Postal voice assistant: Optimized for sorting
The postal industry is experiencing big changes these years with more players at the table, and a decrease in the number of letters being sent. These changes have already forced the Danish division of PostNord to centralize their sorting processes, making postal sorting more complicated for the personnel who now have to overlook much larger geographical areas and sort items into routes that they have no knowledge about.
PostNord er den førende leverandør af kommunikations- og logistikløsninger til, fra og inden for Norden. De sikrer postservicen til private og virksomheder i Sverige og Danmark og er aktiv i alle de nordiske lande som logistik- og godsdistributør.
Centralisation slowed down the sorting speed
At PostNord the vast majority of postal items is sorted by large OCR machines. The remaining, but still significant number of items has to be sorted manually, due to parameters such as deciphering tricky handwriting and handling odd sized items which have proven to be too challenging for the existing OCR sorting machines. Although a bit time-consuming, this wasn’t really a problem until reality knocked on the door forcing the organization to centralize the existing 200 smaller distribution centers to 27 hubs. However, across the Danish zip codes many streets share the same names, requiring every hub employee to sort letters while overlooking several hundred routes in 50 to 60 different zip code areas. This challenge made PostNord go in a different direction using new technology.
Speech technology speeds up the process
Together with a team of developers from Dictus they found a way through the wilderness using state of the art speech technology. In this way, PostNord combines the best of two worlds in a powerful cocktail: The knowledge and experience of the postal workers and the speed and accuracy of modern technology.
“Tekst til tale teknologi gjorde at vi kunne højne produktiviteten, der var meget højere end hvad vi ellers kunne have set os ud i en implementering af et helt nyt setup.”
“Using the data from the big OCR machines in conjunction with a worker that has 10-30 years of experience reading letters and learning areas using the voice system is an unbeatable combination when you have rest mail that is somehow unsortable.“
5 facts on voice sorting
- The address database is imported from the same data used by the large OCR sorting machines
- Mail and parcel data is exported to a central message system for distribution to other systems, eg. route and scheduling
- The voice sorting system used at PostNord works across both dialects and accents in Danish
- The system can be delivered in 20+ languages
- Voice sorting is also an efficient training tool for new sorting personnel
Unique street names recognized without user voice training
Items voice sorted weekly
Why speech technology is a success at PostNord
Implemented overnight for 350 workstations / 1000+ users
Each week more than 1 million items are voice sorted
50000 unique street names are being recognized without user voice training
Through tentative sorting algorithms it is possible to place even partially flawed items on the correct route
Essential tool to make dynamic and automated route and scheduling possible, due to live digital information on every manual item
Before implementing voice sorting items unsortable for the OCR machines had to be sorted manually several times – this process is now reduced to a single sorting procedure giving information such as route, sequence number and QR code all at once