claim reporting. Thereby, suspicious data can be collected and used to contextualise whether the filing is indeed conspicuous.

In an embodiment, the tests further comprise image recognition tests to inform the assessing whether the claim request is legitimate, whereby a digital document being an image is tested for tampering such as by comparing the image with images in a database to identify at least partial matches. Thereby, images may be tested to assess whether they have been at least partially taken from some database.

In an embodiment, assessing whether the claim request is legitimate based on said executing the list of tests comprise a dynamic signal threshold, where which type of signal is transmitted, depends on said dynamic signal threshold that changes over time. Thereby, various contextual factors are taken into account to maximize value of the time of insurance investigator. If they are overworked, more lenient thresholds may be used whereby claim requests that would otherwise prompt review are accepted. In an aspect, the invention relates to a computing device having a processor adapted to perform the steps of the invention.

In an aspect, the invention relates to a computer program comprising instructions which cause the computer to carry out the method of the invention.

In an aspect, the invention relates to a computer-readable medium comprising instructions which cause the computer to carry out the method of the invention when executed by a computer.

BRIEF DESCRIPTION OF THE DRAWINGS

In the following, example embodiments are described according to the invention, where

Fig. 1 is a schematic of a method of an embodiment of the invention,

Fig. 2 is an illustration of a suspicious burglary claim document,

Fig. 3 illustrates an interrogation step of an embodiment of the invention,

Fig. 4 illustrates a computing device for carrying out a method according to an embodiment of the invention, and

Fig. 5 shows a claims dashboard according to an embodiment of the invention.

DETAILED DESCRIPTION

In the following the invention is described in detail through embodiments hereof that should not be thought of as limiting to the scope of the invention.

Fig. 1 is a schematic of a method according to an embodiment of the invention.

The figure generally describes a method for assistive insurance claim investigation. The method of assisting in insurance claim investigation comprises a series of steps outlined in the following.

Firstly, a claim request 110 is received through a channel from a policy holder relating to a pay-out. The policy holder fills out what claim type 111 it is such as a burglary, a fire or a theft, and attaches a number of digital documents 120 with the claim request 110 to document the claim. The documents are needed to verify ownership of the damaged or stolen items.

The method then reviews the claim type 111 of the claim request 110, and retrieves document content and a predetermined list of tests 131 from a database to conduct on the digital document. The list of tests is at least two individual tests.

A processor then uses the list of tests 131 on the digital document 120 to analyse the claim request, using at least the metadata 121 of the digital document itself. The tests can be different types of tests, at least some of which examine the metadata of the digital document to assess whether it fits according to the specific claim type. For example, for a stolen watch it would be assumed that an image document of the watch attached to / included in the claim request would have a date predating the occurrence of the theft. Likewise for a stolen watch it would be assumed that the invoice document showing the purchase of the item would have the name or other information identifying the purchase to the policyholder.

Based on the outcomes of the analysis, the processor assesses the legitimacy of the claim request 110.

Lastly, the method transmits a signal indicative of the assessment of the claim. Preferably, the signal is transmitted to a claim investigator for approval or review. The signal can be one of: a claim verification signal 191 , a claim rejection signal 192 and a claim review signal 193. The claim verification signal 191 will be transmitted when the digital document 120 passes the tests. The claim rejection signal 192 is transmitted when the digital document fails the tests, strongly indicative of fraud. The claim review signal 193 is transmitted in ambiguous situations, or where more information is needed to come to a conclusion. The claim review signal comprises or is accompanied by at least one resolving question specific to the result of the assessing. The at least one resolving question is determined by the ambiguities / data anomalies to be resolved, and the questions are presented to the policy holder for answering. This can be through an e-mail, an SMS or a series of SMS, a robo- call, or through a live questionnaire presented directly in an internet browser window. Thus, ambiguities of the claim documents, such as the metadata-related ones, are resolved quickly and effortlessly and the needed number of clicks for the insurance investigators are drastically reduced. The claim rejection signal preferably also results in at least one resolving question. Instead of the at least one resolving question attempting to clarify an ambiguous situation, however, when prompted from a rejection signal, the at least one resolving question is intended to allow the fraudster to backtrack or further double down on his/her fraud. in the latter case, the specific ambiguities have been sorted and marked for an investigator and will thus produce faster and better claim investigations.

Figs. 2 and 3 will explore a specific example claim request instantiation for the sake of understanding. In this example, the policy holder files a claim request relating to a burglary at his home. The reader will understand that the method naturally applies for different claim types and so on, as well.

Fig. 2 is an illustration of a suspicious burglary claim document 120. For a claim type of burglary, the documents attached to the claim are expected to be receipts or images of the stolen items. Images and receipts would be expected to be older than the date of the burglary, and images would be expected to be taken at the scene of the crime, i.e. , the home address. A database test will investigate the GPS data of the photo metadata and compare the event location of which the policyholder has reported the burglary have taken place.

For one of the designer chairs that the policy holder supplies, e.g. the metadata does not match the claim type. Firstly, the date of the image is four days after the theft, which flags the claim as suspicious. It is not expected to be possible to take a photo of a chair if it is indeed stolen. Further, the geolocation of the image does not match the address of the burglary, which also flags the claim as suspicious. Since a burglary is tied to one location, the items being taken must be confined to that area as well. Of course, these are not necessarily evidence of fraud, since legitimate reasons may exist. Flowever, they fall outside of what is expected.

Depending on the value of the item, the insurance investigators may today simply make the pay-out without ever checking the metadata or document contents. Alternatively, when the insurance investigators find that the price of the items are high enough, they will manually investigate the metadata to see if they match. Only in the latter instance will the insurance investigator find out that the claim could be suspected of fraud.

This means that it may be relatively trivial for fraudsters to file claims of individually insignificant sums to evade scrutiny but derive pay-outs even using fraudulent documentation.

Assuming that the defects or anomalies are identified as discussed, the method as hitherto described can send a review signal to an insurance interrogator indicating that the claim request is not trivial to process because certain anomalies exist, and furthermore it can pinpoint the specific deviations. This review signal may initiate the review process, ensure that the investigator quickly identifies the relevant issues to investigate and assist the investigator in asking the proper questions in the interrogation process. This may not alone speed up the process but also help the investigators focus, and by taking work off their tables, may allow them to look through more claim requests, which allow catching more fraudsters. In this sense, the method can be considered at least a pre-screening process that parses through the information and presents only the most pertinent decisions for the investigators to handle.

Fig. 3 is a schematic of an interrogation step of the claim assessment process according to an embodiment of the invention. The step of interrogation may be performed to resolve some of the ambiguous conditions before transmitting a signal. The signal is preferably eventually transmitted to the insurance interrogator. However, before transmitting the signal and after performing the initial list of tests, the interrogation step may be performed.

The method will be explored and discussed with outset in the burglary example already explored for Fig. 2. The specific interrogation structure will vary depending on specifics as already described. For the contemplated example, as mentioned, a policy holder has filed a claim relating to a burglary and theft of a number of pieces of designer furniture. One of these pieces of furniture had metadata that was suspicious when seen in combination with the claim type of a burglary. On conducting the test, an anomaly may be found in the photo metadata that raises a suspicion. However, the situation may arise from a number of different scenarios. One scenario is that there was no burglary at all, and that the claim request is entirely fraudulent.

Another scenario is where there is indeed a burglary, but the policy holder becomes greedy, and decides to add false items to the claim. He may have gone to a furniture store, and taken a picture of a chair that he then adds to his otherwise legitimate claim request to increase his pay-out.

A third scenario is that when asked to provide photo evidence of the stolen items, the policy holder is not able to find any such image and goes to a store to take a picture of the model of chair.

The interrogation step then gathers further information to assist pinning down which of these scenarios is the assumed right one.

So, as part of the interrogation step, the photo with the anomalous metadata is shown to the policy holder, while a number of questions are asked. A first series of questions could relate to the age of the photo, while a second series of questions could relate to the location. The order in which these are posed may be carefully coordinated according to a predetermined order, or they may be any order.

In any case, a first question 171 posed relates to whether the photo illustrates the item that was stolen during the burglary. If the policy holder answers in the affirmative 171 A, this is suspicious. Continuing in the same series of questions, the policy holder may then be asked a second question 172 relating to how old the chair was when the photo was taken. Perhaps a ‘do not know’ option is allowed here. This may in turn allow selecting an option of ‘this is not my own item, but a similar item’ 172B. Inputting any age answer 172A is highly suspicious, and strongly indicates that the policy holder is committing fraud, since this further cements that the policy holder is willingly informing that the chair is stolen while the metadata informs that the specific chair in the image was present after the burglary. These types of questions are considered implicit questions. The information derived from an implicit question that helps determine which signal to transmit is not the information explicitly asked for by the question. I.e., ‘how old is the chair’ receives an age as answer. However, it is not the age of the chair that is of primary interest. The primary point of interest here is to ask ‘is this really the stolen chair(7), the answer to which helps cement the fraud, since the fraudster is repeating the fraud or providing further fraudulent information, which makes backtracking out of the interrogation harder. However, asking ‘is this really your chair?’ is likely to alert the fraudster that the method is onto him. Thus, indirect questions seem more innocuous. In a preferred embodiment, the interrogation comprises at least one implicit question that help determine which signal to transmit based on data implicit in its answer.

After the date-related series of questions 171 , 172 are concluded, a second series of questions relating to the location of the chair are conducted. A third question 181 is posed as to whether the image was taken at location of the burglary. If the policy holder answers in the affirmative, then this further cements the fraud, since it does not match with the metadata geolocation of the image. If not, then a line of questions may arise that moves towards a less suspicious result.

The interrogation step is preferably performed by connecting an electronic device such as a smartphone or a computer to a database to produce a live interrogation, such as through the internet. This allows seamless interrogation flow, for example implemented on an insurance company website. Other useful implementations are through other digital channels, such as through an automatic robotic telephone call, a short message service (SMS) or a string of such SMS.

It is also possible within the scope of the invention that after any suspicious data / data anomalies are identified, that the list of resolving questions are generated for the insurance investigator to conduct manually, such as through e-mail or over a telephone interview.

The questions 171 , 172, 173, 181 are example questions posed in a manner that are supposed to clarify the issues surrounding each anomaly. The discussed pattern of answering may prompt the transmission of a claim rejection signal. An investigator may preferably go over the interrogation and contact the policy holder to further cement the fraud before potentially involving the police or other authorities about insurance fraud.

Alternatively, the questions may also let the policy holder explain the anomaly. The interrogation goes another way if the first question 171 is answered in the negative 171 B. If the policy holder answers in the negative on the first question 171 relating to whether the image shows the stolen item, the conditions about why the image was uploaded are investigated.

The interrogation method may simply ask a fourth question 173 relating to whether the item is a similar looking item. Answering in the affirmative here may constitute an acceptable answer that moves the assessment towards verification. If the policy holder realises that the image is a wrong image, they may be allowed / prompted to upload an additional image, starting the interrogation over if data, such as metadata or contextual data still mismatches / is anomalous.

The policy holder may then be asked where they have the image from if it is a similar item, or what is on the photo if it is not. If everything is sufficiently explained, the interrogation step may culminate with a verification signal.

Some claim requests may be completely resolved by the interrogation step. This may be the case for some or most claim requests, but is unlikely to be the case for all claim requests. For claim requests not finally resolved, the insurance investigator gains a lot of useful and structured information when starting from the framework of the completed or partially completed interrogation step.

Preferably, more metrics than simply the answers are used as well. Example data sources can be whether the policy holder uses more time on one answer than the others, overexplains himself, goes back and forth in the digital navigation or between individual buttons on a page, and/or transmits a claim request at anomalous times, these are all indications that can be used to contextualise the claim request legitimacy. Most of these interrogation answers are preferably at most supplementary to the assessment, since many other events can explain divergences. However, a granular trust score can be attributed to the interrogation and even to the overall assessment, which may inform whether the claim request is automatically processed or flagged for interrogator review.

Although the focus has been on metadata of the supplied documents, the method may take contextual data into account as well. Contextual data is data like insurer policy age, pay-out frequency and amount, policy outstanding payments, as well as file/document context. File/document context is whether the file/context is recognised from another insurer (such as whether a hash matches a hash in a database). All of these contextual data may be used to perform certain of the tests. Furthermore, the document contents can be included in the analysis e.g. does a picture of a stolen watch actually display a watch and/or does an invoice include the relevant expected keywords usually present in invoices.

Fig. 4 illustrates a computing device 150 for carrying out a method according to an embodiment of the invention. The computing device 150 comprise a networking interface 152 for communicating with other electronic devices, as well as at least a processor 151 and a database 130. The processor performs the steps of the method of the invention. The database 130 comprises a number of claim types 111 and for each claim type 111 , a list of tests 131. The list of tests 131 for a given claim type determines for a claim request made with the given claim type, whether the supported documents / files matches the expectations for that specific claim type.

For example, it can be tested whether images have location and dates that matches, or whether purchase documents have matching dates. The metadata of the supplied documents are tested against the specific expectations of the specific claim type. Since the expectations vary significantly among claim types, it is possible to test much more granularly by contextualizing tests to the specific claim type.

For example, for burglary, any documentation of the stolen items is expected to have an earlier creation date than a burglary date, while an image of a broken- through door of the same burglary is expected to have a timestamp later than at the time of the actual burglary. The metadata of a file will have a creation date and location, and other information, like OS version, that may further pinpoint dates or date ranges for example to help assess whether the claim is suspicious. Many specific tests can be performed to further test files in other ways as well. In certain embodiments, the tests relate to at least one of document/image creation date, creation geolocation, hardware identity, hardware age, image colour profile, OS version that indicates age or creation application if applicable.

The method may assess claim request by using different types of tests too.

In an embodiment, the method comprises claim reporting flow tests 132.

Certain behaviour or aspects of the claim reporting flow may be suspicious. If a policy holder is in the middle of a claim reporting but never finalises it only to restart a new claim request that is similar, this may indicate that the policy holder stopped in the middle of an act of fraud. By itself this does not inform of fraud, and should be used only in context with something else, such as a ‘legitimacy blind alley’ part of the interrogation step described for Fig. 3. The interrogation step may place the fraudster in a position where he realises that he is being asked to the specifics of the fraud he is committing, and he may stop the reporting. If at a later point he reports on a similar or identical claim, this may be taken into account.

Another claim reporting flow test relates to the timing of filing the request. A couple of different time-related parameters may be taken into account.

Firstly, in an embodiment, a test is used that takes the time of the day into account. Time of day correlates to some degree with legitimacy. Claims filed at night are more likely to be fraudulent, while claims filed during the day are more likely to be legitimate. Perhaps legitimate claims are seen more as work that can be squeezed into a daily routine, while fraudulent contemplations have a wholly different psychological rationale.

In the same vein, secondly, in an embodiment, a test is used that takes pay-cycle into account. That is, the time of month or week or time of a bi-weekly schedule. In any case, payment schedules also correlate to some degree with legitimacy. The longer since last pay-check, the larger proportions of fraudulent claims are filed.

Thirdly, certain special events may be taken into account as well. Certain events, such as brand keynotes or product launch events, spark a surge in fraudulent claim requests. Claim requests received before, during or during some period surrounding such event may be considered more likely to be fraudulent.

In an embodiment, the method comprises image recognition tests 133. For example, certain image patterns may strongly indicate image tampering. If a fraudster has modified an image of a wrist by adding a watch to the wrist to supply as evidence of ownership, this may be caught by the image of the wrist that is added to the image of the wrist matching a part of an image that is available online. Such pixel-by-pixel comparison may be almost error-free in detecting image sources, since accidentally replicating the image pattern is effectively impossible. One type of partial image that is useful is for example the reflection on the side of the watch face, although the watch face itself will also strongly indicate fraud if it matches.

Furthermore, the method may comprise dynamic signal thresholds 134. Such dynamic signal thresholds 134 allows time-dependent criteria for what prompts a claim rejection signal 162, a claim verification signal or a claim review signal. For example, for the examples given above, in the period surrounding a product keynote, claims relating to stolen or destroyed electronics may receive more scrutiny before it is considered legitimate. On the other hand, other claims may be paid out more leniently.

In an embodiment, a list of unsolved claim requests is taken into account when determining the dynamic signal thresholds 134. This ensures that the workload of the investigator is taken into account. When there is an overabundance of work, a more lenient approach may be taken where a larger proportion of claim verification signals are transmitted. For example, a claim request price may be assigned below which certain claim requests are approved without substantial review. Such lowermost item value may be raised when the list of unresolved claim requests is long. If there is very little work in a period, the thresholds may lower so that a larger proportion of cases lands in the hands of the case worker or investigator.

In one embodiment, the claim request requires final manual approval before being accepted. In another embodiment, certain claim requests can be accepted automatically without any human intervention at all.

Furthermore, the timings surrounding time of day and pay-cycle may be taken into account when determining the dynamic signal thresholds 134 as well.

Not only the claim type but the document type may inform the types of tests being conducted. Many different types of claims may require similar or identical tests when similar document types are supplied. In an embodiment, the list of tests further comprises a number of document-determined tests.

Document determined tests take certain file/document parameters into account, such as file type, file content or a compressed file hash value, when determining which signal to transmit. Document determined tests comprise tests such as document tampering tests, document reuse tests and optical character recognition tests.

In an embodiment, the list of tests further comprises at least one tampering test. Testing for document tampering relates to reviewing whether the document has been modified or whether it is original. This includes tests like identifying the document creator, such as an image manipulation software, comparing image fragments with image databases or comparing fonts in a document. Some of claim types may require overlapping types of tests.

In an embodiment, the list of tests further comprises at least one document reuse test. Testing for document reuse is useful to identify whether a policy holder is making a false claim based on a previously used file. For example, an ownership certificate of a watch may posted online and several policy holders may fraudulently claim to have lost the indicated watch. Comparing a document or a document key or hash against a database of such previously used documents may allow the method to identify such reuse. Having a central database among insurers facilitates this further. In an embodiment, the list of tests comprises at least one policy holder test. A policy holder test is a test that determines, based on available data the policy holder history. This can relate to contract data in any form that he insurer has access to such as policy age, outstanding payments, payout frequency, payout amount, payout amount over time and payout cycles. Payout cycles refers to the claim request flows that the policy holder has previous filed. The policy holder tests can also relate to social media presence data, web presence data,

Fig. 5 illustrates a claims dashboard (190) according to the invention that is used by an insurance investigator or claims handler. Conventionally, an unordered or chronological list of requests including documents and texts is provided to the insurance investigator to peruse manually during their investigations.

Using the method of the invention, the insurance investigator receives the claim requests pre-assessed. Such a dashboard 190 can be provided on a computing device screen 195 and show individualised lists each having claim requests being similar, such as by assessment. In the shown embodiment, each claim request is sorted into one of: a pre-verified list 191 , a pre-rejected list 192, and a review list 193. The majority of claim requests are typically sorted into the pre-verified list 191 , where the method of the invention has verified that everything seems to be in order. The investigator may either simply see a tally of verified claim requests (not shown), or he may be asked to verify certain simple, predetermined conditions. For example, he may be asked whether an image shows a certain item. This ensures that claims match the products described by the policy holder. For example, a policy holder may have supplied evidence of ownership of a purse and a theft date later than the evidence, as well as a receipt from filing a note with the police about the theft. This all matches the expected evidence. The investigator is then asked whether the image shows the named brand purse. Thereby, legitimate claim requests do not take up too much time for the individual insurance investigator but can instead be quickly processed while ensuring quality processing.

Other claim requests will be sorted into the review list 193. These are the claim requests where the list of tests has identified specific incongruencies, or even where a policy holder has answered during the interrogation step in an incongruous manner or not at all. Images of products may have creation dates later than theft dates for example. The investigator is thus handed a list of specific questions to answer. This may require that the investigator ask the policy holder and/or analyse the answers that the policy holder has given during the interrogation step. Depending on how a consultation turns out or how the policy holder has answered in an interrogation step, the insurance investigator may conduct further investigations or swiftly finalize the case. Thereby even for cases that cannot be resolved without investigator involvement, the method identifies problematic areas of the claim request for the insurance investigator. Thus, the investigator can move deftly through the cases and is directed to where it is needed, reducing the risk of manual errors and significantly reducing the workload on the insurance investigator.

A third group of claim requests are those whose documents have clear evidence of tampering. These are listed in a pre-rejected list 192 of claim requests.

For example, if an image has fragments of known online images, has metadata that mismatches with other metadata of the same file or other such incongruency that only arises with the effort of tampering, the claim request may immediately prompt a claim rejection signal. These may simply be rejected. Alternatively, these claim requests need to be handled more carefully, and they may be handled by a different department with more investigative authority, and/or be filed with the legal authorities or other outside third parties. In any case, such rejection is accompanied with a certain claim request handling escalation. Of course, it is possible that an error is made and then the claim request may be downgraded again, and returned to the authority of the primary insurance investigator. By directly sending this to the fraud division or other such department when the documentation meets certain criteria, more fraud is identified.

As can be surmised, other embodiments of the invention have different versions of the investigation dashboard 190. The dashboard may be split between a primary investigation dashboard showing at least the review list 192 and a fraud dashboard showing the pre-rejected list to fraud investigators. In any case, the pre-verified list may be shown to the primary investigator on the dashboard, or it may be automated fully.

By thus having an investigation dashboard 190 of insurance claim requests, the investigation-relevant data is predetermined for the investigator. This allows him to save clicks in handing the cases, to better identify fraud, and to avoid cognitive fatigue during routine inspections, which might otherwise result in fraudulent evidence slipping through his fingers. Further, since the list of tests can be updated for all investigators at once, the insurance pay-outs can be streamlined, helping the insurers avoid costly mismanagement of claim requests and implementing best practice methodologies instantaneously.

Claims

1. A method comprising the steps:

- receiving a claim request (110) through a digital channel, the claim request (110) comprising a claim type (111 ) and at least one digital document (120),

- retrieve a list of tests (131 ) from a database (130), at least one of which tests or the combination of tests being selected based on the claim type,

- analysing the metadata of said digital document using a processor executing the list of tests, the tests at least indicating whether the document metadata meets metadata parameters of the claim type of the claim request,

- assess (140) whether the claim request (110) is legitimate based on said executing the list of tests,

- transmit a signal indicative of the assessment, where the signal is one of:

- a claim verification signal (191 ) that is transmitted if the tests determine the claim request to be legitimate,

- a claim rejection signal (192) that is transmitted if the tests determine the claim request to be illegitimate and

- a claim review signal (193) that is transmitted if the tests are inconclusive as to claim request legitimacy, where the claim review signal comprise at least one resolving question specific to a data anomaly that resulted in the claim review signal being transmitted, where for certain answers given to the resolving questions by the policy holder, a claim verification signal or a claim rejection signal is transmitted and where

- the at least one resolving question is transmitted through a digital channel to a policy holder having made the claim request.

2. A method according to claim 1 , wherein at least one test from the list of tests is a selected based on document type of the digital document.

3. A method according to any of claims 1-2, wherein the tests further indicate anomalies in document content or whether or not at least two metadata signifying overlapping information matches with each other.

4. A method according to any of claims 1-3, wherein the method further comprise a step of interrogation, wherein the at least one resolving question (171 ) is part of at least one series of related questions (172,173) presented to the policy holder through a digital questionnaire through a digital channel, where answers to prior questions of the related series of questions determine subsequent questions posed in a hierarchical manner.

5. A method according to any of claims 1-4, wherein if a claim verification signal is transmitted, the policy holder receives said claim request verification signal immediately following filing the claim request.

6. A method according to any of claims 1-5, wherein the tests further comprise at least one claim reporting test (132), whereby data specific to the claim request filing is gathered to inform the assessing whether the claim request (110) is legitimate, such as data relating to the flow of the claim reporting or the timing of the claim reporting.

7. A method according to any of claims 1-6, wherein the tests further comprise at least one image recognition test (133) to inform the assessing whether the claim request (110) is legitimate, whereby a digital document is tested for tampering such as by comparing the digital document, such as an image, with images in a database to identify at least partial matches.

8. A method according to any of claims 1-7, wherein assessing (140) whether the claim request (110) is legitimate based on said executing the list of tests comprise a dynamic signal threshold (134), where which type of signal is transmitted (161 , 162, 163) depends on said dynamic signal threshold that changes over time.

9. A computing device (150) having a processor (151 ) adapted to perform the steps of any of claims 1-8.

10. A computer program comprising instructions which cause the computer to carry out the method of any of claims 1-8, when the program is executed by a computer.

11. A computer-readable medium comprising instructions which cause the computer to carry out the method of any of claims 1-8, when executed by a computer.