Using big data to save big money - and hopefully, do some people some good too - is the aim of the Government's social investment strategy.
The fiscal dimension is too easily dismissed. Welfare spending is one of the four big-ticket items in the Budget, along with health, education and superannuation.
The latest estimate from actuaries Taylor Fry of the net present value (NPV) of the projected lifetime costs to the welfare system of current beneficiaries - of whom there are around 550,000 - is $76 billion. If that liability was fully funded the way future obligations to current ACC claimants are, $76b is the lump sum that would need to be set aside.
We can agree, I think, that $76b represents serious opportunity cost, whether to taxpayers who pay for it or to other potential recipients of public spending, like the sick or the ignorant young. Their interests have to be weighed in the balance along with those of beneficiaries.
The actuaries say that compared with what they forecast in their 2012 valuation, before the welfare reforms kicked in, there has been a cumulative $2b reduction in welfare payments over four years, 70 per cent of which can be attributed to the reform package and operational changes.
In the 2015-16 year, they reckon that management reduced the NPV of the future liability by $1.7b. That was offset by (and arguably made possible) an additional $1.5b cost from the child hardship package in last year's Budget.
Another arresting number from the actuaries' report is their estimate that three-quarters of the liability (so, $57b) is attributable to people who entered the benefit system before the age of 20.
We can also agree, I think, that that is way too young for someone to be written off; their fate sealed.
Surely it makes sense, in human and financial terms, to put resources into identifying and helping the vulnerable young, reducing teen pregnancy rates and reducing the numbers classified as NEET (not in employment, education or training).
On the NEET measure, it's a case of so far, so not so good.
In the latest March quarter's household labour force survey there were 86,000 15- to 24-year-olds not in employment, education or training (including 16,000 classified as caregivers, most likely young mothers), representing 12.8 per cent of the age group. That compares with 82,000 or 12.4 per cent in March last year and 76,000 or 11.8 per cent two years ago.
The actuaries' report records a rise of 1000 to 77,000 in the number of people on "work ready" Job Seeker benefits over the 2015-16 year, despite a drop in the unemployment rate from 5.5 to 5 per cent over the same period.
But there had been 20 per cent more people in this category in 2012. "These results are suggestive of a 'hardening' of the segment, where the decrease in numbers seen in the segment over the past few years ... has left a group that has lower than expected exit rates," they suggest.
One of the criticisms levelled at the social investment approach is that such metrics are a poor proxy for social outcomes.
Authorised researchers can interrogate Statistics NZ's Integrated Data Infrastructure, a huge database which brings together information about us from a range of government agencies. It is anonymised, so the researchers cannot identify whose details they are looking at, only that they relate to the same person.
It turns out that some of those who have exited the welfare rolls have not gone into gainful employment, but into a life of crime. The welfare liability estimate does not include what an ex-beneficiary might cost Vote Police or Vote Corrections.
Or they might disappear from official sight altogether. The young guy begging outside Bunnings the other day probably doesn't disclose his earnings to the Inland Revenue.
While a person's information might be anonymised for researchers, it cannot be for the agencies tasked with targeting support.
Clearly, there is a trade-off with privacy concerns here. But the same is true of any number of areas of modern life in the digital age.
Google has just announced it will cease scanning the contents of Gmail users' emails for the purpose of targeting advertising. Well good, though it probably only shows Google already knows enough about us from other sources, like browsing histories.
But the protests of non-governmental agencies, like Women's Refuge, about a potential chilling effect if they are required hand over information about the people they help to the government, or lose funding, have to be taken seriously.
There are also misgivings about the limitations of "predictive risk modelling". If the actuaries have found that lifetime benefit costs are roughly 50 per cent higher for Maori or for those with "child protection history" than for the general population, that does not mean that someone ticking those boxes is necessarily a candidate for "wrap-around support".
But it is not irrelevant, either. It is more akin to risk factors a GP screens for if you go for a medical.
A more insidious danger is that if the Ministry of Social Development's performance is measured in terms such as reducing the number of people drawing such and such a benefit, or the estimated average number of years a client will do so, it could affect in a bad way the approach front-line staff take.
It could, but it need not and should not. The onus is on the department's hierarchy to ensure it does not.
The challenges of getting social investment right are no reason to dismiss the whole approach as hard-hearted or wrong-headed.