By all appearances, Anna Delvey was a wealthy German heiress. Her Instagram account made her out to live a glamorous life, filling her days with VIP parties, stays in luxury hotels and art launches. In her mid-20s, she moved to New York and told her new friends that she had grand plans for her $60 million trust fund.
Who Was She Really?
She was an impostor, a social climber of the worst kind. Her real name was Anna Sorokin. There was no family fortune. A daughter of poor Russian immigrants who moved to Germany when she was a teen, Anna badly wanted to fit in with the societal elites. So she invented Anna Delvey the heiress. Once embroiled in the game, she doubled down on the ruse and took it to extremes.
Anna was eventually exposed after she was caught committing bank fraud while trying to manage her mounting personal debts. She was convicted and sentenced to prison. The people she fooled never did get their money back.
How Did She Fool So Many Smart People?
Anna did just enough to appear as if she lived a jet-set lifestyle, wearing expensive clothes, saying the right things on social media, and even staying at swanky hotels for a short time. Since she could never afford to settle the tabs herself, she lied to borrow from friends and overdrafts from banks. Incredibly, when they were cajoled by Anna to pay for things, no one saw the red flags. This enabled her to cover her tracks and keep the ruse going for years.
A Cautionary Tale?
The content capture industry, long a respectable market with slow and incremental change, finds itself now in the midst of a complete disruption brought on by AI. We’re seeing applications using the latest AI advances in computer vision, machine learning and deep learning to break through the final mile problem of capture: the problem of human intervention. Human touch prevents true unattended document processes, aka Straight-Through Processing, the Holy Grail of RPA.
Deep Analysis coined the phrase “Cognitive Capture” to describe and analyze the emergence of these new capture solutions. New products are pouring into the market applying new AI methods, and the results have been impressive. Training a new document classifier is reduced from days or weeks to minutes, and from 100 samples down to 1 or 2. Data extraction accuracy from unstructured documents leaps from 50-60% to 90% and more. Outliers, errors and exceptions can all be managed by the machine.
Who wouldn’t want one of these? The new products are threatening to displace legacy capture products, some in place for over 15 years. Realizing the imminent threat, legacy capture vendors have jumped on the AI bandwagon, at least from a sales and marketing standpoint. But not every capture product that says “AI/ML” is authentic Cognitive Capture. Like poor Anna, there are some clever impostors who may talk the talk and even dress fine. But after a while they’re exposed. Customers who are not careful will risk wasting both money and time. Who can afford to waste either?
How to Spot the Real Thing
Here are some guideposts I found useful to evaluate the AI authenticity of a capture product.
- Does it use trainable machine learning (ML)? Many capture solutions say they use ML but still depend heavily on Subject Matter Experts (SMEs) to assemble document samples, create metadata or classify document types. With potentially hundreds of diverse document types, this is resource-intensive, time-consuming, and costly. Look for Cognitive Capture tools with highly advanced ML that can automatically learn a new document class with only a few samples, apply learning to new documents on the fly, and augment or even replace SMEs.
- Can it apply multiple AI classification techniques? Look for Cognitive Capture tools that can analyze documents based not only on text, but also on visual anchors, imagery and handwritten information on the document including the presence or absence of signatures. Documents with sensitive information such as handwritten social security numbers can be included in your workflow and easily identified. Regardless of how the information is presented, classification should be based upon all the available information in the document, not just a select subset of that data.
- Does it offer high extraction accuracy with low error rates and at top speed? Most data extraction systems become slower and less accurate in cases such as healthcare claims processing where forms are involved. A Cognitive Capture solution properly deployed will always improve your results and performance. Advertised accuracy rates are a bit like the miles per gallon sticker on a new car. Actual mileage will vary with options, driving conditions, driver’s habits and vehicle condition. This must be demonstrable in your proof of concept.
- Does it provide an accessible and clear-cut user interface for business users? Let’s face it: this is a very sophisticated and complex software operation. Look for a solution with a user interface that doesn’t require a data scientist or programming skills. A Cognitive Capture solution will hide the complexity by using AI to complete basic tasks and to suggest answers. Fine-tuning document classes should be as simple as correcting results by dragging and dropping results from one class to another, then re-running the task.
Authentic Cognitive Capture Tools
In our world teeming with deep fakes and impostors like Anna, it pays to do your homework before you invest in any product. I recommend using due diligence to separate the authentic Cognitive Capture tools from the wannabe impostors. Neutral analysts such as Deep Analysis and AIIM International can help guide you through the maze. In the end, what really matters will be a proof of value / proof of concept that delivers the outcome that best fits your business requirements.