Cost-Per-Observation: Decreasing Exponentially

The computer industry has gotten us accustomed to Moore’s Law. Even consumers who have never heard of Moore’s Law intuitively understand that something about technological progress gives them steadily increasing capabilities for a constant cost, or dramatic reductions in price for a constant capability. And, over a decade or two those trends combine to provide faster performance, larger capacity, smaller form factor, better resolution and more features in personal computing all the while managing to deliver a lower price point. Similar trends are observable across many different sensor types resulting in an exponentially decreasing Cost-Per-Observation (CPO). In this case, observation is being used in the scientific sense, which is “an act of recognizing and noting a fact or occurrence often involving measurement with instruments.” This post will explore examples of decreasing CPO, describe some drivers and multipliers of CPO, highlight higher-order operational goals that can be enabled by decreasing CPO and finally discuss some strategic considerations for businesses that operate in an increasingly observation rich environment.

Examples of Decreasing CPO

The cost of sequencing a human genome has fallen from barely conceivable in 1984, to $3 billion in the 1990s, to a just under $1000 (Not including analysis) in 2014. One day it will be commonplace and so cheap that an individual might just pay for it out-of-pocket for peace-of-mind. It turns out that genome sequencing is not a special case. Across many domains the cost-per-observation (CPO) is decreasing exponentially.

graph of decreasing gene sequencing costs

Wetterstrand KA. DNA Sequencing Costs: Data from the NHGRI Genome Sequencing Program (GSP) Available at: Accessed [June 2015].

Digital imaging of nearly all types is going through a similar curve, carried along by the Moore’s Law scale improvements in the underlying silicon substrate. Increased resolution and better capture rates arrive every year and often at a cheaper price. Or, holding resolution constant the price has been dropping exponentially over the last couple of decades. Digital cameras in every smart phone are so “good” despite being cheap that they are spawning add-ons that turn them into microscopes or infrared heat detectors. Cheap phone based cameras are changing communication through bringing pictures and faces to enhance text and audio. These combinations are making interpersonal communication more personal and add visual authenticity and immediacy to citizen journalist’s reports.

GPS receivers have decreased in cost from $150,000 for the first commercial receiver in 1983, to thousands per receiver in the early 1990s, to under $12 for an embeddable integrated chip in 2014. As they continue to decrease in cost, GPS chips will be embedded in more and more devices. Their time and location data are especially potent when combined with other sensor data in a mobile phone environment.

Accelerometers have gone from over $500 per device to less than $1 per device for MEMs chip based accelerometers that are just bigger than the eye of a sewing needle. At that size companies start producing multi-sensor embeddable devices for example combining a triaxial accelerometer, triaxial gyro, temperature sensors, and an on-board processor running a sensor fusion algorithm all in an embeddable package measuring 38 mm x 24 mm x 12 mm and weighing only 11.5 grams.

As shown in these examples, the cost to obtain and store data has decreased over the years. However, the cost to make use of that data after it has been obtained can still be very expensive.

Drivers and Multipliers of Decreasing CPO

Jumping Substrates: Whenever a sensor transitions to a silicon wafer substrate it connects to the Moore’s law improvements in silicon and gains significant economy of scale. This has been the key turning point and sustaining driver for decreasing CPO for many sensor technologies.

Miniaturization: Miniaturization of sensors has a multiplier impact on decreasing CPO. If only the cost decreased on sensors, usage would still be constrained because the size of the sensors would make them difficult to incorporate in product designs or in some cases would make the devices unusable. The combination of smaller, cheaper and better is transformative for product designers.

Wireless Connectivity: Wireless transceivers have undergone their own miniaturization and cost decreases. These decreases will prove to be a major multiplier of the decreasing CPO in sensors by removing the necessity of running wiring to each sensor. This greatly eases the complexity and cost of deploying sensors in fixed positions within built environments.

Reduced Power Requirements: Concurrent with these other multipliers the reduction in power requirements enables sensors to run off of batteries for extended periods of time and removes additional barriers of routing electrical power to each sensor in a home, warehouse or factory.

Decreasing CPO Enables

What does an exponentially decreasing Cost-Per-Observation enable? What can we afford to do today that wasn’t practical five years ago? What will be possible 5 years from now?

Improved Measurement Reliability: We can afford multiple and diverse measurements of the same phenomena. Multiple data sources of same phenomena create more confidence in our measurement reliability mitigating bias and error that arises from single source measurements.

More Timely Information: When target phenomena are highly variable and fast moving, more frequent measurement and analysis creates more accurate and timely facts and potentially actionable competitive advantage. The classic example of this is Low Latency HFT in the stock market. When the CPO is really low we can afford to sample much more frequently.

Multi-Mode Situational Awareness: Cheap sensors enable more complete coverage in physical space, time and measurement modality (vision, audio, vibration, movement, tilt, temperature, force). This kind of multi-mode coverage in a manufacturing environment will enable the next level of manufacturing automation, efficiency and quality.

Fine-Tuned Behavioral Analysis: Capturing measurements of customer behavior including signals of attention, product consideration, comparisons to alternatives and delays in purchase. This analysis also includes measuring the act of purchase as well as post-purchase actions such as returns, reviews and social media sharing. As consumers increasingly conduct more and more of these shopping steps online the decreasing CPO enables us to capture and analyze millions of these purchase behavior sequences.

Real-Time, Evidence-Based Automated Decisions: We strive to put all these converging observations together with intelligent analytical systems to deliver real-time automated decisions that are just right for that moment, for that person or machine in that specific environment. Decreasing CPO is a pre-requisite for achieving anything close to this vision.

Strategic Considerations of Decreasing CPO

How does exponentially decreasing CPO impact strategic investments and plans for the future of your business?

Dramatic Changes in the Possible & Affordable: Humans tend to think linearly. When we encounter something like the exponentially decreasing CPO we need to get intentional. Develop forecasts and anticipate how it will change assumptions and drivers in your own business. The decreasing CPO combined with miniaturization brings some measurement scenarios from impossible to possible, others change from too expensive to easily justified and yet others enter the trivially cheap zone. Thinking incrementally about changing capabilities will cause you to miss vital opportunities.

Internet-of-Things Finally Arriving: I have been hearing about the Internet-of-Things since 2002. Although one could imagine it then, cost made it impractical. Twelve years later in 2014 it is a very different situation. The Internet-of-Things is making the transition from vision to reality. The landscape will be dramatically different by 2020. What is unique and differentiating today will soon become table stakes. And ecosystems of distributed embedded sensors will flood the air with fresh possibilities for those willing to invest in making sense of it all.

Smartphones as Mobile Sensor Platform: The smartphone already has become the connecting point for mobile, personal sensors such as location and vision. It will continue to evolve into an increasingly powerful personal sensor node, managing some local computation and providing bandwidth and interface to a larger set of services available in the cloud. Understanding mobile phones as a sensor aggregation and gateway platform is critical to creating more intelligent consumer services.

Big Data Inevitable: Decreasing CPO is a primary driver of Big Data type challenges. Big Data requires the significant infrastructure to capture, store, protect and analyze data. Fortunately there are parallel exponential trends driving down the costs of bandwidth, data storage and computation. It is critical that strategic approaches include appropriate investments in supporting IT infrastructure. In some cases cloud computing with its flexible capacity will be the right answer, while for others privacy, security and absolute performance will require dedicated resources.

Develop/Acquire Data Science Capabilities: Magical analytic software packages that ensure you get what you need out of all this new data do not yet exist. Big Data by itself isn’t enough. Teams to make sense of all this data are starting to be labeled as “Data Science.” Data science is more than just statistics or machine learning. It often includes systems engineering with skillful aggregation and integration of data across diverse internal data silos, partner systems and public APIs. Of course all these cheap sensors we have been talking about will be providing a real-time picture of the world of things, people and events and require clever data fusion to make sense of it all. Data Science often includes Natural Language Processing to convert text into structured computable data giving insight into what people are saying about their world. Unlike the old days of stand-alone analysis that ended up in a nice presentation or a printed report, the outcome of data science is often “results” data that gets directly pipelined into real-time applications. That means that your Data Science team also needs domain experts and application programmers. Data Science teams are almost always cross-disciplinary in composition.

Closing Thoughts

I believe that exponentially decreasing CPO is driven by broad scientific and technical progress along with worldwide market competition. This progress on CPO reaches across multiple observational modes, is very robust and immune to the failure of individual companies or technological dead-ends. The savvy business leader will include this as one of the master trends that shape our global future and find ways to profit and grow from the influx of data available through decreasing Cost-Per-Observation.

Further Reading

This is an article in Nature by Erika Check Hayden discussing 3 vendor’s claiming to have broken the $1000 per genome sequencing barrier.

Wetterstrand KA. DNA Sequencing Costs: Data from the NHGRI Genome Sequencing Program (GSP) Accessed [September, 2014].

A TedXAustin lecture by Todd Humphreys about the soon arrival of GPS dots and the competing tensions between privacy and convenience that very precise and very small GPS receivers pose for us.

This academic working paper Statistical Basis for Predicting Technological Progress by B´ela Nagy, J. Doyne Farmer, Quan M. Bui, Jessika E Trancik makes the case that many technologies follow an exponential curve of decreasing cost per “unit” but that some technologies go “super exponential” Read this if you want to dig into forecasting methodology with greater rigor.

Article by Rob Spiegel on developments in sensors. Sensors are Everywhere: From Factory Floor to Your Skin, covers trends such as smaller, smarter, wireless and more robust.