What is the state of AI in business today – and what do mid-market business leaders need to know about AI now?
This article is based on Daniel’s full presentation at NEXT, and on our extensive AI Opportunity Landscape research with enterprise and mid-market businesses.
This article is broken out into three parts:
- The Impending AI Transformation: The impact AI is having on business, how many companies are actually using it, and the challenges they face in building and adopting new AI applications.
- AI as a Competitive Advantage: How to gain a competitive edge with AI by developing data dominance and the necessary critical capabilities.
- An Outline for Getting Started: Recommendations on how to start thinking about AI initiatives and what one should expect about this type of project.
Read the full article below. The complete slide deck from Daniel’s presentation is available on Slideshare:
We begin our explanation of the near-term trends and ROI of AI business initiatives with an overview of the opportunity the technology offers to enterprises and some mid-level businesses:
The Impending AI Transformation
Machine learning and AI have undoubtedly had an impact on numerous businesses since the early 2010s. PwC states that AI could contribute up to $15.7 trillion to the global economy by 2030. In our six years of interviewing AI leaders and researchers, we have found that they are mostly unanimous in thinking that AI will change workflows, processes, and customer service in every industry.
Business leaders can use this impact to their advantage with thorough planning and research on the best possible applications to consider. This has always been the case with any technological paradigm shift, but AI is different in the exponential success it can offer to established enterprises.
Currently, the most powerful and fastest-growing companies are predicated on the development of AI. Google, Amazon, and Facebook are fueled by a constant stream of data and the neural networks built to understand it. While eCommerce, social media, and big tech are important areas in which AI is a clear differentiator, this will slowly become true for every industry.
The State of AI in Business
AI is at the center of attention in terms of new technology for business applications. News outlets, blogs, and Twitter give the appearance that all businesses are using some form of AI and that these industries are transforming right now. However this is largely untrue, and the use of AI in each industry is likely dominated by whichever companies are currently the largest.
This exaggeration creates the perception that smooth AI adoption is as easy as finding a vendor selling a product for a specific business problem. In truth, AI has not actually permeated every industry yet. AI developers are researching ways to more smoothly integrate their applications and offer value to each industry. That said, new applications will still require companies to enact numerous internal challenges to allow for actual AI deployment.
In our research and AI vendor short-listing work with enterprise clients, we advise clients to look out for two factors:
The first is AI talent, which should include numerous employees with master’s degrees or PhDs in AI or some related field.
The second is evidence of meaningful adoption of the developer’s product.
We have found that approximately 33% of companies that claim to sell AI applications even have the requisite AI talent on their research and development teams. Additionally, 33% of those companies with sufficient AI talent show any evidence of client companies adopting the solution. It is apparent that not every industry is seeing a sudden rise in the use of AI, because only a small fraction of developers can even show evidence of a successful sale.
There are numerous reasons why the majority of industries have not yet been transformed by AI. Some involve the difficulties in adapting an enterprise’s tech stack to handle a new application. Others are based solely in ignorance and can be remedied with a detailed but non-technical education about AI. Some industries could benefit from any number of AI applications, but their business need may be too specific to create sufficient demand for it.
AI is not IT:
Iteration required, Data access required, cross-functional teams. A change in culture, workflows, and teams is necessary to enable it. In IT more or less we’re going to calibrate something, plug something in, and it’ll work. With AI we have to do science. We have to make a guess as to what kind of data will work for our use-case.
Can’t just hard code something that will work. Could take months of iteration. Revamp data assets and infrastructure. Figure out what aspects of the data might be valuable. May take up to 18 months to try and see what ended up working or not. Hard to say where we’ll arrive with certainty. Most companies don’t have the stomach to endure that kind of R&D. Any unceartainty regarding ROI is painful.
Adopting and integrating an AI application is not as easy as starting a new IT project. A company can generally expect a new windows application to work immediately in their office. Conversely, AI applications require access to enterprise data and cross-functional teams to experiment with that data. The company’s internal teams would need to spend time on data science experiments in order to make sure the application will work correctly when officially deployed.
In order to enable this type of technology, companies will need to change their team configurations, workflows, and business culture. This is because leadership will need to understand that short-term ROI from an AI application does not usually come in the form of money. Instead, AI research and development can benefit companies by preparing their tech stack for later applications and making their enterprise data more accessible to those who need it.
An AI initiative may require multiple iterations on the same application, thorough research, and ample data to work with. This may take a company months or a whole year to determine whether the initiative was successful. Larger established enterprises may have the funds to handle this level of research and development, but convincing them to take the associated risks may prove difficult.
Many business leaders and decision-makers in the enterprise do not know what AI does, how it works, nor where they could apply it. They may have minimal knowledge about where it is being used in their sector, but they are mostly behind the times.
While getting them to agree to move forward with AI is a formidable challenge, this ignorance can also lead business leaders to show enthusiasm for applications that appear to work but do not solve an important problem. When this happens, a company could choose to adopt a very basic AI application that costs more money than it would ever save.
Current AI use-cases for the enterprise are new and thus not guaranteed to simply work for every company in the same sector. Most AI deployments are experimental (despite the expressed confidence of the vendor), and there are only a handful of use-cases across any given sector that seem to clearly be replicable.
This means that more specific AI-use cases are more likely to have problems or require significant work from the client to make it work for them. More use-cases will likely become more reliable as startups begin to run off their revenue – as opposed to purely living off of venture capital.
AI as a Competitive Advantage
The competitive advantage of leveraging AI does not reside within any individual AI application. This includes chatbots and products labeled, “Cognitive RPA,” which likely does not actually use AI. These types of solutions are not truly transformative to the businesses they purportedly help. Instead, companies gain an advantage from AI through developing critical capabilities within their teams.
The critical capabilities companies that want to leverage AI should build are skills, culture, and knowledge. If business leadership, subject matter experts, and data scientists can develop an understanding of AI and where it can be used, they will be better prepared to adopt new solutions in the future.
Additionally, companies should understand exactly what type of data they have and how it can be used to improve current AI initiatives. Finally, companies need to accept a cultural shift towards in-house experimentation and research. These capabilities are critical to moving into the future alongside AI.
An understanding of one’s enterprise data and how it is structured is necessary in order to prepare an AI application to work within a given business.
Enterprise data often needs to be restructured and stored differently in order to be accessible and understandable to machine learning models. This may require some trial and error, but a given company’s infrastructure will benefit from this experimentation by the time the team gets the AI initiative working. In the future, the company will now be significantly less concerned about whether their application can properly access the requisite data.
Contextual AI Knowledge in Leadership
Business leaders do not need to learn how to code for their AI solutions, but they certainly need contextual knowledge about how the technology applies to their business and what benefits they should be looking for. If a business leader can tell their tech lead exactly what they would like to do with AI and perhaps how, they may have already had some degree of this knowledge.
However, this will mostly come from strategizing and experimenting with new applications over time. This makes it important to keep business leaders informed about what the AI teams are experimenting with and what their results may imply for the future.
Data Science Talent and Cross-Functional AI Teams
Hiring data science talent is an important goal in itself, but the new workers will pose an even greater initial benefit to businesses. They will be able to work with experts from the business area the application is made for, and collaborate on how to best experiment with the data available.
After a few years, data scientists may become experts themselves, which would give them contextual knowledge on what data is required to make an AI application more effective. For example, if a bank was building a fraud detection application, this data scientist may understand which features of the bank’s historical data are the best indicators of fraud.
Conversely, subject matter experts will gradually understand what is important to the data scientists as well. They may see an exemplary instance of fraud and forward that information to the data science team unprompted. This makes for a faster and more knowledgeable scientific workflow for data experiments that will likely lead to more effective AI solutions.
Data Dominance is likely the most lucrative competitive advantage to leveraging AI in the enterprise. Establishing data dominance involves a cyclical process of offering a better product based on AI, gleaning data from the use of that product, and using it to improve the product and gain more customers. This creates a feedback loop where more and more customers come to the company with their money and data, leaving less for competitors. The following is a list of steps for how a large company typically comes into data dominance:
- Collect a large amount of data with an AI product or process. At the beginning, this is sometimes done at a loss.
- Use that data to create a better product or process.
- Scale that product or process to more users or customers because it has become better than competitors.
- Intake data from new users or customers.
- Repeat until nobody can catch up.
Daniel has written an entire article on Data Dominance – it may be worth reading in full for leaders interested in enhancing their competitive position.
Tech Giants and Data Dominance:
Data dominance is how Amazon and Google have risen to the top of their respective fields. In Amazon’s case, they dominated the eCommerce market by operating at a loss for nearly two decades while gathering product recommendation data. This growing store of data, or proprietary data plume, is what can make a company exponentially more powerful than its competitors over time.
The same idea can apply to other businesses as well.
HVAC Company Example:
Consider large HVAC companies as an example. They service large enterprise buildings and build the necessary equipment and piping for eventual repairs. If one of these companies decides they want to try and dominate the HVAC industry, they may try to come up with new ways of gathering important data with which to base their next products on.
A subject matter expert could get the idea to implement internet-of-things (IoT) sensors on key pipes and equipment so that they can detect differences in vibrations moving through them.
This may allow them to find new ways of saving on waste and energy or if the equipment is being used optimally. The company can then take all of the resulting data and use it to train an ML algorithm to find even better ways of saving. This HVAC company may be on the road to data dominance should this idea prove both unique in the field and lucrative.
Getting Started with AI
Businesses should expect adoption of AI applications to be challenging at first. Despite what AI startup press releases say, integrating new AI technology into a business’ tech stack can require a significant trial and error stage of development. That said, this process can be made easier by first developing critical capabilities and educating leadership about how the application should work and what to expect.
It is easy to read a press release and take another company’s word regarding their relatively smooth integration process. This can lead business leaders to move forward with “toy” applications that are not designated to solve a business problem but may yield a functioning version faster than a pragmatically chosen one.
Adopt or Wait?
Most big businesses will likely start the AI initiatives as innovators or early adopters. Mid-level and smaller businesses will usually wait until their desired use-case is more clear and reliable. This is because established enterprises have sizable research and development budgets and talent to adopt the more nascent use-cases.
It may not be optimal smaller businesses to invest in an AI application right away, because they may not be able to afford the data infrastructure overhaul nor the requisite AI talent that comes along with it. It is most likely best to get involved in developing critical capabilities and educating oneself on AI, while holding off on adoption until the time is right.
Next Steps for Business Leaders
- Educate Leadership on what AI can do broadly, use-cases in their sector and adjacent sectors, and give them a realistic understanding of the requirements of adoption.
- Talk to and Study Adopters regarding their success with their applications. This can mostly be done online but is also a good networking opportunity at business events. Gain an understanding of how long it took them to develop and roll out their application.
- Inform Your Plans with consultants, data scientists, and stakeholders. Do not bring in consultants until leadership has a better understanding of how enterprise works, because they may try and sell an application that does not work for the business.
- Decide on Next Steps such as talent and first AI initiatives. Determine the most necessary kinds of talent and what business problems are the most important to solve first.