New MIT research shows mortgage industry could already slash costs - substantially
Artificial intelligence is no longer just a talking point at overseas conferences. For New Zealand’s mortgage industry, it is starting to turn up inside broker CRMs, bank credit centres and non‑bank processing hubs — and new research suggests it could already do a significant share of the work those systems contain.
MIT’s Project Iceberg, a large‑scale simulation of the United States labour market, finds that today’s AI tools are technically capable of performing tasks worth 11.7 per cent of total US wage value — about US$1.2 trillion a year. The model represents roughly 151 million workers across 923 occupations and more than 32,000 distinct skills.
READ MORE: Yes, AI is smarter than you sometimes, but can it help clients with mortgages?
New Zealand’s mortgage sector is smaller, but structurally similar: heavy on documentation, governed by strict rules, and powered by large volumes of repetitive administrative work. The sort of work Project Iceberg finds most exposed.
A closer look at what the study actually measures
Project Iceberg doesn’t try to guess how many people will be out of work in five years. Instead, it asks a more targeted question: for each occupation, what share of its skills are ones that AI has already demonstrated it can perform in at least one real‑world context?
To do this, the researchers treat each worker as an “agent” with a bundle of skills and tasks, then match those against thousands of existing AI tools. The result is the Iceberg Index, a measure of technical exposure — “where AI can perform occupational tasks, not displacement outcomes or adoption timelines.”
For Kiwi mortgage professionals, that distinction matters. The 11.7 per cent headline is not a prediction that one in nine workers will lose their jobs. It is a sign that a sizeable portion of the work sitting inside mortgages — particularly the repeatable, rules‑based parts — could be done by machines today, if lenders and brokers chose to let them.
The visible tech jobs are just the “tip of the iceberg”
Most of the global commentary around AI has focused on coders and data scientists. Project Iceberg suggests they are only the visible edge.
When the researchers look solely at current AI adoption in computing and technology jobs, they find exposure of about 2.2 per cent of total wage value — roughly US$211 billion. They label this the “Surface Index” and describe it as “only the tip of the iceberg.”
Below the waterline, technical capability “extends far below the surface through cognitive automation spanning administrative, financial, and professional services,” bringing total exposure to 11.7 per cent and about US$1.2 trillion in wages.
One summary of the work notes that high Iceberg scores in several US states are driven by “cognitive work—financial analysis, administrative coordination, and professional services.” Swap in “loan processing” and “credit support” and you have a fair description of the back‑office engine that powers New Zealand’s home‑loan market.
What parts of a Kiwi mortgage file are most exposed?
The Iceberg report doesn’t single out “mortgage adviser” or “credit assessor” by name. Instead, it focuses on the kinds of tasks AI already performs well: reading documents, extracting data, applying rules and generating standard text.
Those tasks will be very familiar to anyone in the industry.
Document and data processing
The researchers point out that financial institutions are already using AI for “document processing and analytical support,” and that healthcare systems are automating “administrative tasks.”
New Zealand mortgage operations combine both patterns. Every day, lenders and advisers handle payslips, IRD summaries, bank statements, credit reports, sale and purchase agreements, build contracts and more. They pull out income, commitments and spending figures, key in the numbers, and check that nothing important is missing.
From Iceberg’s perspective, these are prime candidates for automation.
Rules, ratios and triage
The same is true of policy rules and triage. Eligibility against product criteria, serviceability calculations, loan‑to‑value checks and sorting clean files from messy ones are all rule‑driven. Technically fiddly to code, yes — but conceptually straightforward for software built to recognise patterns and apply conditions.
The study’s message is that, measured purely on capability, today’s AI could already handle a surprising share of these steps.
Templates and compliance
Much of the industry’s written output — pre‑approvals, conditional offers, variation letters, adverse‑action notices — follows tightly prescribed templates set by compliance teams. Names, amounts and dates change; the wording doesn’t.
For modern language models, this kind of structured, repetitive writing is very much within reach.
None of this means that mortgage professionals are suddenly surplus to requirements. It does mean that the parts of their day devoted to keying data, sorting documents and filling in standard letters are where AI is most likely to bite first.
A sector with decisions to make
Although Project Iceberg was built to help governments plan, it raises pointed questions for mortgage businesses on this side of the Tasman.
One of the project’s aims is to help policymakers “identify exposure hotspots, prioritize training and infrastructure investments, and test interventions before committing billions to implementation.” That same logic applies to banks, non‑banks and adviser groups contemplating new technology and staffing decisions.
Three questions stand out for New Zealand.
Where, exactly, is your own iceberg?
Because the Iceberg Index is built on skills, not job titles, exposure varies even within a single role. An adviser who spends most of their time in front of clients is in a different position from one who spends hours wrestling with paperwork. A credit assessor handling complex self‑employed files has a different exposure profile from someone assessing straightforward PAYE applications against standard policy rules.
For any given firm, the first step is to break jobs into tasks: gathering documents, entering data, running calculators, checking policy, explaining outcomes, working with borrowers and referrers. Only then can leaders estimate what share of their wage bill sits in work that looks like the document processing and administrative tasks Iceberg highlights.
What will you do with the 10–15 per cent AI can reach?
Iceberg’s headline number poses an uncomfortable strategic choice. Some institutions will be tempted to use AI primarily as a cost‑cutting tool, reducing operational headcount as machines take over routine work. Others may try to hold staffing levels steadier, allowing AI to soak up volume and re‑work while shifting people into areas that resist automation: complex lending, hardship and arrears support, face‑to‑face advice, and outreach to under‑served communities.
In a market where borrowers already complain about delays and opaque decisions, the way that balance is struck will shape reputations as much as bottom lines.
Can you explain your models — to regulators and customers?
The researchers note that traditional indicators such as GDP, income and unemployment explain “less than 5% of this skills-based variation” in exposure. In other words, change will show up deep inside processes long before it hits official statistics.
For New Zealand lenders and advisers, it also raises familiar questions in a new form. If AI is involved in assessing serviceability or flagging files, can you explain its decisions to borrowers and to regulators? Are vulnerable customers being treated fairly? Does automation risk embedding bias between different regions or customer groups?
The window is closing
Project Iceberg is, in essence, an early‑warning system. It was built so societies can prepare for AI‑driven change before it shows up as closed branches, lost livelihoods or unexpected bottlenecks in critical services. The authors put it bluntly: “The window to treat AI as a distant future issue is closing.”
For the New Zealand mortgage profession, that window feels particularly narrow. The sector sits at the junction of finance and paperwork, under increasing regulatory scrutiny and keen public interest in housing affordability.
AI is already capable of doing a meaningful slice of the work behind every approval and decline: scanning and sorting documents, running first‑pass checks, drafting standard letters. The question is no longer whether the technology can do it. It’s whether the industry chooses to use those capabilities to deliver faster, clearer, fairer lending — or simply as a quieter way to cut costs.
Project Iceberg doesn’t dictate the answer. It only makes the stakes harder to ignore.
