The IDEAL Insur­ance Group is rev­o­lu­tion­iz­ing its input man­age­ment through cut­ting-edge AI tech­nolo­gies from Insid­ers. With the intro­duc­tion of the Cog­ni­tive Clas­si­fi­er and a deep learn­ing-based OCR, the insur­ance com­pa­ny achieves a new lev­el of qual­i­ty. The con­clu­sive pro­cess­ing of straight­for­ward issues is now pos­si­ble direct­ly in the inbox – a quan­tum leap in effi­cien­cy.

Thank you for read­ing this post, don’t for­get to sub­scribe!

Based in Berlin, the IDEAL Insur­ance Group offers its cus­tomers var­i­ous prod­ucts for secur­ing bio­met­ric risks and has estab­lished itself as an inno­va­tor in the insur­ance indus­try with new con­cepts such as pri­vate care annu­ity or the first com­plete­ly dig­i­tal insur­ance account “IDEAL Uni­ver­sal­Life.” The com­pa­ny always pri­or­i­tizes the inter­ests of its cus­tomers.

Since 2004, IDEAL has suc­cess­ful­ly uti­lized smart INVOICE for pro­cess­ing incom­ing invoic­es and lat­er intro­duced smart FIX for auto­mat­ed incom­ing mail pro­cess­ing. IDEAL process­es more than 500,000 doc­u­ments annu­al­ly, total­ing around one mil­lion pages. The doc­u­ment intake is high­ly diverse, includ­ing hand­writ­ten let­ters, pho­tographed doc­u­ments, fax­es, emails with var­i­ous attach­ments in dif­fer­ent file for­mats, or embed­ded in the mes­sage, med­ical records, and offi­cial doc­u­ments.

To min­i­mize man­u­al efforts, extract all rel­e­vant data accu­rate­ly, and pro­mote auto­mat­ed pro­cess­ing with a focus on a high dark pro­cess­ing rate, IDEAL was in search of an advanced, intel­li­gent solu­tion. This solu­tion need­ed to han­dle poor­ly read­able hand­writ­ten receipts, low image qual­i­ty in pho­tographed doc­u­ments, or the com­bi­na­tion of text and images on doc­u­ments with­out any issues. A par­tic­u­lar chal­lenge was the desire to con­clu­sive­ly han­dle sim­ple tasks – such as address changes or changes in bank details – already at the first con­tact with the doc­u­ment.

A clear case for the new Cog­ni­tive Clas­si­fi­er from Insid­ers! The Cog­ni­tive Clas­si­fi­er com­bines image-based clas­si­fi­ca­tion with opti­mized text-based clas­si­fi­ca­tion through advanced tech­nolo­gies such as Nat­ur­al Lan­guage Pro­cess­ing (NLP) and Deep Learn­ing. With the recog­ni­tion of visu­al fea­tures, the Cog­ni­tive Clas­si­fi­er can now auto­mate the recog­ni­tion and pro­cess­ing of items such as cer­tifi­cate rolls with seals, notar­i­al doc­u­ments, or pho­to attach­ments. Addi­tion­al­ly, the Uni­ver­sal Inte­gra­tion Lay­er for OCR (ILFO) allows the inte­gra­tion of var­i­ous OCR com­po­nents to address spe­cif­ic use cas­es. IDEAL also intro­duced a deep learn­ing-based OCR along with the ILFO.

“Thanks to the Cog­ni­tive Clas­si­fi­er and ILFO, we have been able to crack the tough nuts in input man­age­ment and make a sig­nif­i­cant leap in process automa­tion. It’s great to be able to imple­ment new AI tech­nolo­gies in close col­lab­o­ra­tion with the man­u­fac­tur­er,” says Carsten Dick­au, Busi­ness Ana­lyst at the IDEAL Insur­ance Group.

Ini­tial­ly, the com­bi­na­tion of the Cog­ni­tive Clas­si­fi­er and deep learn­ing-based OCR was test­ed on the par­tic­u­lar­ly chal­leng­ing doc­u­ment class ‘Death Cer­tifi­cate.’ This doc­u­ment class fre­quent­ly appears in life insur­ance and serves as an exam­ple for all cer­tifi­cates whose extrac­tion is par­tic­u­lar­ly chal­leng­ing with a clas­sic OCR due to print­ed water­marks. In the test, a sig­nif­i­cant qual­i­ty leap was achieved: 97% of death cer­tifi­cates could be cor­rect­ly auto­mat­ed, and in only 3 per­cent of cas­es, a ver­i­fi­er had to inter­vene.

This result thrilled IDEAL so much that the Cog­ni­tive Clas­si­fi­er and opti­mized OCR were intro­duced for all doc­u­ment class­es of the insur­er. By com­bin­ing image and text infor­ma­tion, the Cog­ni­tive Clas­si­fi­er excels even with pre­vi­ous­ly hard-to-clas­si­fy doc­u­ment class­es, increas­ing recog­ni­tion rates by anoth­er approx­i­mate­ly 15 per­cent to near­ly 100 per­cent. The alter­na­tive OCR con­nect­ed via ILFO also con­tributes to improved recog­ni­tion in the case of qual­i­ta­tive­ly poor pho­tos, fax­es, or scans, as well as hand­writ­ten doc­u­ments, sig­nif­i­cant­ly enhanc­ing the high­er lev­el of automa­tion.

With these inno­v­a­tive AI com­po­nents, IDEAL can now process infor­ma­tion that was pre­vi­ous­ly dif­fi­cult or impos­si­ble to extract. The con­clu­sive pro­cess­ing of sim­ple tasks in input man­age­ment has also been imple­ment­ed.

The IDEAL exam­ple demon­strates that pro­vid­ing a plat­form where var­i­ous AI tech­nolo­gies can col­lab­o­rate con­sis­tent­ly brings process improve­ments for cus­tomers through tech­no­log­i­cal advance­ments.