More than $1T is spent annually on cleaning and maintenance. These markets have been relatively unchanged for many years, deploying only small incremental technology changes.
However, there is a seismic shift about to happen in the market. The combination of software, AI, and robotics will dramatically disrupt the market in ways that are difficult to anticipate. Self-driving cars, generative AI, large language models, and AI image/video understanding are all improving at a breathtaking pace. These rapid improvements are about to impact cleaning & maintenance in ways that most people miss. Let me explain how I see this shaping over the next few years.
Cleaning & Maintenance Are Technology Problems
From first principles, cleaning & maintenance are mostly hard technology problems. This is something few people believe! However, if you look deeply at the fundamentals of the problem, you can see a pyramid of technical challenges:
To do cleaning and maintenance well, you need to solve these problems generally from the bottom of the pyramid up to the top.
Understand the Property's Current State
People managing a property and service providers do a lot of labor to figure out what is physically at a property.
How many rooms does it have? What is the roof/floor made out of? When were they last serviced? What is the make/model of the appliance there? How dirty are things?
These kinds of questions are currently not tracked at all by most people in any kind of software tool. When new people arrive at a property, they must relearn all this.
Smart AI software tools and sensors will improve this process. Creating "digital twins" of properties allows people to visualize and track each property's current state in a more intuitive and "gamified" way. This is like implementing a "medical record" of the home, tracking what was there and what was done. The medical record analogy is a good one. Many healthcare professionals resisted moving to electronic medical records, but by making the move, you unlocked giant levers for data mining and improving care. In the same way that the medical field benefited from electronic medical records, property managers can benefit from a digital record of the property.
Extract the Client's Ideal State
If you gave 2 people the same physical property, each would want things done differently. In fact, many people (homeowners, renters, property managers) don't actually know what they want. To get this information, you need to ask intelligent contextual questions to help them discover & articulate what they want. The process to get this from customers should not be having them fill out a giant questionnaire. Such a form couldn't cover all of the scenarios anyway.
Instead, you need to combine information about the specific property, what others like doing, and what else the client has done to make suggestions to the client. Then you collect feedback from the person to update your model of what they actually want. This process is effectively reinforcement learning, and a combination of AI and traditional software are both valuable and necessary here.
Extracting the client's ideal state requires a well-crafted design process to succeed. Leveraging learnings from the mobile gaming community is likely necessary. People don't like filling out forms but happily fill out things for gaming.
Diagnose How to do the Work
Diagnosing something is effectively trying to figure out how to get from the current to the ideal state. Diagnostics is always complex. There are many possible environments and situations. Most people use limited experience and rough heuristics to diagnose issues. But software is getting much better at this fast.
AI-based medical diagnoses are in real-world use now at places like the Mayo Clinic, which has a rules-based AI engine for accurate disease diagnosis. Deploying a similar AI model, but trained for property diagnoses, is technically possible today, although challenging! The main issue is the need for more data about current and ideal states.
Coordinate Materials, Workers, & Info
Coordinating resources with software to complete cleaning & maintenance was not really possible before smartphones. Now it's technically possible but challenging. In practice, the real world is always messy. You need a ton of features to handle the enormous range of potential scenarios that can occur.
Not everything can be deterministic, either! Many things depend on ranges of probabilities to make decisions. For example, when should you notify the client when a cleaner is having an issue and may not make it? Such a simple question varies on the context. If someone is going to be able to come, and you need a minute to check, you should wait a minute. What is the probability that someone else can help? What is the timeline in which we will know more? What are the client's preferences around this type of thing? What kind of client is this? You end up with thousands of potential options for such a simple question. Almost every question relating to cleaning and maintenance has such hidden complexity.
Administrating cleaning and maintenance involves a lot of work with scheduling, communication, accounts payable, issue tracking, inspections, quality control, and payments. These are the most "traditional" software problems of all the problems listed. There are maybe a dozen administrative tasks, each of which has horizontal software providers. Someone trying to solve the vertical stack of cleaning and maintenance must do all these things well. These represent a reasonably known set of problems to work through, but it takes years of work to get it done right!
Do the Physical Work
To do physical labor, right now, humans are the only option. But robots are becoming more capable and will eventually do most labor. I'll discuss this more below.
How Robots Change This Market
Today, humans do most of the physical work in cleaning & maintenance. This will change with robots, which do not need to be perfect to provide value. Here are the stages I see of autonomous cleaning & maintenance:
Phase 0: Little-To-No software
This is the state of the industry today. Software is a minor piece of cleaning & maintenance.
Phase 1: Software & AI to Superpower Humans
This is the phase in which a few companies operate today: making humans smarter with software and AI. This phase will also see a wider sensor deployment to proactively and remotely identify issues.
As we build tools to help coordinate & improve human labor, we as a society are also making progress in robotic labor. Even if we magically had capable robots tomorrow, for them to be successful, we would need things like:
- Maps of the customer's properties
- A list of what the client wants
- Instructions on how to deliver it
- Administrative and coordination tasks
In other words, our progress to supercharge humans helps us progress toward a future involving robots. There is a lot of overlap. This allows us to provide value to customers and progress toward applying robotics to cleaning and maintenance without high-risk robotic work.
Phase 2: Human-Robot Hybrid Operations
The next phase is "Human-Robot Hybrid Operations". This phase is where a human is on-site with one or more robots. Robots can focus on what they are best at, allowing humans to handle the many edge cases that can occur. Having robots help humans is a dramatically easier problem to solve than "robot-only" operations.
A Human-robot hybrid model does not work for cars because you need an expensive human and an expensive robot to do the same task. But it is a great model for cleaning & maintenance! Even if the robot can only do a few tasks, it is still freeing up human time and providing value.
This phase will transform this industry because:
- Here we begin the process of increasing quality and reducing cost simultaneously. This has never been possible in cleaning & maintenance.
- The winner of Phase 2 is likely to be the dominant player in Phase 3 and beyond.
Robotic labor may start as early as 2025, and I expect Phase 3 to last at least 20 years.
Phase 3: Remote Pilot Robots
The next phase of autonomous cleaning & maintenance will be when robots are mostly autonomous with a remote pilot. This phase will occur when the robot can physically get to the property and do all the needed tasks but can't always do it right.
Robots don't need to do everything autonomously. They can rely on a remote pilot for tasks or edge cases. Autonomous service robots can handle latency safely in many situations because stopping for a few seconds is okay when cleaning a home. This is different than the self-driving car market, as cars need to act in milliseconds to avoid accidents.
A single remote pilot could assist customers globally from a single call center at this phase. The economic difference between fully autonomous and remote-piloted robots is likely to be small and diminishing over time, even if the robot needs a lot of supervision.
Phase 4: Fully Autonomous Robots
In this phase, robots will do almost all the physical labor. It's hard to say whether the physical form here is something like Rosie the Robot or a fleet of robots. No matter the form, the intelligence required is close to AGI and will take a while.
But we can make some reasonable predictions about these robots today. They are likely to be expensive at the beginning. Most customers will use them for a limited time each day. A customer using them 2 hours a day would still have 22 unused hours of operation. So it's most likely that most customers will not own these. Rather, they will use a shared service that manages and maintains the robots and provides them in an on-demand and scheduled fashion.
Summary
We intentionally designed TIDY to win big when this seismic shift occurs. The shift may happen much more suddenly than most people expect. For example, Tesla's target release date for its Teslabot is 2027. This could trigger the robotic change. We think the right focus for today is on applying software and AI to cleaning & maintenance. This helps us progress towards a robotic future with low risk and provide customers value.
When this robotics shift does occur, things will progress quickly. 5 years after the launch of robotics in the home, the big winners and losers may already be settled. Google was founded less than 5 years after the open web (Netscape). This is why making progress here is essential. The rewards are enormous, as this change will cause this $1T industry to consolidate into a few dominant players. There is likely one or more trillion-dollar companies made here.
More importantly, this change will dramatically improve society. By making high-quality cleaning & maintenance available to everyone, the world will be better and one step closer to a post-scarcity society.
