How to use case notes for early identification of safeguarding risk

Most practitioners treat case notes as a record of what happened.

Important, yes. Necessary, absolutely.

But still, in most services, they are seen mainly as burdening documentation. Something we complete to evidence contact, capture key facts, and move the work forward. And of course they are that. But I think they may also be something more. I think they may contain patterns, signals, and warning signs that services are not yet making full use of, especially when it comes to identifying safeguarding risk earlier.

This is why, recently, I’ve been working on designing a system of analysing case notes to enable early detection of safeguarding concerns. This is not about replacing professional judgement or reducing human distress to data points. It is about taking seriously the possibility that the stories already sitting inside our records may help us notice sooner, ask better questions, and respond before concerns become crises.

The starting point for building such a system is to develop and validate a safeguarding lexicon. A structured framework of words, phrases, patterns and contextual cues that can help us spot risk earlier through the notes staff and volunteers write after 1:1s and group sessions.

But what I am learning through this process is that this isn’t really about words.

It’s about what happens when services learn to listen more carefully.

Everyone who worked in addiction, mental health or social care services for long enough, knows that people rarely disclose risk in neat, clinical language. They do not arrive saying, “I am presenting with escalating suicidality ideation, possibly subjected to coercive control, and impacted by acute self-neglect.”

They say things like:

“I’ve had enough.”“I can’t go home tonight.”“I’m beyond help.”“I’ve messed everything up this time.”“I don’t see a way out.”

And sometimes they don’t even say that much.

Sometimes the signal is quieter.

A phrase repeated across three sessions.A note about fear tucked inside a paragraph about debt.A passing comment in a group that would mean very little to someone from the outside, but means everything to someone who knows how collapse often sounds, before it becomes visible.

That is the space I’ve been thinking about. Not just how we respond to incidents once they are clear, but how we build systems that help us notice them earlier.

At Betknowmore, like many organisations working close to pain, complexity and shame, so much important information sits in narrative notes. In the free text. In the part of the case note where a practitioner tries to capture not only what was said, but also what was felt or what was implied.

And yet, in most services, those notes are only lightly used as a source of organisational intelligence.They sit there holding patterns of risk. Holding the early architecture of a crisis. Holding, sometimes, the first fragile signs of hope.

So, I’ve started developing a structured safeguarding text-screening framework that can help us identify potential safeguarding risks more consistently and earlier through the language recorded in session notes. The aim is to harnesses semantic frames and use natural language processing techniques to define fine-grained patterns and search them in unstructured data, namely, open-text fields in our records.

The idea is simple.

If certain phrases, combinations of phrases, or patterns of language appear in a note, the system can flag that a team leader and the safeguarding lead for review. This can support the people doing the work by creating an additional layer of attention around risk.

The purpose of a system like this is not to become smarter than the practitioner.

It is to help the organisation become less dependent on chance. Less dependent on whether one worker notices something another might miss. Less dependent on risk sitting silently in a note until it hardens into an incident.

The research in adjacent fields is actually quite encouraging here. Healthcare systems have already shown that free-text notes can reveal adverse events, suicidality, abuse, violence and other safety concerns that structured fields and formal reporting systems often miss. That does not mean there is an off-the-shelf model ready for gambling support services. There isn’t. But it does mean there is enough evidence to build something thoughtful, proportionate and evidence-informed.

In addiction, mental health and social care work, people often disclose risk indirectly. Through debt language. Through shame. Through relationship strain. Through phrases like “I’m trapped,” “I can’t face them,” “people are on my back,” or “I’ve ruined everything.”

If we can learn from that language more systematically, we can do a few important things better.

We can intervene earlier.

We can strengthen safeguarding oversight.

We can spot patterns across cases, not just within them.

We can support staff and peer supporters to feel less alone in holding uncertainty.

And perhaps most importantly, we can start turning narrative data into learning, rather than leaving it as an underused by-product of care.

That matters not only for frontline practice, but for leadership, commissioning and system design. Because one of the longstanding weaknesses in our sector is that we often know far more than we can currently evidence in a usable way.

But unless that wisdom is translated into systems, frameworks and usable organisational processes, it stays trapped in individual experience. And individual experience, however valuable, is not enough to build safe and scalable services.

What excites me about this work is that it sits at the meeting point of compassion and structure.

It says:human judgement matters deeply.And systems matter too.

It says:notes are not just paperwork.They are part of the safety architecture.

It says:if we listen carefully enough, the service can learn.

About the language of relapse.The language of abuse.The language of hopelessness.The language people use when they are close to disappearing, and the language they begin to use when they are slowly, tentatively returning.

That is where I hope this work can eventually lead.

To services that do not only record stories, but learn from them.

To safeguarding that is not merely reactive, but more alert, more connected, more intelligent.

And to a system of support that gets a little better at hearing what people are trying to tell us, even when they cannot yet say it plainly.

Because in this field, safety rarely begins with a form.

It begins with listening.

And the future of better services may depend on whether we can build organisations that know how to do that well.