Deploying AI to supercharge efficiency at Historic Images
Challenge:
Manually creating and recording metadata for thousands of archived historical images was expensive, slow, and prone to errors.
Results:
- Custom AI integration reduced image processing costs by 90% while dramatically reducing turnaround time and potential for data-entry errors.
- Vastly increased capacity: AI solution enables them to process up to 4 million images per year vs. 300,000 per year using manual methods.
- Improved data quality allows more accurate identification and valuation of historic photos for the collector market.
“We were spending thousands of dollars per week … Today, we’re getting a significantly better finished product, for a fraction of the storage costs and a fraction of the production cost.”
— Christopher Galbreath, COO, Historic Images
Historic Images was born out of serendipity. “It started out as a baseball card deal,” said Chris Galbreath, CEO and founder of the company. “We bought a card collection from a photographer, but it turned out his photographs were more interesting than his cards.”
Reselling vintage photos to collectors led to another business opportunity: digitizing and cataloging photo libraries from newspapers, preserving history and rescuing photo prints from landfills.
After a career in the logistics industry, Chris started Historic Images in 2009 just as many US newspapers were downsizing their operations. Major newspaper photo archives, or “morgues,” are vast rooms of filing cabinets filled with decades worth of printed photos. Scanning those photos preserves the images and opens up a new revenue stream for publishers via digital licensing.
Historic Images began working with newspapers like the Detroit Free Press and the Chicago Sun-Times to digitize and catalog their print photo libraries. Major publishers like Hearst Communications followed, along with other organizations with large photo archives.
In the years that followed, Historic Images built a great reputation on their expertise in handling and scanning thousands of literally one-of-a-kind photographs. Christopher, his son Chris, and his daughter Jessica work in the business today, along with a team of employees based in Memphis, Tennessee.
“We were honestly more of a logistics company than a photo company,” Christopher said. “Because of my experience, it was easy to take a crew in, load everything up, get it palletized and out of the building quickly and efficiently. It saved them tons of labor, and it actually saved their archives.”
Based on the demand they were seeing in the market, they hoped to grow the company to process 10 million photos per year. But that growth potential was held back by a bottleneck they didn’t think they could overcome.
Challenge: Expensive, slow manual data entry process
When they first reached out to us at Skeleton Key, Historic Images was spending a massive amount of time and budget to add metadata to its newly-scanned photos.
This vital cataloging information includes a description of who is in the photo and what they’re doing, as well as a content category, keywords, and the names of people shown. In addition, most newspaper photos have typed or handwritten details on their backs that must be transcribed.
As the company grew, metadata processing became a major cost as well as a bottleneck. “Our all-in cost was about $1 per image,” said Christopher Galbreath, son of Chris and chief operating officer at Historic Images. “We were spending thousands of dollars per week.”
While they worked with reliable data-processing vendors, those teams had limited capacity, Christopher said. And the sheer volume of manual work led to transcription errors, misidentification of people in photos, and other issues.
“The accuracy of the information, the data, the bandwidth, the cost – you name it, we were up against it,” Chris said. “We had more employees, more demand, more fixed costs. And we still had to have the photos back to the newspapers within a specific timeframe.”
Other technologies for processing photos and metadata existed, but they were difficult to scale, Christopher said. “There are systems that could handle ten thousand or a hundred thousand images, but when you try to scale to a million, or 10 million which was our goal, it’s a challenge.”
Chris and Christopher were talking about these issues on a business trip when Christopher had an idea. They weren’t very familiar with custom software or AI, but Christopher believed technology could offer a solution to these and other challenges. But neither of them knew how to get started.
Solution: Custom AI integration transcribes data
Our custom software solution performs a variety of tasks as scanned images are uploaded from workstations, including resizing photos, assigning them to specific campaigns, and populating asset records including image metadata. For this, we leveraged a large language model (LLM) hosted in the cloud to receive queries via an AWS Lambda function. Images are run through a pretrained Amazon Nova model which doesn’t involve a public instance, ensuring data privacy and security. It also allows for granular reporting on job status, time required, and other key metrics.
In addition to processing images in a fraction of the time, the trained AI model can recognize a phenomenal level of detail by analyzing a photo and any text it recognizes, including image captions on the back of news photos. And with a variety of training data to draw on, the LLM can add context and details that people might have missed when adding metadata to photos manually — such as the make and model of a car, or the breed of a dog. This richer cataloging data helps both Historic Images and their clients get more value out of their photo archives.
Results: Costs cut by 90%, positioned for future growth
Better quality for a fraction of the price
Historic Images’ up-front investment in custom software and AI solutions has more than paid off, Christopher said. “We get a significantly better finished product, all stored in one place, for a fraction of the cost.”
It’s also allowed them to improve the data quality in their own archive of prints for sale. “For older photos, you don’t get all the same tags and information. You get more misspellings, you get more misstatements,” Chris said. “For every photo we still have a copy of, we’re rerunning it through the new software so that all of the data will be uniform.”
Future-proof data storage
Our custom solution also solved a secondary issue for Historic Images. “We had a fragmented data storage system,” Christopher said. “A lot of our newer data was being stored in the cloud, but a lot of our older data was strictly on drives in a server at the office. We always ran the risk of that server failing.”
Without a coordinated storage system, transferring large amounts of digital images back to clients was difficult. In the past, they had solved this issue by shipping large numbers of physical hard drives. Our solution gives them centralized, secure storage and reduces the risk of data loss.
These improvements not only allowed them to take on more work with confidence, but also positioned them to achieve their vision of success — including buying out their chief European competitor.
“When we started, the idea of custom software was pretty new to us,” Christopher said. “Skeleton Key made it easy, and they gave us the confidence to know our team can do this.”
“Skeleton Key’s people are some of my favorite people on Earth,” Chris said. “This has changed everything for us.”
We help you find the right approach to leveraging AI and custom software
When AI is applied properly to the right kinds of business challenges, the results can be transformational. But the first step is asking the right questions and knowing where, when, and how LLMs and other tools could help your organization.
Our team has the technical knowledge and the business acumen to understand which problems AI and custom software can solve. We build solutions that evolve both at the pace of technology and in line with your needs.
Book a consultation with us today and learn how we can help your team achieve great things.
