Walk into the offices of our social workers, and you might think you have entered an archive, or perhaps a library. Drawers and shelves are full of paperwork, and just like an archive, these rooms hold diligent, detailed records of the lives of the children we serve. The documents record the highest highs and lowest lows of every child who walks through our doors. Now a new technology promises to lift that weight, and we are trying to find out what it would cost.
Why the paperwork exists
My name is Gayane Aghabalyan, and I have worked as a development officer at the Child Development Foundation (CDF) in Armenia for nearly five years. CDF operates six child development centers in Armenia’s capital, Yerevan, and in the regions of Vayots Dzor, Ararat, and Kotayk. Over the years, I have gone from executive assistant to program lead, working at every level of the organization, from helping write grant proposals to designing the programs that keep these centers running.
In each center, we take pride in our thorough documentation: Initial assessments, individual development plans, monitoring records, family consultation notes. Each child has a robust history that helps us see points of progress, the paths we have taken, and where we should head. This is not bureaucracy for its own sake. When a psychologist, a speech therapist, and a social worker support the same child, these records are what hold the work together.
A big part of my job has been preparing results-based reports, and those results rest on paperwork. Collecting the data we need is a constant struggle, and so is turning it back into a usable form. Whenever I asked colleagues for the same information, it came back in a different shape each time. The idea for this project grew from that small, ordinary frustration: There had to be a better way.
For years, paper was the best means we had. Over time, some of the data was digitized. Now, in the age of new and faster technologies, namely artificial intelligence, the question has changed: How much of this information can be digitized? And how much should be?
The temptation
Using this technology is tempting. An AI tool can analyze years of records in minutes, not months, and sketch a roadmap for a child’s development. It can turn a tired social worker’s evening of note-writing into half an hour. These tools are free, accessible, and require no IT department, which means any practitioner could quietly start using them tomorrow.
But at what cost? These files hold diagnoses, family circumstances, confidential information, and the personal histories of children in vulnerable situations. The privacy concerns around AI are well known. So instead of jumping on the AI bandwagon, we chose to take a step back to understand exactly how our workflows function, how these tools handle data, and what our privacy responsibilities are.
What we found
The project began this spring with a needs assessment of how our centers actually document their work, conducted together with an external expert on digital tools. One finding stood out. Our paper records are uniform: The same forms, the same structure, in every center. Our digital files are not. Digital skills were never a required qualification for social workers in Armenia, nor a subject in their training, so each specialist developed their own way of working. One keeps files in folders by year, another by child, a third keeps notes in formats no colleague could easily understand. The dedication is equal everywhere; the systems are not.
This told us something important. The problem is not that our staff need an AI tool. The problem is that they were never given a digital foundation to build on, and dropping the most powerful technology of the decade onto uneven ground would only widen the gaps.
The difference is visible in small, everyday things. In one center, a specialist keeps every child’s records in clearly named digital folders and can find a document from two years ago in seconds. In another, a colleague equally devoted to the children stores notes in a way only she can navigate, or keeps some records only on paper because that is what she trusts. Neither is doing anything wrong; they were simply never trained to a shared standard.
Scaffolding before tools
So the project starts lower and slower than the name “AI integration” suggests. For example, our first training did not start with AI at all. It started with a simple question we asked the staff: What takes you the longest in your documentation, and what do you wish were easier? From there, we will look together at what these tools actually are and how they work. However, we will agree on one firm rule before anyone begins using an AI tool: No real information about a child ever goes into an external tool. Only once that foundation is in place do we move, carefully, to testing.
Staff trainings will continue in the fall, followed by supervised piloting of AI support in narrow, clearly defined tasks, always with anonymized information. The lasting product will be an internal AI and ethical use protocol: A document that tells every current and future employee what is acceptable, what is not, and who decides.
