Are There Challenges To Automating Product Onboarding?
CrowdANALYTIX creates automation solutions, and our most popular offering is DataX, an AI-driven product data onboarding solution for retailers and distributors. There are countless benefits to automating this process, and the ones we tend to focus on are increased speed, increased accuracy, lower costs, and of course scalability. But are there challenges when a business chooses to automate their product onboarding?
Of course there are! And we want to be transparent about them.
Automation is not instant.
Every company wants a solution that will instantly digitally transform their business, deliver ROI immediately, and require no onboarding period. Sounds great, right? Unfortunately, this is more of a pipe dream than a realistic expectation, even with AI on your side. Our DataX product onboarding solution is capable of achieving 90% automation within six weeks for companies that have access to high-quality data, but we still have to begin by doing everything manually. During the 6-week ramp-up period towards DataX automation, AI algorithms have the time to learn proprietary taxonomies, organize data, and begin taking over human tasks like data validation and completion. We take over from day one, but our AI takes over slowly during a weeks-long learning period.
Output depends on input.
The level of automation possible varies for each unique business, and depends heavily on the quality of the data available. Some retailers and distributors have suppliers that are able to provide very complete information, regardless of its format, and DataX organizes this data for use. However, if the suppliers or manufacturers do not have quality data, there is a limit to the level of automation available. DataX can handle data in virtually any format, data that is disorganized, and data in massive amounts. But if a single piece of information — for example, the width of a window frame or the color name of a rug — simply does not exist, DataX cannot generate it from scratch.
Similarly, the level of accuracy possible is dependent on the data quality and quantity available. As you will read below, CrowdANALYTIX can usually offer an accuracy rate of 90%, which is far better than the accuracy of a human being. But if the data our AI receives is poor — incomplete text, blurry PDFs, corrupted images, and more can be issues — or is in inadequate quantities, then the onboarded product data produced may be incomplete, inaccurate, or ambiguous.
Expect more errors initially — It’s part of the process.
Just as full automation cannot be implemented instantly, the full potential of AI accuracy cannot be achieved instantly, either. The humans who begin the AI implementation process have to learn each company’s taxonomy and custom business rules in order to build the necessary algorithms for automation, and as most companies have already discovered, human beings are error-prone. They typically get 70% of a given task correct when they are giving it the average amount of attention and care.
Luckily, humans are able to learn over time… and so can our AI. Even as we shift to automated data onboarding, errors will occur as the AI learns, processes more data, and makes adjustments based on feedback. But unlike humans, AI can be brought to about 90% accuracy under the right conditions, giving us better-than-human accuracy in most instances. We can never promise more than 90% accuracy, because we do not (and cannot) create error-free systems. Better-than-human is as good as it gets, especially since the only two choices we have are human and AI efforts!
Adjustments will be required.
Do not expect to implement a solution like DataX and never look at it again. All intelligent systems, for all purposes, require tuning and adjustment over time. Why? Because input data is guaranteed to change over time, and that can cause the initial AI algorithms to decrease in accuracy over time. Unlike software, AI does not perform at 100% consistency over a long period, because AI deals with live, shifting data. Many companies also need to adjust their taxonomies, definitions, and business rules over time due to mergers and acquisitions, customer preferences, and changes in demand.