AI mortgage processing: banks hand loan files to AI agents

AI mortgage processing is no longer a pilot. On 13 October 2026, TD, the sixth-largest bank in North America by assets, said it now uses agentic AI in two more steps of its home loan work. Five days earlier, the software company MeridianLink put a document agent for mortgages into service with its first customers. In the United States and Canada, AI agents now read borrower files, flag what is missing and prepare summaries for a person to approve.

This matters far beyond North America. A home loan is the slowest part of most property sales, and the buyer often waits on paperwork. Here is what changed, why it matters, and what it means for banks, agencies and buyers in Morocco and the Arab world.

What happened in mortgage AI this autumn

Four announcements show the same direction.

• TD (13 October 2026). The bank added two agentic AI capabilities to its real estate secured lending. For mortgage funding, the AI compiles client and property details into a short case summary, so staff can finish the first review in up to half the time. For mortgage discharge, it gathers what is needed to prepare the discharge statement, and staff check the result before it goes out. TD says a first agent, launched in May 2026 for pre-adjudication, cut average processing time from 15 hours to under three minutes.
• MeridianLink (8 October 2026). Its Doc Agent for Mortgage, nicknamed “Millie”, checks submitted documents for completeness and quality, extracts the data so nobody retypes it, and turns lending rules into plain-language document requests for the borrower. It is live with early customers, including Credit Union of Colorado, with general availability expected in late 2026.
• Blend (July 2026). Blend moved its Autopilot assistant from pilot to production with five lenders, after a four-month preview on more than 25,500 loans. It reviews documents as they arrive, checks them against lending guidelines, flags missing or inconsistent information and calculates qualifying income before the file reaches an underwriter.
• nCino (August 2026). nCino released Mortgage MCP, which lets lenders connect AI agents to its mortgage software through the Model Context Protocol. Agents act only within the user’s existing permissions, every action is logged, and high-impact actions can require human confirmation.

These figures come from the companies themselves. They are not independent audits, and TD’s 15-hours-to-3-minutes result covers one step, not the whole loan.

Why agentic AI fits mortgage work

A mortgage file is a long chain of small checks: payslips, bank statements, identity documents, the sale agreement, the valuation, insurance. Each check is simple. The chain is slow because people wait for each other and for missing papers.

That is exactly the kind of work an AI agent handles well. Unlike a chatbot that only answers questions, an agent can open a document, compare it with a rule, write a request and update a system. We explain the difference in AI agents vs chatbots. Three patterns repeat in every announcement:

• Catch missing documents early. The biggest delay is a file that comes back incomplete. Agents ask for the right paper on day one.
• Prepare, do not decide. The AI writes the summary or the calculation. A person still approves the loan.
• Connect through standards. Blend and nCino both use MCP, the same protocol property platforms such as Lofty adopted this year. Our guide What is MCP explains how it works.

What agencies and buyers can do now

Banks will set the pace of AI mortgage processing. Agencies and buyers can still prepare for it today.

1. Give buyers a document checklist on the first contact. Most delays start with a missing paper. A clear list, sent in the buyer’s language, saves days.
2. Collect documents digitally and name them clearly. Clean, readable files are what an AI agent at the bank can process quickly.
3. Know the bank simulators. Several Moroccan banks already offer online simulations. Point serious buyers to them early.
4. Answer fast, then hand over. A buyer who asks about financing at night should get a first reply at once and a follow-up from a person the next day.
5. Keep advice human. The rate, the choice between classic and participative finance and the final decision stay with the bank and the buyer.

That is the role of Quick Agent, our assistant for agencies: it gives the first reply on WhatsApp, email and Instagram, logs each request and prepares the follow-up, so no financing question waits until morning. For context on public support for buyers, read our article on Morocco’s housing aid.

Conclusion

In 2026, AI agents started doing the slow, repetitive part of mortgage work in North America: reading documents, chasing missing papers and preparing summaries, while people keep the decision. Morocco has the volume and the first local tools. The agencies that prepare complete, clean files will feel the benefit first. kawnium has no link with TD, MeridianLink, Blend, nCino, AFDAL or the other companies mentioned.

Sources

1. TD Bank Group – TD expands agentic AI across mortgage operations (13 October 2026)
2. Business Wire via Yahoo Finance – MeridianLink Doc Agent for Mortgage now live with customers (8 October 2026)
3. National Mortgage Professional – Blend takes agentic AI live with first mortgage lenders (9 July 2026)
4. nCino – nCino releases Mortgage MCP to connect AI agents (7 August 2026)
5. Médias24 – AFDAL launches a voice AI agent for home loans (8 May 2026, in French)
6. Le Matin – Bank credit at the end of August 2026, Bank Al-Maghrib figures (5 October 2026, in French)
7. La Vie éco – Participative housing finance reaches 32.3 billion dirhams at the end of August (2 October 2026, in French)