Articles

Can you rely on your structured data to “teach” your AI?

Can you rely on your structured data to "teach" your AI?

By Cheryl Ahrens Young, IGP, CSM, CSPO, CTT+, ERMm, ECMp 
[email protected]

The tenets of records management are to provide a reliable, authentic record with auditable integrity.  The assumption that unstructured records (a contract, for example) are less reliable than structured data (databases, for example) needs to be investigated more thoroughly.  In the past few years, I have been involved in a number of projects where the underlying assumption that the data held in line of business applications, such as Oracle and SAP, were accurate and complete.  This assumption proved to be costly, in terms of both time and money, as projects were delayed and additional resources needed to identify and remove real duplicates or rename non-duplicated information and to standardize naming conventions so that the project could move forward, often after the go-live date. 

There are so many adages that apply – “pay me now or pay me later”, “assume = ass u (and) me”, “garbage in/garbage out”, it’s almost funny (almost).  If you have lived through a project where an underlying assumption proved false, you understand the pain.  In many cases, the test environment uses a database of made-up data, so as to not compromise security requirements.  In any project, though, live data is used in the production environment.  My experience has been that the data needs to be thoroughly examined, and tested, before the go-live date in order to celebrate a successful implementation.

How could the involvement of a records management professional at the inception of creating a new database or rollout of a new project avoided this pain?  We’re trained to create classification plans that make business sense, understand the implications of abbreviations and acronyms, and, know that a record will lose its value over time and will need to be discarded.  Dealing with the here and now, in migration or business process automation projects, we also understand the importance of cleaning and purging information during the development stages, prior to testing, so that the testing is a reflection of the usefulness of the application and not the integrity of the data. 

During the Summer Conference this year, the importance of the profession was emphasized by the speakers in topic ranging from Tyrene Bada’s RIM 101 to Andrew SanAgustin’s Where is the RIM/IG Profession headed and a great panel discussion on RIM in the Real World.  A common theme: ROT exists!  Records retention schedules are necessary to defensibly dispose of that ROT.  Getting rid of the ROT improves the overall results in both AI and human production.

Another lesson learned: AI prompts and any resulting transactions are records!  Update your retention schedule to include these!