AI could do nearly 12% of your work already, says new study

For UK mortgage lenders and brokers, artificial intelligence is no longer a distant fintech buzzword. It is becoming a practical lever for how loans are originated, underwritten and serviced — and new research suggests it could already perform a sizeable share of the work your people do today.

MIT’s Project Iceberg, a large‑scale simulation of the U.S. labour market, finds that current AI tools are technically capable of performing tasks worth 11.7% of total wage value, or about US$1.2 trillion annually. The model represents around 151 million workers across 923 occupations and more than 32,000 skills. 

READ MORE: Yes, AI is smarter than you sometimes, but can it help clients with mortgages?

While the study is U.S.‑based, its implications travel well. The UK mortgage industry is built on exactly the kind of work Project Iceberg finds most exposed: document‑heavy, rule‑driven processes carried out at scale in lenders, building societies and broker networks.

What Project Iceberg actually measures

Project Iceberg does not simply speculate about “jobs of the future”. It compares, task by task, what workers do today with what existing AI systems can already do.

Each worker is modelled as an “agent” with a bundle of skills and tasks, mapped against thousands of AI tools in production. 

The key metric, the Iceberg Index, measures technical exposure: the share of an occupation’s wage bill tied to skills where AI has already demonstrated usable performance in at least one context. The authors stress that it “captures technical exposure, where AI can perform occupational tasks, not displacement outcomes or adoption timelines.” 

For UK mortgage employers, that means:

• The 11.7% figure is not a forecast of redundancies; it is a measure of how much work could be automated with today’s tools.
• The gap between technical possibility and your current operating model is now a strategic choice, not a constraint of the technology.

The real exposure is in back‑office, not just in tech

The public debate around AI has focused on coding assistants and tech layoffs. In Iceberg’s terms, that is just the visible “surface”.

Looking only at current AI adoption in computing and technology jobs, the researchers find exposure of about 2.2% of wage value — roughly US$211 billion — which they call the “Surface Index”. They describe this as “only the tip of the iceberg.” 

The real mass of capability “extends far below the surface through cognitive automation spanning administrative, financial, and professional services,” bringing the total to 11.7% and around US$1.2 trillion in wages. 

In other words, AI is already strong at exactly the kinds of tasks that dominate mortgage work:

• Reading and extracting information from documents
• Applying policy rules to standard cases
• Producing structured outputs such as letters, illustrations and reports

One summary notes that high Iceberg scores are driven by “cognitive work—financial analysis, administrative coordination, and professional services.” Swap in “affordability checks” and “case processing” and you are essentially looking at the back‑office of a lender or large brokerage.

Mortgage tasks AI can already do

The report doesn’t speak specifically about mortgages, but the tasks it highlights map neatly onto UK mortgage operations.

Areas with the highest near‑term automation potential include:

Document and data processing

• Reading and classifying payslips, bank statements, ID documents and property details
• Extracting income, commitments and other key fields into decisioning systems
• Flagging missing items or inconsistencies in application packs

Rules‑based assessment and case handling

• Eligibility screening against product criteria and policy rules
• Standard affordability checks using prescribed formulas and buffers
• Simple case triage and routing to the right underwriter or team

Standard communication and reporting

• Generating suitability letters, ESIS documents and standard customer correspondence
• Producing recurring MI reports from defined templates

The study notes that financial institutions already deploy AI for “document processing and analytical support,” and healthcare systems automate “administrative tasks.” Mortgage operations combine both patterns: financial logic on top of dense documentation.

This does not mean entire roles disappear overnight. It means:

• High‑volume, low‑complexity work in case processing, underwriting support and servicing is immediately exposed.
• Human effort will be pulled towards complex, borderline and vulnerable‑customer cases, broker relationships and exception handling.

The strategic question for UK mortgage employers

Project Iceberg was designed so governments can “identify exposure hotspots, prioritize training and infrastructure investments, and test interventions before committing billions to implementation.” For UK mortgage lenders and brokers, it effectively poses the question:

If AI can already perform a double‑digit share of the work behind your loan book, how will you use that capability?

There are three broad strategic options.

1. Cost‑first automation

Use AI primarily to reduce headcount in operations and servicing, driving down the cost‑to‑income ratio. This delivers fast savings but comes with risks:

• Loss of experience in complex underwriting and arrears management
• Customer frustration if vulnerable or unusual cases get stuck in rigid automated flows
• Conduct and regulatory risk if decision logic cannot be clearly explained to customers or the FCA

2. Productivity and service uplift

Hold overall headcount roughly steady while:

• Letting AI absorb future volume growth in standard business
• Re‑deploying people into relationship‑heavy and judgement‑heavy work: complex lending, retention, arrears support, broker service and cross‑sell

This path targets both efficiency and customer outcomes, but requires deliberate redesign of roles, training and incentives.

3. Business‑model change

Use AI as a trigger to rethink the broader model:

• What should be fully straight‑through (for example, simple remortgages), and what should always involve human advice?
• Where do you want to compete on speed and cost, and where on bespoke structuring and relationships?
• How can you use freed‑up capacity to build new offerings — such as proactive affordability coaching, real‑time portfolio risk monitoring or embedded mortgage solutions with partners?

Project Iceberg does not tell you which answer is right. It tells you the decision is now driven by strategy and regulation, not by whether the technology works.

Turning Iceberg into a mortgage planning tool

To make this research useful at board and EXCO level, mortgage employers can take four practical steps.

1. Quantify your own exposure by process

For originations, underwriting, completions and servicing:

• Break down roles into concrete tasks: data entry, document review, basic vs complex decisions, customer/broker contact.
• Classify tasks as high / medium / low AI suitability, using the patterns Iceberg highlights (document processing, administrative work, standard analysis). 

Estimate what share of total wage spend sits in high‑suitability tasks. That becomes your own internal “Iceberg Index”.

2. Model scenarios before you invest heavily

Run 3–5 year scenarios:

• Conservative: AI as an assistant, with minimal structural change
• Targeted automation: a defined percentage of high‑suitability tasks transitioned to AI, with headcount held flat as volumes grow
• Aggressive: large‑scale straight‑through processing for standard cases, with corresponding role redesign

For each, estimate the impact on:

• Cost‑to‑income and processing times
• NPS and broker satisfaction
• Workforce size and skill mix
• Conduct and operational risk

This mirrors how Iceberg is used in the public sector: to support “evidence-based planning as AI capabilities expand across the economy.” 

3. Decide your people and conduct stance up front

Technology teams will naturally focus on systems and vendors. Boards need parallel clarity on:

• What commitments you will make on retraining and redeployment for staff in highly exposed roles
• How you will maintain human oversight and explainability in credit decisions, in line with FCA expectations
• Where you will insist on human contact (for example, arrears management, vulnerable customers) even if automation could be technically applied

Without those decisions, AI adoption is likely to be driven piecemeal by cost pressures, increasing conduct and reputational risk.

4. Put AI exposure on the risk and remuneration agenda

Boards and risk committees should regularly see:

• An estimate of the share of wage bill and decision flow handled by AI‑suitable tasks now and under plan
• The control framework for AI in underwriting and servicing
• How management is balancing efficiency gains with customer outcomes and regulatory expectations

Remuneration committees should understand whether incentives favour blunt cost‑cutting or balanced, long‑term value creation.

The MIT team behind Iceberg is blunt: “The window to treat AI as a distant future issue is closing.” For the UK mortgage industry, operating in a tightly regulated market with thin margins and demanding customers, that window may be even narrower.

AI can already handle a meaningful slice of the work behind every mortgage you write. Whether that leads to faster, fairer and more sustainable lending — or to rushed automation and avoidable mis‑selling risks — will depend on the choices mortgage employers make now, before the iceberg is directly in their path.