Address recognition, delivery errors, extra staffing costs, manual keying and more. Let’s look at four ways artificial intelligence (AI) can help to solve the postal business challenges of seasonal peaks
Every year, global seasonal mail peaks, principally during the Christmas period (October-December), continue to stretch postal sortation systems, resources and staff to the absolute limit.
In the UK, during the Christmas peak, Royal Mail processes up to 10 million parcels per day, compared to an average of 5-6 million parcels per day in non-peak months.
A study commissioned by FedEx estimated the total number of parcels handled by all carriers during the 2024 peak season in the UK was 1.29 billion items over the three-month period. That’s a 10.9% increase from the previous year.
The surge in parcels is driven by continued e-commerce growth, fueled further by the now familiar seasonal, time-limited online sales events. As a result, the UK is now the busiest market in Europe for this period, accounting for 21% of total European deliveries: that equates to some 12 parcels per person. Just Europe’s peak season parcel volume alone could now fill Wembley Stadium 34 times over.
The sector’s seasonal business challenges are real. So, it’s no surprise that operators are now looking to AI to improve sortation performance, increase accuracy, reduce errors and minimize operating costs.
Business challenge 1: Improving automated mail sortation
Traditional optical character recognition (OCR) systems have long been used in postal sortation to interpret (read) handwritten and printed addresses and codes on letters and parcels. However, conventional OCR systems often still struggle with recognition and/or accuracy in several scenarios. Handwritten text, especially when contrast is poor – for instance black ink on a red envelope – may be unrecognizable when scanned (binarized), requiring the item image to be inspected by a human operative to manually key in the correct details.
Printed address labels also pose problems. They may be crumpled, be very small, have poor print quality, include unusual foreign or stylized fonts, or appear shiny or smudged. Similarly, items like magazines that are wrapped in clear plastic are also notoriously difficult for OCR systems to interpret: the system cannot correctly identify the relevant address label area against the visual ‘noise’ of the package.
Historically, human inspection has provided the essential contextual information and judgement required to find and check the address data to resolve issues. Is the package the correct way up? Has the OCR scanned the region of interest? Is something obscuring part of the address?
Now, using machine learning, AI software solutions can be trained to do the same, boosting the performance and capacity of existing automated mail sortation assets.
Making recognition smarter
Systems like Lockheed Martin’s image-based AI platform, Minerva, reduce the noise on complex parcel labels – shiny plastic packaging, cluttered magazine wraps and seasonal decorative envelopes – so existing OCR engines can read them correctly first time.
This approach uses AI-enabled software to pre-process and optimize the image of each parcel or letter before it is passed into the existing OCR engine. Algorithms are trained to locate, identify and isolate the delivery address block, pinpointing the address information and other data such as a barcode.
Using the clean, uncluttered address block image area, the system can create a clear, high-contrast image of the mail piece, fully optimized for the operation’s OCR recognition solution.
In the few exception cases where an image still cannot be recognized automatically by standard OCR, the system can also emulate human keying operators, automatically inspecting, address matching and keying the correct details at speed.
Just as a human can manually inspect a package or letter and determine the address and assess its size, weight and condition (for example if its damaged), AI, trained on suitable data, now offers similar capabilities.
AI also presents the opportunity to automatically manage routing for certain items through the sortation system to minimize rough handling or extract damaged packages before they are dispatched.
Business challenge 2: Automating cross-border customs data capture
The volume of cross-border parcels is also on the rise. Postal operators must capture customs contents declaration information from labels and recover VAT if required. Using the same region of interest label recognition, the AI can extract the data and pass it to an OCR solution to be read, captured or sent to a human operative to be keyed.
Business challenge 3: Reducing costly delivery errors
Research by Loqate found that when addresses are inaccurate or incomplete, 41% of deliveries are delayed and 39% simply fail. Delivery errors are both costly and damaging. In the USA, 8% of first-time deliveries fail, at an average cost of US$17.20. In the UK, the failure rate is 6% at an average cost of £11.60 (US$15.25) each.
The costliest failure isn’t when automation can’t read the address – it’s when it misreads it. Parcels sent to the wrong end of the country can ‘loop’ in the network: sorted to the wrong depot, returned to the hub, sorted incorrectly again and delayed repeatedly. Such errors are costly to resolve, because operators only discover the mistake once the item physically arrives in the wrong place.
Clearly enhancing OCR address reading accuracy even a small percentage by using AI has a major impact: less wasted fuel and CO2, happier customers and retailers, fewer refunds, and increased loyalty and retention.
Business challenge 4: Reducing reliance on seasonal staff
To manage the peak festive surge, postal operators need to hire and train thousands of seasonal workers every year. In 2024, the UK’s Royal Mail payrolled 16,000 additional temporary staff between late October and January 2025 to assist 85,000 postal workers sort the seasonal mail and parcel uplift. In the USA, USPS hires up to 50,000 seasonal workers during the holiday period (from Thanksgiving to early January).
Recruitment can be challenging, and labor costs are rising. Applying the average postal role wages and seasonal contract duration terms for each country, these hires demand an annual investment in seasonal labor of circa £43.2m (US$56.7m) by Royal Mail and US$288m by USPS.
Clearly, as AI-enabled workflow enhancement becomes more widely adopted, it has the potential to both reduce the number of seasonal roles required and increase the productivity and efficiency of those in post.
Software as a Service AI
The modular, Software as a Service model approach to AI adopted by Lockheed Martin’s Minerva suite provides the opportunity to instantly enhance existing sortation system accuracy, throughput and performance, and, depending on the application, without the need for extra hardware or complex integration.
The pre-processed, AI-enhanced image files created are platform-agnostic, compatible with any OCR system. The solution can readily be pre-trialled using problematic data, providing a clear measure of the improvement that is possible in just a few days. Coupled with a simple subscription use model and no lengthy contract commitment, software-only AI solutions promise to transform efficiency and profitability for postal operators.
With varying postal volumes and rising seasonal labor costs, using AI to optimize existing sortation systems makes sound commercial sense. And it could truly transform peak postal seasons, for all involved.
About the author
David Woodward is a seasoned IT and program management leader with over 30 years of experience spanning defence, logistics and infrastructure sectors. Currently serving as systems solutions programmes manager at Lockheed Martin UK, David heads the Systems Solutions portfolio, delivering advanced image analysis and data services technologies to national logistics operators and beyond.