Don’t Let AI Kill Your Next Deal
You knew it was going to happen. Everybody is using AI to answer questions, help them write better, and create ultra-realistic photos and videos. It was just a matter of time before someone asked AI what their home is worth or how much a home should be listed for.
Sounds like a fair question. But it’s a little like asking WebMD to diagnose a medical condition without ever seeing a doctor. You might get some helpful information, but the answer is only as good as the information available. AI can’t walk through your house, see the updates you’ve made, evaluate its condition, or understand what’s happening in your local market. Those details often make a big difference when determining value.
I decided to write about this because of something that happened recently during an appraisal. The assignment was for a home that was being purchased. The property was under contract, and all of the comps were indicating a value lower than the contract price. I asked the agent if they would like to provide me with the comps that they used in coming up with the list price.
Normally, when I ask this, an agent will provide me with a printout of their CMA or, at the very least, a list of the MLS numbers for the properties they considered. But this time I received a printout that was obviously from an AI model like ChatGPT.
Before I dig into this topic further, I will say that AI has its place and can provide helpful information on certain topics, and it can help us to perform some tasks by doing a lot of the heavy work. I think it can make us believe that it knows more than it does by feeling authoritative; however, we must realize that it does have limitations, and we need to recognize them.
To accurately price, or appraise, a property, local knowledge and expertise are needed. This knowledge is needed to choose the right data and to apply judgment to the information. To be blunt, AI does not have this type of ability. Not yet, anyway. It’s also important for the user of the AI model to know the right questions to ask.
Getting back to my recent experience, it turns out that the information the AI model collected to support its value suggestion was inaccurate. In addition, some of the “comps” were not even in the subject property’s competitive market area. All of this combined resulted in the property being priced too high. The appraisal came in lower than the contract, and while the property did sell, it was for much lower than the list and contract price.
Why Accurate Pricing Matters
There are numerous things that can happen if a property is not priced accurately. As you read in my example, the appraisal may come in lower than the contract, which can kill the deal unless it
can be negotiated.
While the buyer got a better deal, the seller lost the money they were expecting. This can destroy an agent’s credibility and affect their reputation.
AI can be used as a tool to aid in what we are doing but we need to be careful not to ask it to do things it’s not capable of. It can handle some of the tasks that are involved in pricing a property, but asking it difficult questions, especially without complete information, can result in inaccurate answers.
5 Pitfalls of Letting AI Price Your Listing
1) AI doesn’t “know” the market
In the past, I have written about what a competitive market area is and how it is used to choose comps. The information the agent gave me in the example I just described, which was provided by the AI model, did not recognize what the competitive market area for the subject property was.
One of the sales that AI provided, while being somewhat close in proximity, was not even in the same school system as the subject property. The market area was totally different, and the comp sold for a much higher price than what a similar home in the subject’s area would sell for. This resulted in the list price being skewed too high, which caused problems as I noted.
AI did not recognize that the school system or municipality mattered. These are characteristics that drive property values in most areas. If you don’t account for these differences, then a property can be priced inaccurately.
Searching for comparables is more than just looking at what nearby homes sold for. While proximity does have its place in comp selection, other factors must also be present, and that’s where properly defining the competitive market area comes into play.
In case you are wondering if these same rules apply to vacant land as well, it does. The process of finding comparables for land is the same as for houses.
2) The data AI relies on can be wrong
Just like the infamous Zillow Zestimate, AI uses information from public records, like square footage. Public records are famous for having incorrect square footage information on a house.
If you rely on public records for the property you are pricing and the comps, that is a double whammy. The price per square foot of a sale that is calculated by dividing the sale price by the square footage can be wrong, and then if you multiply that by the incorrect square footage of the subject property, that can result in a property value indication that has no resemblance to reality.
3) Price per square foot becomes meaningless
While it can be a reliable and relevant statistic to use in pricing a home in some situations, this only occurs when that number is accurate. If the square footage of the subject property and the comps is wrong, then the price per square foot is irrelevant.
As I stated above, this is also true for land. The amount of acreage and the price per acre must be accurate to nail down pricing.
On another note, the price per acre can be misinterpreted if there are variations in acreage between the subject and sales. Larger parcels typically sell for less per acre compared to smaller parcels, when everything else is equal. AI most likely does not recognize this and can apply a larger or smaller price per acre adjustment when estimating the value of land.
4) AI may confuse above-grade and below-grade areas
Speaking of square footage, one of the issues I saw with the square footage that AI used in my example was that it combined all of the living areas together. It added the heated and cooled areas in the basement to the above-grade area, which is a big no-no.
While all areas of a house are included in the final value, the basement area typically contributes a different amount than the main levels. If you combine them and apply a single price per square foot adjustment, the basement may be overvalued.
These inaccuracies will be found out if an appraisal is required for financing by the buyer. The appraiser will separate the basement area from the above-grade area, which can result in differences between the appraisal and contract price.
5) Macro market statistics may be different than local level numbers
It is important to recognize that macro market trends may be different than local level trends in certain cities, neighborhoods, or subdivisions. While the overall trend for the Birmingham, Alabama real estate market may show an average appreciation rate of 5 or 10%, this may not translate to the local level.
Certain cities, neighborhoods, or subdivisions may be appreciating or depreciating at a different rate. To apply a certain market conditions adjustment to a property regardless of where it is located may not provide an accurate picture of what is happening in the subject’s area.
It is important to analyze the market trends of the competitive market area of the subject in order to get a more accurate understanding of what is occurring at a local level. This will give us a better understanding of supply and demand as well as appreciation rates.
Another Story of AI Craziness
Believe it or not, I have another AI story to share. This one did not involve a real estate agent using AI. Instead, it involved the property owner.
Several years ago, I completed an appraisal on a property. Recently, a real estate agent contacted me about that same property. During our conversation, I learned that the homeowner had asked ChatGPT what the property might be worth.
Just as I mentioned earlier in this post, ChatGPT responded with a very confident answer. It even provided some general support for a value that was outrageously higher than the value I had previously appraised it for. The problem was that the estimate was not supported by market data, and the AI did not have detailed information about the property itself.
The reason I am sharing this story is simple. No matter what question you ask an AI model, it will try its best to give you an answer. To do that, it fills in the gaps with the information it has available. The response may sound convincing and well thought out, but that does not mean it is accurate.
I believe this is something real estate agents are going to encounter more often. Homeowners will ask an AI model what their home is worth and then form a preconceived opinion about value before ever speaking with an agent.
In some ways, this is the Zillow Zestimate 2.0.
The challenge is that it can make it harder to have a productive conversation about pricing a home for sale. The homeowner may become attached to the number the AI provided and may be reluctant to consider other information that points to a different value.
If an agent prices the property based solely on what the homeowner wants to hear, rather than what the market data supports, the home may sit on the market because it is not priced correctly.
As real estate professionals, we have a responsibility to educate the public about the limitations of AI tools. As I have explained throughout this post, AI can be helpful, but it is not a substitute for local market knowledge, property inspection, experience, and professional judgment.
Determining an accurate listing price takes much more than asking a one-sentence question to an AI model. It requires analyzing the property, understanding the local market, and interpreting the data correctly. That is still something that requires human expertise.
What Accurate Pricing Actually Requires
I’ve covered this in previous blog posts, but I’ll review it again here. I’ll try to link to as many previous blog posts as possible in case you want to read something more in-depth about the specific topic than what I am covering here.
Accurately pricing a listing starts with clearly defining the competitive market area. This is an area that goes beyond the subdivision or neighborhood and reaches to specific areas the buyer would consider if a home in the subject neighborhood was not available.
After you have the competitive market area defined, you must then filter the sales by bracketing the subject property characteristics. This includes items such as square footage, location, bedroom and bath count, age, and other property-specific features like amount of land, pools, barns, etc.
While it would be nice to have identical properties to the one we are pricing, this is not a perfect world, and this rarely happens. The sales comparables will have differences in these features, and it is important to understand this and adjust for any differences between the subject and sales.
Generally speaking, a comp that has a superior feature to the subject property requires a downward adjustment to make it more similar, and a property with an inferior feature requires, or lacks a feature the subject has, requires an upward adjustment.
These adjustments are made to bring the comps more in line with the subject and narrow the sale price down. This adjusted sale price then provides a better indication of what the subject property could sell for.
I know that agents may have a difficult time making adjustments to sales like appraisers do, but there are other options. Appraisers utilize a quantitative method to apply dollar adjustments to sales that reflect the contributory value of features or the lack thereof.
Agents who are not familiar with how to determine these adjustment amounts can use the qualitative method. Qualitative analysis involves using quality ratings based on how the sales compare to the subject property.
An example would go something like this: If a sale is better than the subject property in a certain feature, then rather than making a downward dollar adjustment, you would just add a negative (-) sign next to the feature. You would continue to do this for each feature that you have found has meaning to buyers.
After you have added either a positive or a negative sign next to significant features, you would then tally them up and note the net result. If you had 5 positive signs and 3 negative signs, the end result would be 2 positives. By doing this for each sale analyzed, you will get a better understanding of how similar each sale is to the subject property. You would then reconcile to the properties that were the most similar.
In addition to understanding why adjustments need to be made to the sales, it is also important to understand submarket trends where the property is located, not just trends for the overall city. This is where knowledge of the local area, as well as professional judgement, excels over AI models and algorithms.
Conclusion
There is no doubt that AI will continue to improve and expand in the years to come. There is a place for AI in the process of pricing a property, but only as a starting point for the analysis that is best done by a local real estate expert. AI is great for analyzing large data sets that can provide statistical analysis of sales that are located in the competitive market area of the subject. This analysis can help us understand trends that will contribute to a more accurate pricing strategy.
If I can answer any question you may have about comp selection or the pricing process, or if you need an appraisal, feel free to reach out, and as always, thanks for reading.
If you liked this post, subscribe by email (or RSS feed).
Speak Your Mind