A PhD-level data scientist, built into the organisation.
Executive summary
Not a cheaper version of an existing task — a capability you could not previously staff.
OASIS represents a province-wide network of agencies supporting people with developmental disabilities. Its leaders coordinate through shared lists where compliance questions, policy decisions, and operational pressure all surface as ordinary messages. The signal in that traffic is rich. Reading it well, continuously, is the hard part — and the skill to do so is scarce and costly.
So OASIS put its organisation on a knowledge graph and gave it a data scientist agent: a teammate that reads the entire network of people, topics, and threads, and returns plain-language decisions instead of more reports. In one recent fortnight it proved the point. A wave of provincial compliance and coverage questions arrived across several lists at once. To anyone reading their own inbox it felt like a busy two weeks. To the agent it was something specific — an emerging, cross-functional risk forming across executive leadership, HR, and finance, detected against a quiet baseline and surfaced while it was still unfolding.
OASIS estimates that staffing this kind of continuous monitoring and analysis by hand would take on the order of 100 hours of specialist data-science work per week, which is why most organisations never attempt it. That figure is not hours removed from an existing process — the capability did not exist to be staffed. That is the shift a COO should notice: not a cheaper version of an existing task, but a capability you could not previously put on the org chart.
The enterprise problem
The real operating picture does not live in the quarterly deck.
In any large organisation, the real operating picture lives in the messages, threads, and requests moving between people every day. That is where risk first appears, where processes quietly break or get worked around, and where a few people end up absorbing far more coordination than anyone has noticed.
The raw material is already there. What is missing, in almost every organisation, is a continuous, high-quality way to read it: to see themes forming across silos, to tell a genuine shift from a noisy week, and to turn that reading into a decision about who needs support and what needs to change. A COO could hire a small data-science team for occasional deep dives, but the skill is rare, the cost is high, and the work is never continuous.
OASIS took a different path. Rather than commission another report, it built the data scientist into the platform.
The trigger
A busy fortnight to everyone reading their inbox. Nothing in particular to the network.
Over one recent fortnight, OASIS member agencies began raising a cluster of provincial compliance and coverage questions: obligations that were shifting, how they interacted with billing and liability, and the policy and training implications that followed. The questions did not arrive under a single tidy heading. They appeared as separate threads on the executive list, parallel chains in finance, and related questions in HR and training.
To each person reading their own inbox, this registered as a hectic stretch. Across the network, with no one positioned to see all of it at once, it registered as nothing in particular.
What the agent saw
Not more email. A pattern.
Reading the network as a graph, the agent compared the fortnight against a quiet prior baseline and watched the compliance theme rise from near-zero to a dominant share of activity — an order-of-magnitude shift concentrated in days. More importantly, it saw that the shift was not contained in one place. The same cluster of topics was being discussed in parallel by executive leadership, finance, and HR, with shared participants and shared language tying the threads together.
The agent named it for what it was: a single cross-functional risk forming across the organisation, not three separate busy patches. It surfaced this in near real time, rather than weeks later through anecdote.
A theme invisible the week before becomes the dominant signal. The y-axis is a relative index, never raw counts.
Flashpoints and smouldering issues
The same week behaved two completely different ways.
The agent then looked at how each thread behaved — how many people joined, how fast the first reply arrived, how activity was distributed through the day. That let it separate two very different things. For a COO, the distinction is the insight: it shows, at a glance, where the organisation’s attention was correctly aligned to risk and where it was lagging behind it.
An immediate swarm
Some threads drew a swarm of senior people within minutes, replies landing back to back, often spilling into the evening. Attention was correctly aligned to the risk.
- Senior participants join within minutes
- First reply in minutes, not hours
- Activity sustained late into the evening
Important, and left waiting
Others were smouldering — important questions about policy and training that sat for hours before anyone answered, despite carrying real operational weight. Attention was lagging behind the risk.
- Real operational weight, slow first response
- Hours of silence before engagement
- Easy to miss until it escalates
Workload and strain
A few people were holding far more than anyone had noticed.
Because the organisation was modelled as a graph, the agent could also see who sat at the centre of the response. Combining how central a person was to the critical threads with when they were active and how substantial their messages were, it surfaced a structural fact: a small number of people were absorbing a disproportionate share of the coordination — much of it after hours, and on the most demanding questions.
Reported anonymously and structurally, that is exactly the signal leadership needs. It points to where support should be added or load redistributed, before quiet effort becomes burnout, and it makes visible the informal coordinating roles that traditional reporting never captures.
Nodes sized by relative involvement; edges weighted by shared participation. The marked hub shows where a small number of people absorb a disproportionate share of the coordination. No names, no specific topics.
How it works, in plain language
The organisation on a graph, read continuously.
The organisation's operating reality is modelled as a knowledge graph — people, teams, threads, topics, and time, with the relationships between them. The data scientist agent reads that graph continuously, using graph data science to find what is forming, what is anomalous, and who is connected to what.
A knowledge graph of the organisation
People, teams, threads, topics, and time, modelled as a network with the relationships between them. The operating reality, not the org chart — who actually talks to whom, about what, and when.
Graph data science, continuously
Centrality, community detection, embeddings, and change-point detection run against the graph to surface what is forming, what is anomalous, and who is connected to what. It does not stop at "activity went up" — it explains why, and shows which parts of the organisation are pulled in.
Decisions, not dashboards
It says what to do next: where attention is misaligned to risk, who is overloaded, which question needs a policy decision. Leaders stay on top of the loop, deciding and acting. The agent does the monitoring and analysing beneath them — work no one has to remember to start and no one has to learn to run.
Impact
A capability that did not previously exist to be bought.
Before this, answering the simple but vital questions — did something actually change, or is this just a loud week? who is overloaded? is this one issue or three? — would have meant a specialist exporting data, categorising it by hand, building a one-off analysis, and explaining it, if the organisation had the people to do so at all. OASIS does not. The capability is net-new. Now it runs continuously inside the platform, and that time is spent deciding and acting instead.
An assumptive figure: the specialist data-science effort it would take to staff continuous monitoring and analysis of the full communication graph by hand. Not hours removed from an existing process — there was no before.
Generalising across the enterprise
The same agent, in any function where people coordinate through messages.
The OASIS story sits in member services, but the pattern is general. The ingredients do not change: put the operating reality on a knowledge graph, give an agent the ability to read it with graph data science, and let it turn patterns into recommendations.
Customer operations
Detects emerging pain points across channels, routes them, and tracks frontline strain.
Incident management
Maps threads and on-call activity to find systemic issues and under-resourced teams.
Supply chain
Surfaces cross-functional risk — from supplier delay to quality issue — before it lands as a missed target.
Close
You stop guessing at what is happening.
This was not one lucky analysis. It is a structural change in what an organisation can do: a PhD-level data scientist built into its fabric, mapped to everything, working without pause.
“When your organisation lives on a knowledge graph and an agent can read it with the judgement of a data scientist, you stop guessing at what is happening. You watch the organisation think, in real time, with a data scientist beside you.”
— OASIS × Karmaflow
Questions a COO asks
What an organisational intelligence agent actually does.
Can an AI agent detect cross-functional risk across teams from internal messages?
Yes. OASIS models its people, teams, threads, and topics as a knowledge graph, and a data scientist agent reads it continuously. In one recent fortnight the agent detected an emerging compliance and coverage risk forming in parallel across executive leadership, HR, and finance — a single cross-functional risk that no individual inbox revealed — and surfaced it in near real time rather than weeks later through anecdote.
What is graph data science, and how does it let an agent read an organisation?
Graph data science is the discipline of analysing a network: centrality, community detection, embeddings, and change-point detection. Applied to an organisation modelled as a knowledge graph of people, teams, threads, topics, and time, it finds what is forming, what is anomalous, and who is connected to what, measured against a baseline of normal activity.
How is this different from a dashboard or a BI tool?
A dashboard reports that activity went up. The agent explains why, shows which parts of the organisation are pulled in, and says what to do next — where attention is misaligned to risk, who is overloaded, which question needs a decision. It returns decisions, not more reports, and it runs continuously without anyone having to remember to start it.
Can it tell a genuine shift from a noisy week?
Yes. The agent compares current activity against a quiet prior baseline using change-point detection, so a real, order-of-magnitude shift is separated from an ordinary busy stretch. It also distinguishes flashpoints — threads that draw immediate senior attention — from smouldering issues that carry real operational weight but sit for hours before anyone answers.
Can it show who is overloaded or at risk of burnout?
Reported anonymously and structurally, yes. By combining how central a person is to the critical threads with when they are active and how substantial their contributions are, the agent surfaces where a small number of people absorb a disproportionate share of the coordination, often after hours, so support can be added or load redistributed before quiet effort becomes burnout.
What would it cost to staff this kind of monitoring by hand?
OASIS estimates that continuous monitoring and analysis of the full communication graph would take on the order of 100 hours of specialist data-science work per week. That skill is scarce and expensive, which is why the work was never done. It is a net-new capability built into the platform, not a cheaper version of an existing task.
About OASIS
OASIS represents a province-wide network of agencies supporting people with developmental disabilities, coordinating policy, compliance, and operations across its members.
About Karmaflow
The AI agent platform underneath. Businesses use Karmaflow to put their operating reality on a knowledge graph and give it agents — including a data scientist that reads the whole organisation and returns decisions, not dashboards.
Put your organisation on a graph.
Give it an agent that reads the whole network — finds risk before it surfaces, sees who is overloaded, and returns decisions instead of dashboards.
- Case Study
- Customer Story
- OASIS
- Knowledge Graph
- Graph Data Science
- Organisational Intelligence
- AI Data Scientist
- Operational Intelligence
- Conversation Intelligence
- Risk Detection
- Member Services