The well­be­ing ser­vices coun­ty of Cen­tral Fin­land and Tieto­evry test the use of AI for analysing cus­tomer feed­back

The well­be­ing ser­vices coun­ty of Cen­tral Fin­land has suc­cess­ful­ly car­ried out a pilot project in coop­er­a­tion with Tieto­evry to find out if gen­er­a­tive AI could be utilised to clas­si­fy open cus­tomer feed­back data and respond to cus­tomer feed­back more quick­ly.

The well­be­ing ser­vices coun­ty of Cen­tral Fin­land is respon­si­ble for the social, health and res­cue ser­vices of around 273 000 peo­ple in cen­tral Fin­land. Feed­back on ser­vices is col­lect­ed through a vari­ety of chan­nels: a form on the web­site allows feed­back on all ser­vices in the well­be­ing ser­vices coun­ty, while SMS feed­back and tablet devices are used for lim­it­ed ser­vices. Open feed­back is also avail­able through each chan­nel.

”The analy­sis and clas­si­fi­ca­tion of open feed­back came up in our dis­cus­sions with Tieto­evry, and we decid­ed to inves­ti­gate whether gen­er­a­tive AI could be used to clas­si­fy open cus­tomer feed­back data,” says Jaana Pel­tokos­ki, Ser­vice Man­ag­er at the Infor­ma­tion Man­age­ment ser­vice area of the well­be­ing ser­vices coun­try of Cen­tral Fin­land.

”Cen­tral Fin­land is a cus­tomer who bold­ly want­ed to pilot new tech­nol­o­gy to clas­si­fy cus­tomer feed­back data and take the ini­tial step in using AI to improve res­i­dent ser­vices,” prais­es Niina Siipo­la, Head of AI and Data Solu­tions at Tieto­evry Care.

”For years, Tieto­evry Care has been build­ing the foun­da­tion for using data and AI in social and health­care ser­vices. Our team has a long expe­ri­ence in the use of AI and the devel­op­ment of lan­guage mod­els, and there­fore we know where AI is suit­able and where it is not. Based on this pilot, it seems per­fect­ly suit­ed for automat­ing man­u­al analy­sis work. While AI can­not solve all chal­lenges, it can act as an effec­tive assis­tant in this case, sav­ing resources and help­ing to pri­ori­tise feed­back,” adds Siipo­la.

The pilot utilised the Microsoft Azure Ope­nAI ser­vice, where the clas­si­fi­ca­tion was entire­ly done by gen­er­a­tive AI using large lan­guage mod­els. In the pilot, more than 4 900 open, anonymised cus­tomer feed­back giv­en in 2023 were clas­si­fied into dif­fer­ent cat­e­gories, such as appoint­ment, queu­ing, tele­phone or cus­tomer ser­vice, or encounter. Most of the feed­back was relat­ed to recep­tion, treat­ment, encoun­ter­ing and treat­ing the patient, or cus­tomer ser­vice. Feed­back was also divid­ed accord­ing to tone into pos­i­tive, neu­tral and neg­a­tive.

The results were pos­i­tive­ly sur­pris­ing.  The large lan­guage mod­els per­formed very well in clas­si­fy­ing open feed­back data, and they even recog­nised the tone of voice of the feed­back. From the cus­tomer feed­back data, it was also pos­si­ble to deter­mine, with rel­a­tive reli­a­bil­i­ty, what type of mea­sures should be tak­en.

”It’s great that we had the oppor­tu­ni­ty to par­tic­i­pate in this pilot with Tieto­evry. It gave us insights and per­spec­tive on how AI can be used in open feed­back and text analy­sis. The auto­mat­ed analy­sis and pro­cess­ing of open feed­back clear­ly showed that infor­ma­tion can be gen­er­at­ed faster than cur­rent­ly for the use of ser­vices and as a basis for ser­vice devel­op­ment,” says Pel­tokos­ki.

Read the full arti­cle here.

Source (text and image): Tieto­evry