He Replaced a Lead Manager With AI: What It Actually Saved

By Mike DeHaan, Dan Austin, Dylan Koch · October 13, 2026

Most "AI in real estate" talk is theory. On this episode, Dylan Koch walked through something concrete: he built his own SMS follow-up automations with Claude, stopped paying a lead manager, and cut about $1,200 a month in overhead. Mike DeHaan and Dan Austin pushed back on how to frame that, then turned to what's happening on the debt side right now.

What did Dylan automate, and what did it replace?

Dylan said the unlock was mundane: the browser matters. He moved to Google Chrome, installed the REsimpli extension so Claude could interact with his CRM, and from there he described the workflow in plain language — if it's a new lead, follow up twice a day with these SMS templates; once a condition is met, kill the automation, hand it to Connor, tag him.

The result was that he let his lead manager go and dropped business overhead by roughly $1,200 a month. He was upfront that it wasn't purely an AI decision: he said the writing was on the wall because she had taken another job, and her call volume and new-lead numbers had already dropped off.

The tradeoff he named: the old role did calls and texts. He isn't having AI make calls, so this is text-only. Because his operation filters mainly for highly motivated sellers or high-spread leads, he said his acquisitions manager can handle most of the live conversations. The sticking point is 10DLC messaging compliance if he ever leaves his CRM — Mike and Dan's suggestion was to pipe texting through a third-party call system like JustCall rather than build it. For context, Dylan said his REsimpli plan runs $4,800 a year, about $400 a month.

• Browser and extension choice determined how much Claude could actually do
• Automations were described conversationally, not coded from scratch
• Text-only follow-up; live calls stayed with a human
• 10DLC compliance is the barrier to leaving a packaged CRM

Is cutting headcount the right way to think about this?

Dan argued it isn't. His framing: don't look at AI as a cost reduction, look at it as an efficiency gain. You may still need a given role, but your existing people absorb more capacity. Dylan said he's already seen that with his acquisitions manager, who used to log in to 70 tasks, half of which can now be automated away.

Mike gave an example from his own operation. A repeat document like a loan extension is now one button: the system reads the uploaded loan docs, knows the extension terms, auto-fills, confirms the recipient, and sends. A five-to-ten-minute task becomes ten seconds. He also described using an email agent to reconstruct why an $80,000 wire went out weeks earlier — it traced the chain and produced a likely answer in about fifteen seconds, a job he said would have taken him at least half an hour.

On accuracy, Mike was blunt that these systems make mistakes and need review. When his team launched their internal build he was fielding several bug fixes a day; now it's closer to two a week, and the remaining issues are narrow single-use cases. Dylan said running comps is one of the last things he'd hand over, because value can change street by street and bad comps make everything downstream wrong. Mike's counter: let it pull comps, then correct it with the ones you'd actually use and explain why, so that reasoning carries forward.

What's happening with lending right now?

Dylan said a Cincinnati multifamily owner he knows — a 66-to-68 unit class A value-add bought roughly two years ago — was approached through a broker about a refinance with a large insurance company acting as the direct lender. Dylan said the loan amount discussed was around $6 million, the terms looked close to agency debt but without recourse, and the quote was a ten-year rate around 6.25%. Dan noted insurance companies are commonly investors in debt funds, but going direct to the borrower was new to him. With the ten-year treasury around 5.25%, Dylan's question was whether that spread pays the lender for the risk.

On the private side, Mike said DSCR shops are shutting down entirely — one of their own investors emailed that they were closing their DSCR business. His read: either the spread between what capital partners expect and what borrowers will pay no longer works, or they originated loans sitting on a warehouse line that are now underwater as rates moved.

Mike also said United Wholesale Mortgage notified him it is bringing back essentially no credit floor on its conventional, owner-occupied products, with pricing and LTV adjusting accordingly. His view is that the borrower pool has shrunk and large shops are reaching further down to generate revenue. Dan tied it to conventional transaction volume being very slow.

The number operators should note: Mike said he saw the average conventional mortgage around 7.875%, while non-QM business-purpose investor loans are pricing roughly a full percentage point cheaper — a reversal of the historical relationship. Dylan's own cautionary example was a lease-option deal where he took an interest-only loan tied to prime; he's already absorbed two rate increases plus a surprise mechanical repair, and he's not sure his tenant-buyer will still qualify in two years. Mike's take on fixed versus adjustable: people avoid locking a rate because they fear being stuck, but he doesn't see a return to 4%. If it happens, pay the prepayment penalty or wrap it into the new loan — if the deal had real equity going in, you'll be fine.

The full episode has more on the automation build, the insurance-company debt quote, and where the hosts think private capital is moving now. Listen to EP 513 of Collecting Keys.

AI & TechPrivate Money & Lending

Educational content from the Collecting Keys podcast. Not financial, legal or tax advice.