AI might be great - but can your staff actually do anything with it?
Mortgage lenders and originators across the United States who believe they are winning the race on artificial intelligence may be measuring the wrong things, according to a new report by Section AI that should prompt some uncomfortable conversations in C‑suites.
A survey of 5,000 knowledge workers at large companies in the US, UK and Canada finds that three years after ChatGPT's launch, only 2.7% of employees qualify as "AI practitioners" – people who have genuinely woven AI into their workflows and are seeing real productivity gains. Just 0.08% meet the bar for "AI experts". In total, 97% of the workforce is using AI poorly or not at all.
For US mortgage firms that have spent the past year rolling out AI tools, drafting usage policies and running training sessions, the findings are a stark reality check: most employees still don't know what to do with AI in the context of their actual jobs.
From basic prompts to real productivity
The report, produced by AI transformation firm Section, argues that "AI proficiency" meant one thing in 2025 and something much harder in 2026.
Last year's focus was on basic literacy: understanding what AI is, how to avoid data leaks and how to write a decent prompt. Many firms can now check those boxes. Staff know how to ask AI to clean up an email to a broker or summarize a property appraisal.
But the bar has shifted. Proficiency now means incorporating AI into meaningful, value-adding tasks every week – not sporadic experiments. For mortgage firms, that might mean automated extraction of income and employment data from bank statements and pay stubs, first‑pass affordability and DTI checks, automated risk flagging in edge cases, systematic review of compliance documentation, or AI‑assisted fraud detection in application packages.
The survey suggests most organizations haven't come close. Seventy percent (70%) of workers are "AI experimenters": they use AI for very basic tasks such as summarizing meeting notes, rewriting emails and getting quick answers. Another 28% are "AI novices" who rarely or never use AI. Less than 3% are practitioners or experts.
The impact on time saved is predictably modest. Twenty‑four percent (24%) of workers report saving no time at all with AI; another 44% save less than four hours per week. Only 6% say they save more than 12 hours weekly – the kind of number that might actually show up in processing costs, turn times and pull‑through rates.
The "use case desert" in mortgage operations
The central problem isn't that people can't prompt. It's that they don't know what to use AI for in their specific role.
Across respondents:
- 26% say they have no work‑related AI use case at all.
- 60% say their existing use cases are beginner‑level.
- After reviewing 4,500 reported work use cases, researchers judged only 15% likely to deliver genuine ROI.
The most common "most valuable" use case is using AI as a Google search replacement, cited by 14.1% of workers. Draft generation follows at 9.6% , grammar and tone editing at 5.7%. Basic data analysis sits at 3.8%. Task and process automation – the category mortgage firms should care most about – accounts for just 1.6%.
Overall, 59% of reported use cases are basic task assistance, more than a quarter have no meaningful role in larger workflows, and only 2% are rated as advanced.
For a US mortgage lender or broker, that often translates into loan officers asking AI to polish a borrower email, but not using it to systematically extract data from 1003s, pre‑populate condition checklists, flag potential fraud patterns or generate first‑pass underwriting narratives. AI remains a smarter spell‑checker rather than a re‑engineered origination or fulfillment engine.
Executives optimistic, frontline unconvinced
Perhaps the most uncomfortable finding is the gap between what leaders think is happening and what employees actually experience.
Among C‑suite respondents:
- 81% say their company has "a clear, actionable policy that effectively guides AI use."
- 80% say tools exist with a clear access process.
- 71% say there's a formal AI strategy.
- 66% feel encouraged to experiment and create their own AI solutions.
- 48% believe there's "widespread adoption with open sharing of use cases and best practices."
Individual contributors – the non‑managers doing most of the daily work in processing, underwriting support, servicing and quality control – see things very differently:
- Only 28% agree there's a clear, actionable AI policy – a 53‑point gap with the C‑suite.
- Just 39% say tools exist with clear access, versus 80 percent of executives.
- 32% say there's a formal AI strategy, compared to 71% of leaders.
- 25% feel encouraged to experiment and build solutions, versus 66% of the C‑suite.
- Only 8% think there's widespread adoption with open sharing of best practices, compared to 48% of senior leaders.
For US mortgage firms, where much of the repetitive, rules‑based work – income verification, document collection, file prep, compliance checks, QC reviews – is performed by individual contributors, that pattern should be deeply concerning. The people whose work could most benefit from AI automation are the last to get tools, training and managerial expectations.
Training isn't closing the gap
The report doesn't suggest firms are standing still. Sixty‑three percent (63%) of respondents say their company has an AI policy, 50% have access to a tool and 44% receive AI training.
These investments do make a difference. Employees at firms with a company AI strategy are 1.6 times more proficient than those without; those with tool access are 1.5 times more proficient; those who've been trained are 1.5 times more proficient; and those whose managers expect AI usage are 2.6 times more proficient than employees whose managers discourage it.
Yet after all that effort, the average worker who has undergone AI training scores 40 out of 100 in proficiency. Most remain in the "experimenter" band – they know what an LLM is and have a handful of low‑stakes use cases, but haven't started to explore intermediate or advanced applications.
The explanation is straightforward: most training programs are still aimed at the wrong target. They teach access, safety and prompting – how to use AI – but not how to identify and redesign workflows where AI can actually remove work or accelerate decisions.
Knowing that AI can summarize a document isn't the same as knowing how to redesign the mortgage application pipeline around it.
What US mortgage leaders should do next
For mortgage executives, the report points to several necessary shifts:
Stop measuring success by logins and course completions. If only 15% of use cases are value‑driving, adoption metrics are misleading. Start tracking time saved per loan file, cost per funded loan, days to clear and exception rates – the metrics that actually matter.
Treat use case development as a managed competency. Build function‑specific use case libraries for loan officers, processors, underwriters, closers, post‑close and servicing teams. Make use case development a formal responsibility for team leads and regional managers, not a side hobby.
Prioritize individual contributors. The people doing the most repetitive work have the least access and support. Standardize AI tool access across roles, ensure fair reimbursement policies and require every manager to identify and track at least three AI use cases for each direct report.
Shift training from "how to prompt" to "how to redesign workflows." Teach teams to map their own processes, identify bottlenecks and time sinks, and test AI in controlled slices of work. That's where real productivity lives – not in better‑worded emails.
Close the executive awareness gap. Leaders who rely only on adoption dashboards will keep overestimating progress. Shadow staff as they attempt to use AI in daily work. Insist on hard measures – hours saved per file, reduction in rework, faster decisioning – not just platform usage stats.
Accept that the proficiency bar will keep rising. The gap between casual experimenters and real practitioners will widen as AI capabilities advance. Build continuous learning infrastructure now – communities of practice, peer coaching, internal certification paths – rather than treating AI as a one‑and‑done training topic.
The bottom line for US mortgage firms
The report's underlying message is simple but uncomfortable: AI transformation in mortgage isn't primarily about buying technology. It's about redesigning how work gets done – loan by loan, workflow by workflow, role by role.
Generative AI can already read, extract, classify and draft at a level that would have seemed impossible just a few years ago. Whether that translates into lower cost per loan, faster turn times, better pull‑through rates and genuinely improved borrower experiences will depend far less on what tools are purchased in 2026, and far more on how quickly the industry can move beyond experimenting at the edges and get serious about redesigning the core.
The tools are ready. The question is whether mortgage leadership is ready to do the harder work of changing how the business actually operates.
