I built FindRE, an AI-powered CRM that a real estate investment team runs acquisitions in, while acting as their fractional CTO. Before that they ran the same work across a marketing CRM, a separate dialer, a texting app, an e-signature service and spreadsheets. This guide is what I learned about where AI genuinely helps an investor, where it is hype, and how to set it up so it keeps working after the demo.
Where AI actually helps an investor
Most of an investor's day is not deciding. It is reading, calling, texting and remembering to follow up. That is where AI pays, because those are the jobs that follow a pattern but still need words.
| Job | What AI does well | What a person still does |
|---|---|---|
| Finding leads in public records | Reads lists, notices and records that a person would never get through, and turns them into clean, structured leads | Decides which lists matter and what a good lead looks like |
| Working the phone | Writes the call summary and proposes the next step, so nothing lives only in someone's head | Makes the call, has the conversation |
| Replying to sellers | Drafts the reply to an incoming text or email in the tone you set | Reads the draft and sends it, or does not |
| Keeping follow-up alive | Nothing. This is rules, not AI, and it is the part that makes the rest worth having | Sets the rules once |
Notice the last row. The single biggest improvement for the team behind FindRE was not a clever model. It was a rule engine that runs the predictable steps, so a follow-up never depends on someone remembering. AI wrote the summaries; rules made sure the summaries led somewhere.
What AI is bad at
- Knowing your buy box. A model can score a lead against criteria you write down. It cannot know that a certain street never sells, or that a particular type of seller wastes your time. Keep the criteria in your hands and let the AI apply them.
- Valuing a property. It can prepare the arithmetic and gather comparable sales. The decision needs local knowledge and facts that are not in any record.
- Negotiating. A drafted reply is a starting point. The moment a conversation turns into a negotiation, a person should be writing every word.
- Being trusted blindly. Anything a seller sees should be approved by a person. The time saving is nearly the same as full automation, and the risk is a fraction.
The National Association of Realtors keeps a useful, sober overview of where AI is being used across the industry. Read the tool pitches against it.
Buy a tool or build a system?
Start by buying. If your team is small, your process is ordinary and one of the rented tools fits, connect AI to it and stop there.
Build when the process is the advantage. The team behind FindRE had a way of working that no single product matched, so they had stitched together five tools and were losing leads in the handoffs between them. Every handoff was a gap: the call happened in one tool, the notes lived in another, the contract went out from a third. A system built around the process removed the gaps, and AI could then sit where the work happens instead of in a separate window.
The test is simple. If you are paying for several tools, moving information between them by hand and still losing leads, you have outgrown buying.
How to set it up so it keeps working
Pick one job
Start with the job that eats the most hours and has the clearest pattern. For most acquisition teams that is the call summary and next step. Do not start with "an AI that finds deals".
Put the data in one place
AI is only as good as what it can see. If the call log, the lead record and the message history live in three tools, the summary will be thin. One system first, AI second.
Keep a person on anything a seller sees
Summaries and next steps can be automatic. Replies are drafts until a person sends them. This one rule prevents almost every embarrassing story you have heard about AI in sales.
Cap the spend
Model providers bill for what the AI reads and writes. Set usage limits and track the cost per feature from the first week, so the bill never surprises you.
Read the log
Keep every AI output and what the team did with it. The drafts people rewrite tell you what to fix. The summaries nobody opens tell you what to remove.
What it looks like in FindRE
In FindRE the AI work is deliberately narrow. The team calls and texts sellers from the browser, and every call and message is logged against the lead. After a call, AI writes the summary and suggests the next step. When a message comes in, it drafts a reply for a person to review and send. A rule engine runs the routine follow-up. Skip tracing is built in, so a new lead arrives with contact details attached rather than as a name on a list.
Alongside the CRM, pipelines read public notices and lists and turn them into clean lead records, which is the reading-heavy work a person would never finish by hand.
Nothing in that list is exotic. It works because each piece has one job, limited power and a person where it counts. The team moved off their old marketing CRM, and what began as one team's system is now offered to other investors as a product.
Where to start
If you are running acquisitions across a patchwork of tools, the first step is not AI. It is a look at where leads go cold, followed by a decision about whether to connect AI to what you have or build one system around how you work. That look is what I do under custom CRMs, and the AI pieces are what I build under AI agents. You can read the full story of the build in the FindRE case study.