Getting Deep on AI — Buy, Build or Acquire
STRATEGIC SOURCING CONSIDERATIONS
The total addressable market opportunity (TAM) for AI driven services in the enterprise sector is forecast to be worth $52bn BY 2021 (IDC — Worldwide Spending on Cognitive and AI Systems). Regardless of the absolute market value that is reached the general direction and magnitude of opportunity is very clear.
From my exploits with Pimloc it’s evident that AI driven services can already provide significant operational efficiencies — initially to augment existing processes and workflows; and in some cases, to replace them but I am also now seeing AI solutions being used to create new businesses and business lines that were not previously possible.
There seems to be little doubt that AI is driving a market inflection that will create vast value and fundamentally change the way we work, live and interact. It will create new businesses and disrupt old ones; those that put the right capabilities in place now will yield major dividends into the future as their systems learn and grow over time.
But how should businesses approach their strategy for AI sourcing?
A QUICK NOTE ON LANGUAGE
AI is the umbrella category that everyone recognises, it’s taken on a very general meaning and is very often over-hyped and over-promised. Literature and film have romanticised and warned of its capabilities since man started to write stories.
In many cases, what started out as an ambition to build general AI (HAL 9000, Twiki and Her) has become a sea of chatbots and recommendation engines with only a few specialist companies pursuing the meta-challenge of solving ‘General AI’.
Machine Learning is a subset of AI that is starting to get more awareness outside of specialist circles, the ability of a machine to learn based on a specific range of inputs (sound, images, numbers). It is one of the ways you can create narrow AI solutions (ie. checking for a specific step on a production line).
Deep Learning is a subset of Machine Learning that is used to build very deep mathematical models that are capable of representing more complex real-world challenges; i.e. being able to tell the difference between all animals. Pimloc uses Deep Learning to recognise and classify a broad range of content within images and video for a range of real-world business applications — this is a ‘type’ of AI but we are not trying build an ‘AI’.
Diagram: Example of a Convolutional Neural Network used for training Deep Learning systems
With the right team and access to relevant training data deep learning can be very effective across a broad range of applications that deal with the input, analysis and tracking of data (an area that is critical to the future success of most modern businesses).
Deep Learning, or a derivative of it, is likely to be the area that adds most practical value to the business world in the next 5–10 years — companies with lots of rich proprietary data (or an ability to synthetically generate it) are in a strong position to disrupt their competitors and build future advantage.
UNDERLYING MARKET CONSTRAINT
The major constraint on businesses wanting to develop their own solutions is access to top deep-learning talent — some estimates put the total global talent pool at 20k people (MikeDrew, OdgersBerndtson) with the majority recruited directly into the large global tech players (Google, Facebook, Baidu).
Although research has been happening in the area for the last 50 years the field of Deep Learning only really gained acceptance in 2012 with the introduction of accessible GPU processing and programming tools, access to large data sets and the development of deep learning network frameworks. Because of this the most experienced teams in the market have up to 6 year’s hands-on development experience with most being more recent graduates (the talent pool has been growing year-on-year with the popularity of the field).
This scarcity has driven up salaries and in turn created a very constrained market for talent that is having a major impact on the market provision and development of AI services.
The situation is manageable for the leading global tech firms (large salaries, scope to publish research and visionary projects) and a few specialist start-ups who can also offer pioneering/visionary projects with equity based rewards — the largest challenge I see is for the majority of more mature businesses who would greatly benefit from having internal AI capabilities but cannot currently attract and retain the right technical talent to build it.
TWO DISTINCT MARKET SEGMENTS
From my last 18 months’ endeavours with Pimloc speaking to businesses across a wide range of sectors it seems that most companies fit into two main categories:
1. The Enlightened: Companies who have already tried to create new AI driven services and know the challenges
2. The Explorers: Companies who have not yet tried to create AI driven services but believe they can
These are both made up from similar companies it’s just they’re on different parts of the enlightenment journey. The challenge for a pioneering start-up selling business services in this space is that every meeting is likely to yield a positive response based on the capability of the ‘technology’ to improve a business’s processes/products and services… but not necessarily any direct projects. Many of the businesses have not yet fully considered their options and are not yet ready to buy, the process takes time.
Many of these early stage conversations involve education, explanation and demonstration of what is possible — whilst valuable, these discussions can result in handing over business ideas that the company then (rightly) is forced to review against their other sourcing options. Whether to Build the solution themselves or to Buy (it’s usually far too early in the process for them to consider Acquisition). Many businesses ‘appear’ to be in the fortunate position to try out a few market solutions whilst trying to build their own capability alongside.
It appears that there are lots of options but the reality is not that simple.
THE ENLIGHTENED
Those that have already tried to build their own solutions
This group of companies are the ones who instantly see the value in our Pimloc proposition. They fully understand the constraints of Building their own solutions and have tried most of the generic cloud based solutions already — as a result they are better informed about what they need:
- Many have tried to build their own Deep Learning teams and know first-hand how hard it is to recruit, manage and retain a specialist deep learning group (especially straight out of University). They also know how difficult it is to design, create and train good deep learning solutions
- Many who have used leading AI cloud services to get up and running are now worrying about the strategic risks of building core business capabilities on-top of 3rd party APIs from Google and Amazon (and the associated data security risks)
- Many who have used online APIs know first-hand the limitations of existing SaaS services (especially in the image recognition and classification domain) and the amount of effort required to build usable solutions on-top of them
- Many who used pay-as-you-go Cloud AI services to get their businesses started (as they seemed convenient) are now being crippled as their costs have scaled over time with their success
- Many who are dealing with big data (video streaming, IOT) now fully understand the bottle-necks and costs involved in upload and GPU processing in the cloud and are looking at alternative platforms with more edge based processing and hybrid local/cloud solutions
These companies know enough to understand how deep learning services can provide them with real strategic advantage in the future but unless they have unique access to top talent they also now know that their best options may be to Buy or Acquire services they can tailor and own. They have a much better handle on where the value lies in their existing business data and what they need to start capturing for the future.
I’m now coming into contact with much higher proportion of these companies each week, a major shift from when I started. It’s taken some of them over 12 months to go through their own learning cycle — which whilst valuable in its own right could have been time spent developing solutions that would have already been delivering to their top and bottom line.
Education rates are increasing with business buyers becoming more informed about what they need and where to get it; I think we’re getting close to a tipping point where AI based business services are going to take-off in the broader market.
THE EXPLORERS
Those who have not yet tried to build their own solutions
This group have a general understanding of what is possible but still believe everything they need can be easily put together through open source libraries and APIs (ie. we can build it ourselves). They are usually just starting the journey of discovery and will create some proof-of-concept work with their existing development teams, try out online APIs and may even try to recruit their own specialist teams. From discussions with these businesses these efforts have proved great for end-of-year accounts statements and marketing but they have struggled to get innovations directly into commercial product and services.
This learning cycle is certainly part of the innovation process and is valuable, it creates internal business capability and educates teams on the complexities of what’s involved. They get to smoke test a range of new solutions and can start to work through the performance metrics that matter for their commercial products or services.
All of this means they are in a better position to Buy or Acquire if they later work out that the strategy of Building is too slow and not delivering — but they lose time and money (12+ months)
Many of the companies I met 12 months ago that were in this category are now coming back to the market to Buy solutions they can integrate directly to their commercial workflows, processes and product. They now understand the strategic value they have intheir data and the specialist value partners have in their expertise and platforms — they recognise that both are needed to create real product and service innovation.
WHAT TO BUY?
When it comes to Buying deep learning services, there are a range of choices that can be made depending on your requirements for IP ownership/control and the uniqueness of your needs.
Kicking-off some internal trials with open source solutions and online APIs is a great way of getting started but you may want to do this in order to work through your more detailed business requirements; it can inform your strategy but you may need to be careful that over time it doesn’t become it.
A businesses requirement for IP ownership and control alongside the uniqueness of their required solutions will determine the best Buy options for them. If you’re building strategic advantage for your business you may want to own the IP that is created over time or at the very least protect it from competitors.
The option to Acquire Deep Learning businesses with these capabilities is not open to all sizes of company and usually requires a period of Buying their solutions first. To date it’s been the large technology players who have been driving acquisitions (and so called ‘acquihires’) but I think we will start to see the next waves of M&A activity happening with more vertically integrated specialist businesses across non-tech sectors who realise the complications in building their own teams and platforms.
The journey is still in very early stages for most companies. Many traditional businesses are still struggling with their overall digital transformation and are already layering on a need for AI. It’s not a linear progression but both areas require a more informed and holistic approach to business data management and a strategic approach to sourcing new capabilities and competences.
If you want to have a conversation about what is possible or to discuss your content intelligence needs please drop me a note and I’ll follow-up directly.
Simon, simon@pimloc.com,
This story is published in The Startup, Medium’s largest entrepreneurship publication followed by +368,675 people.
Subscribe to receive our top stories here.
