Why Model Performance Beats Features: Collov AI's Growth Playbook

0:00Today I'm speaking with Xiao Zhang, co-founder and chief executive officer at Colof AI. We apply those spatial AI models to streamline the design process, also to streamline the virtual staging processes. So that helps a lot of real estate agents and interior designers. And Xiao, how did you come to decide that the problem you wanted to tackle was exactly that one with AI of spatial design? Normally, a real estate agent, they need to run a combination of the furniture and then hire some photographers to take nice photos. So that process costs like several thousand bucks, also takes several days.

0:35With the help of the model, we can shrink the timing to be several seconds, and also we can shrink the cost from several thousand to like less than$1 per several images. So that is a huge increase of the efficiency.

0:57Hey, everybody. Thanks for listening. Today, I'm speaking with Zhao Zhang, co-founder and chief executive officer at Coloff AI, a fully integrated platform that is leveraging AI to provide a one-stop shop for users to design, shop for, project manage, and acquire funding for any residential remodeling product. How are you doing, Zhao? It's a pleasure to have you here. Yeah, cool, cool, cool. Very nice to join in this conversation with you, Anders. Yeah, so very nice to share my experience here. For sure, that's what has us most excited as well. So before we get into the deeper questions, Yao, tell us a little bit more about yourself.

1:35Where are you located? How did Colov AI come to be? That would be a nice way to put our audience in context. Hi, everyone. My name is Xiao Zhang. And so right now I'm based at San Francisco Bay Area. So the Silicon Valley. Yeah, so a little bit of my background. So I previously studied the applied physics PhD at Stanford University during the past like several years. So during that research period, so I was doing the AI for science, the interdisciplinary research, applying the AI models, especially some AI models, which is very skilled at understanding the space, those kind of spatial model to resolving some science or physics problems.

2:20And then after the graduation, so I'm thinking about how to better utilize the technology to apply them into resolving some real life problems. So for example, so the real estate and home renovation industry, so it really requires a good model understanding of the space and then do an automated design process. So in that case, I was applying those kind of AI models into resolving this problem in the industry, in the real estate and home renovation industry. So we apply those spatial AI models to streamline the design process, also to streamline the virtual staging processes. So that helps a lot of real estate agents and interior designers, so largely increasing their efficiencies.

3:05So right now we have over like say 10K real estate agents and also like numerous interior designers subscribing and using our services and also served more than like hundreds of enterprise clients. So right now we are at the Series A platform stage. We're happy to share some more details about our startup journey. Okay, very nice. Very nice. That is so cool. And Xiao, how did you come to decide that the problem you wanted to tackle was exactly that one with AI of spatial design? Like, did you analyze different ways to get into the market and working with AI and you prefer that one or like how did you come to decide that?

3:46Back to the point that when I was about to graduate, so I did like think about like how to better utilize the model, especially the spatial focus model, how to better utilize the capability of that AI model to resolve some real life problem. And then so the first one I thought about the real estate industry, and so for this industry, it really requires a lot of a nice understanding of the space, because no matter the indoor design, outdoor design, and also the furniture combination, all those kind of things really requires a very nice understanding of the space. And that is very much linked to my kind of expertise.

4:27So my AI model is more skilled at understanding the space. So that is a very good feat to applying the model into resolving the industry domain-specific problems. And also, secondly, for this industry, the market is very large. So the real asset industry has like over a trillion dollar market size. So in this industry, the efficiency is not that high enough. So people are using some traditional way of, say, like doing the interior design, doing the virtual staging or showcasing their properties. For example, when they're doing the virtual staging services or doing the staging services, normally a real estate agent, they need to run a combination of the furniture and then have people helping them moving into the property they want to sell.

5:17And also like the filter need to hire some photographers to take nice photos. So that process costs like several thousand bucks in the United States. And also like that process also takes several days. And then so right now with the help of the model, we can shrink the timing to be several seconds. And also we can shrink the cost from several thousand to like less than one dollar per several images. So that is a huge increase of the efficiency. So with those two points, the large market size and also the space that we can increase the efficiency. So combining those two, we think this is a nice industry.

5:56becomes a spatial model in two. Oh my God, that sounds so cool because when I hear you talk, Zhao, I only recall in my experience because I don't have a lot of what's per se expertise in spatial design and interior decoration, but I have used ChatGPT and sent it a picture of my living room and said, hey, could you maybe redecorate, remodel this in a certain style? And from what I'm imagining, Colab AI is going maybe through that route. Is that what you're saying that is efficient? It's making it more efficient for people to just take a picture and remodel whatever they want. Yeah. So basically, our product is like, we want to make that very streamlined, very easy to use.

6:38So the real estate agents and interior nurse agents need to take a photo and then upload into the website. And then choose the paper style and room type they want to change the photo into. And then just click the generate. And then several seconds later, you will see an essay looking images. And also, as Andrews just mentioned, using the chatGPT, those kind of large foundational model, they also launch very powerful models. However, for those models, they're kind of a generally applicable model into more scenarios. However, for the domain-specific version, for example, if you want to upload an image through chatGPT and then let it change to a specific style of design.

7:21So in a lot of cases, the room structure is not maintained really well. And also, the style you want to change might not be as you expected. So in that case, for our company, we really did a lot of innovation into our AI model architecture. So in that case, it has a really nice controllability during the generation. It will 100 % maintain the overall structure of the room. And also like it will make the furniture combination looks really, really real, looks very similar to a real photo. Yeah. So that's what we did. And then that's why we win the class. Oh, my God. That's so amazing. OK, so like I'm understanding that maybe OpenAI and those models that you find that aren't yours are more of a one size fits all per se.

8:10And it's just a picture that will work for everybody just a little bit. but what you're providing is a bit more tailored to the specifics of what each user is going to want. So, okay, that is something that sets you apart from all the AI noise and static that is within the market right now. And Yao, if I may ask, what do you believe is the secret to building a platform that handles design, shopping, project management, and even funding all in one place? Yes, one of the most important characteristics for building a startup and then organize everything, I think is kind of more to be persistent, like during this kind of startup journey.

8:50Yeah, for example, when I build a startup, this startup has been last for about like three to four years already. So there are a number of times that we have went into trouble. Sometimes the trouble might be we are close to running out of cash. And sometimes the trouble is the business model might not work so well. So there's an example. So back in the first year of my business startup, our business model is kind of largely different from what we are right now. So back then, our model is like our AI is not that powerful enough yet back in 2022. tool. And then, so what we did is like, we hired a team of contractor interior designers.

9:35They used our AI design tools to help them finish the interior design and then that will be presented to the end consumer. So back then, the AI is not powerful enough to automate the process so that we have a human in the loop process back then. And then with that kind of model, so we saw that kind of business model might work. So we try to put more like promotions, marketing, like investment. So that try to increase the revenue. However, then we find out that business model, the kind of the unit economy model is not kind of positive. So the more we invest in marketing, the more clients we get, the more we lose.

10:18So actually like for serving each clients, so we need to spend more, like more than more than we earn. So the reason is that back then, the more clients we have, the more interior vendors we need to have. The management costs kind of increase with our growth rate of the revenue. So the management costs increase. And also since that's a human in the loop process, so the more clients we get, so the quality of the service will kind of like varies since we will have a larger team of like interior designers. Sometimes with a small team, the quality can be guaranteed. However, with a larger team, sometimes the quality might not be always that good.

11:01So in that case, we found out like we met a problem that we run, we kind of burn more cash and then the more clients we get, we lose more money. So we are in the circumstance that we are about to run out of cash back then. So if we just try to give up, then everything kind of ends. However, we just kind of not giving up. So although we didn't find some good solutions back then, however, once we persist several more months, and then suddenly the chat GPT is launched. So everybody's attention has been drawn into the AI field. So we are successfully getting some more investment. And then also there is also a breakthrough in our model innovation.

11:51So in that case, our model can successfully automate the design process and then can serve numerous number of clients by automated process. So with those kind of help, so we kind of revamped the business model to an AI-centric automated process and also we got the funding so we can continue. And then these two years, we grew very nicely. and then we are in a variable state. So basically, so the persistence is really important. So sometimes you will never know that when you will meet some kind of problems, but if you kind of never give up and then there will be opportunity that you can size. Yeah, so that's my kind of experience.

12:35Of course, no, and that's beautiful. Like you're telling me, Xiao, that you were working with AI back in 2022. That's back when people didn't even know that ChatGPT was gonna exist. So I'm very glad that you've been able to like hang on the curtails of this whole AI wave. And now Kolob has been benefited thanks to that. And you've been able to grow even more. That's amazing. That's amazing. And talking about your consumers, Zhao, and how you mentioned that you had to maybe invest a bit more in marketing and in showing your product. How are you able to educate consumers about spatial design intelligence when most people have never heard of it?

13:10Or the max they've heard of is like me, just using ChatGPT for two pictures. Okay, yeah. So I think when we try to communicate with the consumers, we will not show you those kind of fancy words like the spatial design things or some spatial AI model. Yeah, because a lot of our clients, they are aged, like say, 40 to 60, like real essay agents. So they are not that kind of used to the fancy or kind of the buzzwords. So in that case, we really need to make our website very, very easy to understand. And also we want to make our process very, very easy to use. So that's also the reason that right now our product process is very streamlined.

13:56The customer just upload a photo and then click the generate, then you'll get the results. Very streamlined, very easy to use. So I think if your audience is very tech-centric, then you might make the world more fancy or make the process a little bit complicated. That's fine. However, if your customers are not so used to the tech field, in that case, try to make the process very streamlined and try to make it very easy to understand. That will really help the customer adoption rate. Okay. Okay. No, I totally understand. Because to the ordinary consumer, if you use too many big words, you'll lose them in the first tab that they open.

14:37And they'll be like, hey, I don't understand this. And then they'll go do something else. So I totally get that. I totally get that. And how about when it comes to the automation, how do you balance automation with the personal touch that homeowners want for their space? Which is what I believe is what you mentioned sets you apart because you're able to tailor what the AI is going to, for lack of a better expression, what the AI is going to spit out. The human touch in the process is also important. So for now, we have a very quick streamlined process of AI design, the customer out of the photo, and then after that, AI will generate the results in several seconds.

15:19However, after that, we will allow the customer to do a lot of edits based on using the help of the AI tool. They can do a lot of edits. There will be human touch after the initial generation. So that will let the customer to edit as they want. And then this kind of editing process is also built based on what I just mentioned, based on the experience back in 2022. So originally, we are providing the AI tool to help the designers to do the interior design more efficiently. And then right now, since we have a streamlined AI design process, so we will first show the AI design results in several seconds to the customer.

16:04However, after that, we will use a human-in-the-loop process using the AI tool to help the customer to do their edits, whatever they want. So that's kind of how we design the whole process. Okay. And Xiao, do you also help the customer? What's it called? Because from what I kind of read and what I might have understood correctly or wrongly, Do you also provide contractors and suppliers for the customer to recreate in live what the AI showed them with your company? So there are a lot of contractors, suppliers, real asset agents. So they are using the tool so that they will not only show a kind of a design proposal, virtual design proposal, but also after that, they will follow the design proposal and then put the furniture to the place that AI design showcased.

17:00So that's a process to convert from a virtual design to a kind of the real like um fringer combination or the real like uh real home based on the design's guidance yeah okay cool and how do you handle or is it very complex integrating all these contractors and suppliers to your platform and financing it into one seamless experience so originally there are not a lot of users like using that well so sometimes uh uh people are attracted to our website, they might subscribe. However, the next month they will turn. So they will leave the website. So that's always the case back in the first year when we launched the automated design process.

17:44And then we very much treasure the customer feedback. So we will talk to, say, five to 10 customers every day. So we will kind of summarize what's the problems they have, and they try to categorize that to several different aspects to improve our model, improve our products. And then we find out, for example, there is one major issue that they don't use really well, which is when we generate the AI design. So originally, our model is not like maintaining the room structure while doing the generation. So sometimes the window will change to a wall, the door disappears. So in that case, the customer can now use the well.

18:29Yeah. And then based on those kinds of customer feedbacks, we kind of iteratively improve our model and improve our products. So after several rounds of improvement of the model, the products, based on the customer feedbacks, you find out gradually the customer, more and more customers, they are used to our tools. And then we have a kind of a growth rate over month over month. And also, since we are continuously talking to 5 to 10 customers per day, we see that the customer satisfaction rate has been increased by a lot. So I think the trick is listen to your customer and then improve the model based on different customers' feedbacks and then try to make your model applicable to different customers' needs.

19:18And then after that, you will see the customers are satisfied and then different kinds of customers, they are utilizing your product well. Okay, cool. Wow. AI really is super interesting because at the beginning, I feel it just does what it wants. But having a team like yours behind it, then actually tailoring it to what the feedback of the customers is, it really provides a much more wholesome experience for every single person that uses the platform because it will give them a better, what would I say, like a better rendering of what they want? Yeah, for sure. Yeah. So here I actually have a quick point to add.

19:55So in this wave of Gen.AI startups, I think the model performance really is a key issue to determine whether you can win the customer or not. Yeah. So for example, since we are in the Silicon Valley, we are always talking with other Gen.AI startup founders. For example, one of the very popular startups called Cursor, the AI coding company. So they are recently growing the revenue really fast. So right now they are at about like 500 million annual recurring revenue. Just realized this in two years. So after they founded the startup, really kind of like a very high speed growth. So what they did is something similar.

20:37So originally, so their model is not that catered to the customer needs. And then after several rounds of improvements, they make one function really useful to use, really easy to use. They made the function like the engineers, they write half line of the code, and then they use their model to click the tab key on the keyboard. And then the AI model automatically completes the whole line of code. So after they make the model to be twinned, to be having a very stable performance of the auto-code completion feature, so they see a dramatic growth of the user adoption rate, and they reach a very high annual return revenue.

21:21So for us, we find a similar kind of phenomenon. So previously, our AI generation always have the room structure to change during the generation process. However, once we make our models architecture to be innovative and then once we make our generation to be stable, we see a very high speed of growth rate. And also we see a high conversion rate after that. So in this way of generating AI, I would say the AI model performance really is the key scene that matters to determine whether you can have a nice growth rate or not. Okay, no, I totally understand that. I totally understand that. And Yao, I have in the back of my head another question because I know you work a lot with AI, automation, and that is something that nowadays is being very assimilated to not needing that much of a people workforce or an actual human behind what is being done.

22:21And my question is, how do you know when you've got to just keep using AI or you have to actually expand your team and that you actually need somebody to come and maybe tutor the AI or do something else to make sure that what it's outputting is okay? Do you have a certain process for that? Yeah, yeah, for sure. Yeah, so like nowadays, since AI gets more and more powerful, so like there is a question like whether we want to have a really large and powerful team or we want to keep the team lean. So I think for our choice, it's like, so right now we keep the team relatively lean, but like it doesn't mean that we don't need like human or we don't need people to tutor the AI process.

23:04It doesn't mean like that. It doesn't work like that. So I think the trend of the team management or the team carrying process is kind of something like, so you need the top talents in your team. So who understand the AI, who knows how to tutor the AI to do either some functioning of the AI model or just doing some prompt engineering to make the model to work better. So I think right now we keep the team lean. However, we will have each of the members in the team to be very talented to understand both AI and the customer. So that once we want to launch a new AI feature, our team members, they will be using the AI tools with the help of the AI model.

23:54So they can tune the model to be catered to the customer's demand, customer's requirements. So that's how we manage the team in the area of AI. Okay, amazing, amazing. And okay, we're almost at time, but I would like just to ask one more question, and this is for the audience. I believe you already answered this one when you said the perseverance that needs to never alter. But if you had to give only one piece of advice to entrepreneurs who are building integrated AI platforms for the complex consumer decisions such as spatial design, what would it be? What would the advice you give anybody else that is at the start of your journey that you've already been on since 2022 and before?

24:44So I think right now, since people are doing attention to the AI models to say that whether the model is really powerful, the model has some unique capability, that really attracts a lot of attention. However, I think if you want to do a business or do a startup, I think it's always very important to listen to your customer and then to be a customer-centric startup. I think no matter if it's an AI startup or kind of other type of startup, listening to your customer is always the top one thing. So right now in the area of AI, people's attention are too much drawn into the AI model capability. However, you really need to cover the AI model to make that protocolized and then to make that very easy to use to your targeted audience.

25:34In that case, you can really win the customer. So no matter your technology is good or not, or no matter how good your technology is, you always need to pay really a lot of attention into the customer satisfaction, the customer service part. So to serve your customer well, then that's really the key thing. You can be in the customers. Yeah, that's kind of the advice I give to the tech founders who is willing to do some ASR. I love it. I love it. You heard it here first. And I believe this is something that everybody knows deep in their hearts, but it might be forgotten at times. And it's the importance of customer centricity because we really are working for the customers and trying to provide a product that they like.

26:18And if they like it, then the platform will rise with more users and more people. So I believe that is a very nice way of looking at it because you don't lose what is important in all this process. Because you might have a very good product, but if it doesn't satisfy the proper consumer need that you want to, then it might fail. So I'm very glad that's your perspective. And please, anybody who's hearing this and you're starting out in the tech industry space with AI, always remember, keep your customers at heart. You heard it here from Xiao first. And without further ado, we're going to say goodbye.

26:53Please make sure that if you have any spatial design needs and if you just want to go and check Colab AI, do you have any words, Xiao, that you would like to direct the audience that they could go and look at the product, maybe test it out? Like if you are interested in the spatial design or like if you are in the real estate industry doing like, say, the poverty listing or if you are real asset agents and also if you are like, say, interior designers, contractors, as long as you are in the industry and you are interested in the AI design process, you are very welcome to try out our product, which is kolo.ai.

27:31Search kolo.ai. You will log into our website and then experience the power of AI immediately. Yeah, thanks so much for listening. Thank you so much. And thank you, the audience, for listening. make sure to go check out kolov.ai. And don't forget to subscribe for this episode in the future of consumer marketing so that we can keep on hearing the best experiences from all the talent that comes on this show. So thank you so much, Shao, for coming on. We'll make sure to keep in contact. I hope the best keeps on going for kolov.ai. Yeah, no problem. My pleasure. That's a pleasure.

28:14Thank you.