Why it takes a long time to deploy AI in long-term care

This article is about introducing AI in long-term care facilities. Typical functionality is fall detection, prevention of bedsores, and sleep quality assessment. The article is of interest to three types of readers: care home operators, system integrators, and VCs.

1. Care home managers and their advisors.

One of the biggest problems for long-term care operators is staff shortage. AI running on smart sensors makes caregivers 20-30% more productive, so care homes are better off embracing the new technology:

• The advantage of today’s artificial intelligence (AI) is that it tells with great accuracy if a resident needs support or instant help. In the past, false alarms were an issue. Today, with AI, this is solved: Fall detection generating only one false alarm per three months? Check, done.
• The reliability of today’s AI means that caregivers can change the way they work. Routinely inspecting the well-being of residents at night is no longer required, which is where the jump in caregiver productivity comes in.
• In addition, the technology brings benefits to the residents themselves. When they fall, an alarm goes off within seconds. There is no need for fallen elderly patients to lay on the floor waiting hours for help.

However, suppliers to care homes can have different agendas, which care home managers and their advisors should be aware of. I will explain these agendas in this article.

2. System integrators, especially those specialized in security

For system integrators, the highly competitive security business is a “red ocean,” whereas the adjacent long-term care homes market form an emerging blue ocean. This, of course, is due to the recent introduction of AI into this market vertical. Thus, there is money to be made by system integrators open to service long-term care facilities. Read on.

3. Venture capital investors

The introduction of AI in long-term care should be of great interest to you:

• First, healthcare software is sold in a recurring license agreement. Companies with recurring license income are acquired for great prices.
• Second, caregivers need to be trained to use the AI, which makes it sticky and leads to great retention statistics.
• Third, there is a long-term shift in demographics, where the market for long-term care doubles until 2050. As a result of these market drivers, the long-term care software space is on fire, with for instance the recent acquisitions of Oslo-based Senso by Nordic Capital and San Francisco-based Ouva by Avasure (both in July 2024).

Still, VCs often remark that so much is happening that they find it challenging to pick the future winning technology type, such as wearables, radar, or optical sensors. This article answers that question for them.

The problems in care

The problems in care are overwhelming. The core problem is “double aging,” meaning:

• There is an increased demand for care due to the increase in the number of very old individuals and increased life expectancy, and the baby boomers are now starting to need care.
• Decreased supply of caregivers due to the declining birth rates over the past 50 to 70 years.

To quantify the problem, the Netherlands (where I am from) spends 4.1% of its GDP on long-term care. Norway and Sweden spend 3.5%. That is a lot. The average spend by developed countries on long-term care is 1.8% of the GDP. That is still a lot.

Moreover, the problem gets worse in the coming years. Globally, I count a Total Addressable Market (TAM) of 63 million care beds. This number will grow to 121 million beds in 2050. The TAM will double.

How things are done, today

Long-term care facilities are serviced by system integrators, who often have a long-term relationship with each other. The system integrator installed the telephone, Internet, and televisions. The system integrator is transparent in how she makes money. Through:

• Purchasing products on behalf of the care institute, to which the system integrator provides a guarantee, for which she adds 20% (and may get a kickback from the product manufacturer),
• Hourly wages for installation of these products,
• Maintenance contracts.

The current types of technology sold this way to long-term care facilities are underwhelming. A typical list of products installed in long-term care facilities:

1. Active alarm buttons, allowing a resident to press and ask for help – which is often felt as stigmatizing and which requires the resident to carry the alarm button;
2. Bed sensors such as infrared motion scanners or pressure-sensing mats, generating an alarm if a resident gets out of bed or if a cat jumps on the bed;
3. Acoustic monitoring, generating an alarm if a sound coming from a microphone exceeds a threshold;
4. Infrared motion scanners positioned next to the bed, generating an alarm if something close to the bed moves;
5. Wearable bracelets with gyroscope and accelerometer, offering the same type of alarms as radar;
6. Geofencing, where an area is delineated in a camera view. If the pixels change intensity, an alarm is sent.
7. Radar, where radio frequency waves combined with AI detect if someone falls fast – but which fails to recognize if a resident collapses slowly;

None of these sensors works perfectly by itself. Therefore, rule-based logic is added. For instance:

• Send a “fall detected” alarm when the motion sensor near the floor gives a signal, but the motion sensor at shoulder height no longer provides a signal.
• Send an “out-of-room” alarm when the bed sensor gives an out-of-bed alarm and the door sensor detects that the door has changed state.

These sensors require installation, maintenance, and support, as does the self-developed rule-based logic. This is a good source of income for the incumbent system integrators.

Three reasons why AI takes so long

First, the critical thing to notice is that all of these detectors above can be replaced by a single sensor: the camera. As I stated in the beginning of this article, modern artificial intelligence can articulate with excellent reliability what goes on in the room:

1. If the resident is on the floor and needs urgent help.
2. If the resident (who may have bedsores) should be rolled over to the other side,
3. A report on the quality of sleep of the previous night,
4. If the resident tries to get out of bed and needs assistance to prevent a fall,
5. If the resident’s time spent in the bathroom is too long, indicating a possible fall,

And so on. This can be detected through a decent camera with night vision and AI.

And that is the problem: The single sensor replaces all the other sensors. This means it replaces the income of the incumbent, traditional system integrator. This is why some of these traditional integrators are reluctant to embrace AI.

The situation is similar to often repeated stories, like Dell vs. Compaq, Netflix vs. Blockbuster, Charles Schwab vs. Merril Lynch, Kodak, and Polaroid vs. the digital camera. A newcomer, in this case, AI in care, eats away at the current business of existing system integrators.

As a second reason why AI takes so long, I bring up the famous MIT Professor of Robotics, Rodney Brooks. He states that software adoption is a trillion times faster than hardware. ChatGPT reached a hundred million users in two months. It will take another 30 years before all cars are electric, while the US Air Force still flies 60-year-old B52s.

Similarly, the AI software runs on cameras, which require physical cabling and assembly. This takes time. My company, Kepler Vision Technologies, develops this type of AI specifically for long-term care facilities; click here for more information.

Third, there is the master-apprentice relationship in care, which is how young caregivers learn their profession. The advantage is that they learn hands-on experience from experienced caregivers, ensuring they are well-prepared for real-world challenges. The drawback is that the master never learned to work with AI. Learning new skills, like working with AI, is therefore problematic.

This is why I see the introduction of AI in long-term care taking a long time. And if you have feedback on this article, send me a private message.