Built an AI Deal Analysis Tool for Fun

Austin, Texas
Rehabbing & House Flipping
1yFollow

Built an AI Deal Analysis Tool for Fun

Chris Magistrado
Investor · Member since 2022 · 78 posts · 42 votes

I built something to make deal analysis easier, and I’m pretty proud of it. It’s still rough and has plenty of room for improvement, but I can upload photos of a distressed property, tax records, comps, lender rates, and local rehab costs.

From there, it analyzes the photos to identify repairs, explains why they're needed, estimates costs for each item, calculates the total rehab budget, determines ARV based on comps, and provides an expected return—basically automating a lot of what spreadsheets already do.

It’s not meant to replace an analyst, and of course, everything still needs to be verified. It won’t catch everything in photos yet, and comps can be off, but as a way to speed up deal reviews and analyze more properties in less time, it’s been pretty useful. It can even learn from past deals to help refine the analysis over time.

I built this mostly to better understand the components of a good deal. Now that it’s up and running, I’m wondering—are there other tools out there that do something similar?

5Reply
41 views

Most Popular Reply

Jeff S.
Lender · Los Angeles, CA · Member since 2009 · 1k+ posts · 2k+ votes
1y

Sounds interesting, @Chris Magistrado. Of course, anyone or anything can make predictions, but the proof is in the pudding. I'd be curious how accurate your tool is with a comparison to real-world deals after they are sold. That is, how well did your deal analyzer estimates match the actual rehab costs, holding costs, ARV, and profit? A detailed before-and-after P&L comparison would be useful here to show accuracy and weaknesses.

I have a relatively sophisticated rehab spreadsheet I’ve been enhancing for years, and I know it’s reasonably accurate. It only knows what it’s told, however. Automating what spreadsheets already do is not useful since spreadsheets are already automated. I don’t think that’s what you meant. The top three expenses in a flip are the rehab costs, lender fees, and sales commissions. These typically account for about 80 to 85% of all expenses in a flip.

If you know the lenders' terms and the loan duration, lender fees, the second greatest expense, are easy to calculate. Agent sales fees, number three, which are normally a percentage of the ARV, are similarly trivial. You don't need AI for those. This leaves rehab costs, number one, and ARV, which is where the skill comes in.

Estimating today's ARV is fairly straightforward. A tool to accurately predict the ARV in 6 months or in a year, as well as the project's rehab costs, would be extremely valuable. I'm curious how well your tool could do that.

Will it tell you to go back and take a few more photos to get a better idea of the rehab costs? Since it will evaluate photos, bonus points if it could look at the MLS and recommend the best finishes and paint colors to maximize the ARV. Or the best MLS photos to take, how many, as well as staging styles. More bonus points if it could look back at the borrower's track record and predict how they will do on a particular type of rehab? I can ask for the world, can't I?

Everyone and their brother is now offering AI tools to evaluate loan documents and bank statements. Something to accurately evaluate a flip would be a welcome addition.

I’m curious to follow your progress, Chris. Sounds great.

1Reply
See this reply in the discussion

7 Replies

Jump to latestLatest
  • Michael Wyatt
    Lender · Boston, MA · Member since 2021 · 125 posts · 64 votes
    1y

    Backflip has a decent analysis tool on their app. Not to the extent of uploading photos and explaining repairs (that's pretty cool) but running comps, estimating ARV, reasonable rehab budget, profitability, etc.

    1Reply
  • Jeff S.
    Lender · Los Angeles, CA · Member since 2009 · 1k+ posts · 2k+ votes
    1y

    Sounds interesting, @Chris Magistrado. Of course, anyone or anything can make predictions, but the proof is in the pudding. I'd be curious how accurate your tool is with a comparison to real-world deals after they are sold. That is, how well did your deal analyzer estimates match the actual rehab costs, holding costs, ARV, and profit? A detailed before-and-after P&L comparison would be useful here to show accuracy and weaknesses.

    I have a relatively sophisticated rehab spreadsheet I’ve been enhancing for years, and I know it’s reasonably accurate. It only knows what it’s told, however. Automating what spreadsheets already do is not useful since spreadsheets are already automated. I don’t think that’s what you meant. The top three expenses in a flip are the rehab costs, lender fees, and sales commissions. These typically account for about 80 to 85% of all expenses in a flip.

    If you know the lenders' terms and the loan duration, lender fees, the second greatest expense, are easy to calculate. Agent sales fees, number three, which are normally a percentage of the ARV, are similarly trivial. You don't need AI for those. This leaves rehab costs, number one, and ARV, which is where the skill comes in.

    Estimating today's ARV is fairly straightforward. A tool to accurately predict the ARV in 6 months or in a year, as well as the project's rehab costs, would be extremely valuable. I'm curious how well your tool could do that.

    Will it tell you to go back and take a few more photos to get a better idea of the rehab costs? Since it will evaluate photos, bonus points if it could look at the MLS and recommend the best finishes and paint colors to maximize the ARV. Or the best MLS photos to take, how many, as well as staging styles. More bonus points if it could look back at the borrower's track record and predict how they will do on a particular type of rehab? I can ask for the world, can't I?

    Everyone and their brother is now offering AI tools to evaluate loan documents and bank statements. Something to accurately evaluate a flip would be a welcome addition.

    I’m curious to follow your progress, Chris. Sounds great.

    1Reply
  • Chris Magistrado
    OP
    Investor · Member since 2022 · 78 posts · 42 votes
    1y
    Quote from @Jeff S.:

    Sounds interesting, @Chris Magistrado. Of course, anyone or anything can make predictions, but the proof is in the pudding. I'd be curious how accurate your tool is with a comparison to real-world deals after they are sold. That is, how well did your deal analyzer estimates match the actual rehab costs, holding costs, ARV, and profit? A detailed before-and-after P&L comparison would be useful here to show accuracy and weaknesses.

    I have a relatively sophisticated rehab spreadsheet I’ve been enhancing for years, and I know it’s reasonably accurate. It only knows what it’s told, however. Automating what spreadsheets already do is not useful since spreadsheets are already automated. I don’t think that’s what you meant. The top three expenses in a flip are the rehab costs, lender fees, and sales commissions. These typically account for about 80 to 85% of all expenses in a flip.

    If you know the lenders' terms and the loan duration, lender fees, the second greatest expense, are easy to calculate. Agent sales fees, number three, which are normally a percentage of the ARV, are similarly trivial. You don't need AI for those. This leaves rehab costs, number one, and ARV, which is where the skill comes in.

    Estimating today's ARV is fairly straightforward. A tool to accurately predict the ARV in 6 months or in a year, as well as the project's rehab costs, would be extremely valuable. I'm curious how well your tool could do that.

    Will it tell you to go back and take a few more photos to get a better idea of the rehab costs? Since it will evaluate photos, bonus points if it could look at the MLS and recommend the best finishes and paint colors to maximize the ARV. Or the best MLS photos to take, how many, as well as staging styles. More bonus points if it could look back at the borrower's track record and predict how they will do on a particular type of rehab? I can ask for the world, can't I?

    Everyone and their brother is now offering AI tools to evaluate loan documents and bank statements. Something to accurately evaluate a flip would be a welcome addition.

    I’m curious to follow your progress, Chris. Sounds great.

    You’ve brought up some of the challenges I faced in this. Accuracy of both the input data and the results.

    First is the providing information it comes up with that isn’t there. Certain models like the ones ChatGPT are trained on do this, but other LLM likes Gemini don’t. It will only respond with what you give it, nothing more. This leads to better accuracy. I have a friend who uses this to summarize legal proceedings for his business. (He’s not a lawyer so it’s not life and death. 😂)

    Second component is understanding what the AI indersgands when it views a photo and how accurate it is. This again has to do with how well the model is changed. When reviewing test results across multiple LLMs, one performed significantly better in this challenge, which means I used that one for the image progressing. I’m sure video will follow once better base models release, maybe even this year.

    Third is your existing P&Ls. If you have historical data of what flips you’ve done, it can be trained off that, and only that as you desired. You can also just tell it what percentages you want to hit.

    Fourth, you can adjust the prompt so it will tell you directly that it needs more information about certain things, and can suggest more investigation into areas of your flip to increase the confidency score of different parts in it’s rehab analysis. You can also request the confidence score by percentage if you want, and adjust it accordingly.

    The way to test accuracy in Machine Learning, is by training the data on most of the deals. The you take a few deals, only give it them input, in this case photos, what returns you want, loan details, etc, and compare your final results with what they got. It’s not a crystal ball and it’s not going to catch major shifts in the market. That’s a whole other level of having to monitor the news, which I think is much more of a challenge. 

    As for what it does, it’s nice to get a generally idea of what needs repair and areas to consider and what to look at. If you want an instance of this, happy to chat and see if it works for you. 

    0Reply
  • Chris Magistrado
    OP
    Investor · Member since 2022 · 78 posts · 42 votes
    1y
    Quote from @Michael Wyatt:

    Backflip has a decent analysis tool on their app. Not to the extent of uploading photos and explaining repairs (that's pretty cool) but running comps, estimating ARV, reasonable rehab budget, profitability, etc.



    Yeah the uploading photos part is probably the most exciting part about it!

    I'll checkout Backflip! Looks like they're trying to sell lending services through their app.

    0Reply
  • Carlos Collier
    Investor · Prescott Valley, AZ · Member since 2017 · 27 posts · 13 votes
    1y

    Hey Chris, this sound like an awesome addition to a tool set. I've been using Dealcheck.io to do my deal analysis, but would be interested in hearing more about what you used to build your tool, or even getting hands on with something similar of possible! :)

    0Reply
  • Chris Magistrado
    OP
    Investor · Member since 2022 · 78 posts · 42 votes
    1y
    Quote from @Carlos Collier:

    Hey Chris, this sound like an awesome addition to a tool set. I've been using Dealcheck.io to do my deal analysis, but would be interested in hearing more about what you used to build your tool, or even getting hands on with something similar of possible! :)


    I'll check out Dealcheck! And I'll send you a connection!

    0Reply
  • Mario NajarroPro Member
    Real Estate Agent · Stafford, VA · Member since 2016 · 6 posts · 1 vote
    1y

    Hi Chris, I just came across your post. Interesting enough, I've been trying to create a similar product, however, a little more robust. I'd love to bounce ideas off of eachother. 

    0Reply
Join the conversationCreate a free account to reply, vote on answers and follow this thread.
Log InJoin for Free