The Social Investment Agency presents predictive risk modelling as neutral, necessary, and life-saving. It will prevent harm, ensure accountability, and guide evidence-based decisions about who poses a danger to children. The predictive risk models sit atop the Integrated Data Infrastructure, a system that collects, links, and stores every interaction whānau Māori have with state agencies across health, education, justice, benefits, housing, and employment. This data flows one way: into permanent profiles that follow whānau across generations.
So let us take them at their word.
If these tools truly identify risk and prevent future harm based on historical patterns of behaviour, then let us apply them where the evidence demands. Let us model the state. The state prosecutes whānau Māori for abuse and neglect, then removes tamariki from their care. Yet the state itself is a failed parent and documented perpetrator of harm. The cauldron calls the kettle black but refuses to apply that logic to itself.
The State as High-Risk Entity
Using the SIA’s own framework, what does the data tell us about New Zealand state agencies as predictors of harm to children?
Between 80 and 90 percent of Māori gang whānau members experienced abuse in state care, according to the Royal Commission of Inquiry into Abuse in Care. Our ‘Gang Whānau Healing Intergenerational Trauma’ research confirms this pattern. The state removed these children, claiming to protect them. Instead, it subjected them to violence, neglect, and institutional abuse that scarred them for life. Now, as adults building families and healing from colonial trauma, that same state uses their childhood abuse (the abuse the state inflicted when it removed them from their whānau) as evidence they pose a risk to their own children.
The IDI ensures this abuse becomes permanent evidence. Every interaction (a parent seeking family violence support, accessing addiction services, asking for housing or financial assistance) enriches the data profile rather than being recognised as help-seeking behaviour. The same system that fails to track institutional patterns of harm meticulously catalogues every moment of struggle in Māori whānau lives.
In predictive risk modelling terms, this is what researchers call a “feedback loop.” The state creates the harm, records the harm, then uses that record to justify further intervention. If we were modelling any other entity with this pattern, the algorithm would flag it as extremely high risk.
Nine percent of children in Oranga Tamariki care were abused or neglected in the 2023/24 year (507 children). This represents a significant increase from 2019, when almost six percent of children in care were harmed. For children returned home, thirteen percent were harmed within the monitoring period. In youth residential facilities, twenty percent of children were abused. Of the harm in these facilities, 84 percent was caused by other children and eleven percent by staff members. Current, documented, measurable harm occurs under state supervision.
The data on long-term outcomes is even more damning. Māori who spent time in the Oranga Tamariki system are nine times more likely to be in emergency housing, six times more likely to be hospitalised for self-harm, and significantly more likely to be victims of crime than Māori who were never in state care. Ninety-two percent of young Māori in youth justice had previous care and protection reports. Being in state care is linked to higher rates of prison admission, unemployment, and suicide.
If we applied the SIA’s actuarial logic to this data, the state would be classified as a forward liability. It causes measurable harm, destabilises families, and produces negative long-term outcomes at rates far exceeding those of the communities it targets. The fiscal cost alone should trigger intervention. The state spends hundreds of millions managing the consequences of its own failures while refusing to measure the return on that investment.
The Colonial Project as Risk Factor
Now let us expand the model. What if we treated colonisation itself as a risk factor and measured the state’s historical patterns of harm against Māori?
Land confiscation. Forced removal of children over multiple generations. Suppression of te reo Māori. Economic dispossession. Over-policing. Disproportionate imprisonment rates. Erosion of tikanga-based systems of care and accountability.
Systematic patterns of behaviour documented across more than 180 years. The evidence shows clear intent, consistency, and measurable outcomes. Every one of these actions causes intergenerational trauma, family separation, economic instability, and disconnection from cultural identity (the very factors the SIA now uses to classify whānau Māori as high-risk).
If a Māori parent had this track record, they would be under surveillance. If a gang member had this pattern of behaviour, they would be monitored, restricted, and subjected to intensive intervention. Yet when the state demonstrates these same patterns across generations, it positions itself as the protector, the expert, the authority with the data and the tools to determine who is safe.
Accountability Deficit
Through the Integrated Data Infrastructure, the SIA demands transparency from whānau Māori. It integrates data across agencies to build comprehensive profiles of individuals based on their involvement with police, benefits, health services, education, and child protection. It links this data across generations. It stores it. It analyses it. It uses it to make decisions about the future. Whānau cannot access the full picture of what’s held about them, cannot correct inaccuracies, cannot see how agencies share their information, and cannot opt out. The surveillance is comprehensive, permanent, and unidirectional.
Yet the state exempts itself from the same scrutiny.
If we had applied risk assessment models to senior officials throughout their careers, perhaps those who later engaged in misconduct would never have risen through the ranks. If predictive analytics had flagged patterns of institutional failure, perhaps Oranga Tamariki would not have consistently failed to meet minimum care standards for five consecutive years. If data-driven intervention redirected resources based on evidence of harm, state care abuse would not increase year after year.
The standards placed on Māori parents are not applied within the institutions that judge them, pathologise them, or seek to fix them. The surveillance is unidirectional. The data flows one way. The purpose of these tools is to protect the state from accountability.
What Predictive Risk Modelling the State Would Actually Look Like
If we were serious about using these data systems and predictive tools to prevent harm, here is what we would model:
Risk of child removal based on agency behaviour: Which Oranga Tamariki regions have the highest rates of unsubstantiated investigations? Which social workers have the highest rates of removing Māori children compared to non-Māori children in similar circumstances? What are the predictive indicators that a child removed by the state will be harmed in care?
Risk of institutional abuse: What are the risk factors within state care facilities that predict abuse by staff or other children? Which residential facilities have the highest rates of harm? What systemic patterns precede those incidents?
Risk of long-term harm from state intervention: Can we predict which children removed from whānau will experience worse outcomes than if they had remained at home with support? What is the threshold at which state intervention becomes more harmful than the original situation?
Risk of discriminatory practice: Which policies and practices correlate with disproportionate Māori involvement in state systems? Can we identify which agencies, regions, or individuals demonstrate patterns of bias in their decision-making?
Risk of policy failure: What is the predictive validity of past government interventions in Māori communities? Which policy approaches have consistently failed to improve outcomes? What are the indicators that a new policy will replicate past harms?
These are the models we do not see. These are the algorithms the SIA will never build. Predictive risk modelling was never designed to scrutinise power. It was designed to manage populations and justify intervention in the lives of those already marginalised while shielding institutions from examination.
The Evidence Our Research Provides
Our research with gang whānau shows what happens when the state refuses this accountability. Wāhine told us about Police sitting outside their homes. Social workers arriving without notice. The most senior people in OT being directed by Police to target them. Every minor injury becoming a file note, an investigation, and reason to uplift tamariki and mokopuna. The surveillance drained their energy and capacity to do the healing mahi they were already engaged with across their gang whānau.
These wāhine were building strong, culturally grounded families. They were working, paying taxes, raising tamariki, and breaking cycles of trauma. But the state treated their own childhood abuse (abuse the state inflicted when it removed them from their whānau) as evidence their children were unsafe.
Many lost children to state care based on assumptions and scaled assessments of the likelihood of causing harm. Many of those children were harmed in care. The very outcome the removal was supposed to prevent was caused by the removal itself.
In our Hauraki Māori Weathering Cyclone Gabrielle research, we documented how whānau-led responses saved lives while state agencies arrived late. In He Whare, He Taonga, we showed that housing discrimination and institutional failures (not Māori parenting) drove instability. In the Takatāpui, Rainbow and MVPFAFF+ Survivor Report, participants told us that whānau support and community-controlled services were the strongest protective factors, while state intervention consistently caused them harm.
The pattern across all our research shows the deepest harms come from state institutions, not from whānau.
Why This Matters
The Social Investment Agency positions its approach as innovative, evidence-based, and child-centred. But evidence is only evidence if you are willing to see what it shows. The data overwhelmingly demonstrates that state intervention in Māori families produces negative outcomes. It increases harm. It destabilises whānau. It perpetuates trauma across generations. Yet the response is always more surveillance, more data collection, more risk assessment tools aimed at Māori communities.
The SIA practices evidence-resistant decision-making dressed up as science.
If the SIA genuinely believed its own rhetoric about data-driven decision-making, it would model the state. It would use predictive analytics to identify which agencies and policies pose the greatest risk to tamariki Māori. It would implement transparency mechanisms that allow communities to see how decisions are made and who is accountable when those decisions cause harm. It would redirect resources away from interventions that consistently fail and toward whānau-led, community-controlled approaches that the evidence shows work.
The Refusal to Model the State Is the Point
The refusal to apply predictive risk modelling to the state reveals the entire logic of the system. Colonial states do not build tools to examine their own violence. They build tools to manage the populations they have dispossessed, displaced, and harmed.
The SIA knows what the data would show. The state is the highest-risk entity in the lives of Māori children. State care produces worse outcomes than remaining with whānau. Surveillance destabilises families rather than protecting them. Institutions designed to help Māori perpetuate cycles of harm instead. That is why the model will never be turned around. That is why the surveillance remains unidirectional. That is why the accountability only flows one way.
One of the wāhine from our gang whānau research spoke about this and how state systems approach whānau Māori: “They don’t fucking work for us. What they do is they open up our mamae. It’s like they come to you, they open up our fucking wounds with a surgical, unclean fucking knife. They whip it open for us, get us to share what’s going on, and then have no idea about how to fucking close the bastard back up. There’s no resolutions for it.”
There is no genuine apology in the Social Investment Fund. There is no behaviour change. Only new language for the same systems of control.
Another participant said something else that cuts to the heart of this: “Our jails are full of our men and women that are holding intergenerational trauma that was inflicted by them to begin with. Yes, some of us have become our own abusers, but where the fuck did that infiltrate into our bloodstream? We know exactly where it came in.”
Our mokopuna carry whakapapa older than any government, stronger than any algorithm, more enduring than any policy. They are part of a history that has survived every form of state intrusion and will continue to survive. Forward liabilities? Lab rats? Data points to be calculated, cohorts to be managed, risks to be modelled? No. They are descendants of ancestors who navigated this whenua long before these institutions existed.
The evidence across all our research shows tamariki Māori grow up stronger with their own whānau. The state has repeatedly and measurably proven that it cannot be trusted. If we applied the same tools, standards, and scrutiny that the SIA applies to Māori communities, the data would be irrefutable.
The state is the risk. The state is the harm. The state is the biggest cost. And no amount of data will make that truth disappear.
—
Author’s Note: This analysis is written in my independent capacity as a researcher. Te Whāriki Manawāhine O Hauraki engages with the Social Investment Agency as part of our broader work. This engagement reflects a pragmatic reality: if we do not participate, funding will go to Pākehā organisations or those who claim to represent Māori interests without genuine connection to our communities. Our participation allows us to advocate for and provide services to our own people, even as we critique the fundamental assumptions and power structures of these systems.
POUHINE 