Press enter or click to view image in full size
The combined rise of artificial intelligence, along with improved data capture and storage, is increasing our ability to automate more and more of the key functions within a company. If we believe that this trend of automation will continue to hold, it seems possible that one day we’ll be able to have a fully autonomous company: an AI managed company, supported by robotics, that performs all of the core activities of a company — i.e. read no humans employed.
What would a company where no humans work look like? What are the core activities & roles within a company, and how is automation affecting them? Approaching this systematically, I listed out these two dimensions: activities that a company performs (e.g. building a product, distributing a product), and roles that we commonly find in companies (Marketing, HR, Product). I found that this exercise provided an interesting framework for thinking about automation and enterprises more broadly.
This post shares some of the observations and thoughts I came up with.
Cookie-Cutter Companies
An interesting first principle is that companies are fairly consistent in terms of both the core activities they perform, and the departments/functions they have within them — regardless of industry.
At a very high level, I would say the core activities of a company are the following:
Core Activities
- Figure out what to make / what service to provide
- Build the product / provide the service
- Get the product to people
- Support the people using the product/service
- Support the company itself (e.g. make sure there’s money, make sure the people working there are happy, allocate the right tools / space needed to get things going, hire new people / get more money to grow)
This list seems fairly simple, but representative of what most companies do, and indicative of the required abilities of an AI to perform the activities humans currently perform within companies.
Core Departments / Functions
Let’s next consider the core departments / functions within a company (e.g. Customer Service, Marketing, HR etc). When you actually list the typical departments of a company you realize there aren’t that many unique functions, and that they recur across companies. I don’t know the historical aspects of each of these departments, but company org charts have probably looked largely the same for at least the last 50 years in terms of what departments are listed. Even across industries, the departments a company will have are largely the same. That stood out to me.
I’ve tried to list the core functions/departments that consistently appear across companies:
Press enter or click to view image in full size
Furthermore, I think that we can link the departments, to the core activities that they perform.
We can also consider these given departments, across different industries.
This visualization introduces interesting ways to evaluate the evolution of a company over time. For example, if your company specializes in e-commerce marketing automation, do you start going after other departments in the e-commerce space such as e-commerce logistics automation, or do you push down into other industries e.g. pharmaceutical marketing automation.
The Path to Automation: Role By Role, Activity By Activity
Certain roles within a company, and specifically certain aspects of those roles, are easier to automate than others, and so that’s where tech companies have started to dive in. People who have those job functions, will need less and less time to perform aspects of their jobs as a result of automation tools. The consequence being that they will have more time to focus on other high value aspects of their function in the company, or that the company will just overall require a smaller team to fulfill the goals it has in that particular function.
Marketing Automation
The area I am most well versed in is web/digital startups (e-commerce, apps etc.). In that area, I believe that over the past few years we have seen the first wave of automation, and that wave was particularly focused around marketing departments.
I’ve worked extensively on products for two companies that billed themselves as marketing automation companies. The tools are primarily targeted at publishers and e-commerce companies. These products would allow companies to automatically track how users engage on their websites, and then automatically analyze those users’ actions, and send them relevant marketing materials via email, and other channels: e.g. promotions, follow up actions etc. As you can imagine, before these tools existed, humans must have been spending a lot of time trying to send follow up emails — so the value proposition these tools have in terms of saving human labor is substantial.
The software some of these marketing automation companies offer are becoming so plug-and-play that you can pretty much drop a few lines of code on your site, and that enables their software to monitor users’ behavior, and automagically create and send them personal, relevant communications to further re-engage them with your product. Bluecore, Sailthru, HubSpot, Marketo, and RichRelevance are a few examples of applications in this space.
These platforms sell the notion that an algorithm is better than humans at monitoring and interpreting customer behavior, and deciding on the best action to take given that behavior. By studying large data sets of user behavior, their algorithms are better at predicting the best marketing strategy to take for each user. For example if a particular user abandons their cart, the algorithm knows that it’s ideal to reach out to them exactly a day later, with a 25% off coupon, on items they left in their cart.
These types of marketing decisions are particularly well suited to algorithms running on large data sets of historical user behavior. At scale it would be nearly impossible for humans to do what these marketing automation companies do: monitor the behavior of millions of users and send personalized marketing communications to each of them.
Of course, these types of automated responses to customer actions and behavior is only one aspect of what a person in the marketing role would do. There are other aspects to the marketing function that for now remain difficult to tackle with automation. For example, creating unique, fresh marketing communications for special events and partnerships, identifying relevant customer segments and how to communicate value propositions, branding & positioning, and much more. But the point is that slowly, algorithms & AI are starting to outperform us function by function, aspect by aspect, leaving an increasingly smaller domain where humans alone can outperform AI.
Press enter or click to view image in full size
I joked with one of these companies, that the product we should build for customers is just a dashboard that has a Green Light to indicate everything is alright, and the algorithm is doing its job, or a Red Light that would only appear if there was ever some issue with the software. The users should never do anything themselves, and just leave the algorithm to do its job. We didn’t go with that design, and for me it was mainly a joke about automation and what design in the automated world looks like ‘Stay calm, and don’t worry.’ Not the design I’d opt for — I am firmly in the camp of thought that humans should understand what the software is doing, the model it’s working with, and to have the ability to augment and adapt it.
Thankfully, there are currently many studies that suggest that for the time being humans in tandem with AI may often be able to outperform either humans alone, or AI alone at many categories of tasks: e.g. playing chess, and likely many of the types of things we do at our jobs . This TED talk provides an interesting overview on the conversation of human-computer cooperation.
In December I went to a dinner hosted at Young & Rubicam (Y & R), one of the largest marketing and communications agencies in the world. The dinner was attended by executives from several marketing automation companies, digital agencies, brands and startups. The topic of marketing automation seemed to be at the heart of the event, and it was interesting to see the split between people who were convinced that eventually algorithms would be able to perform large swathes of the marketing function better than people, and those that argued that humans would still outperform algorithms in choosing meaningful strategies to help companies communicate and reach customers. This topic definitely stirs people’s emotions — no one wants to feel like parts of their function are being automated away. But, the key is to take the time we free up through automation, and drive it towards new high-value areas/functions.
One of the comments I made at the dinner, was that the limitations of what can be automated by AI stems from how the AI itself operates. For example, most of the marketing automation platforms take the simplistic approach that the best strategy for today, is whatever strategy worked best yesterday. This idea that today is like yesterday is a simplistic model of the world. These algorithms are far from building a sophisticated model of how the world works. They do not take into account the various factors that influence change, and they are unable to reason about why today is different than yesterday. The strategy that will work best in the present, is not necessarily the strategy that worked best in the past. Moreover, if history does not repeat itself, and there really are Black Swan type events that have no historical precedent, then historical statistical analysis will inevitably fall short. Nate Silver’s book The Signal and the Noise is an excellent read that suggests that our predictive powers are strongest in domains where we have a good grasp of the model of the underlying forces that drive the system. We fare poorly in domains where we just look at historical data to predict the future without any understanding of what the model that generates that data is.
I would argue that humans are still some ways ahead of algorithms in modeling the world, simulating possible future worlds, and detecting patterns in our observations that may indicate where things are heading. Not to mention, ultimately, human beings are the ones who shape the world — the aggregate forces we observe influencing our environment are only the effect of the collective decisions of individuals.
Automating Office Management
Moving beyond marketing, where do we currently see AI & automation in the enterprise. In many ways, my friends’ company Managed by Q, which bills itself as “Smart office cleaning and management”, is an attempt to automate tasks commonly associated with Office Managers in the Operations & Logistics departments of companies. With Q, it’s easy to make sure that the office always gets cleaned on time, that there are readily available supplies, and that repairs happen in a timely manner. They make it easier for Office Managers to perform several aspects of their job, in turn freeing them up to do other high value activities. I give this as an additional example of a company that comes to mind that I see already actively automating aspects of a department.
Legal, Sales & Customer Service
The three departments that I believe will see a strong wave of automation in the next few years are legal, sales and customer service. I believe that these departments have many aspects of their jobs that are well suited for automation using big data analysis & artificial intelligence.
There is a confluence of several trends that I think combine nicely to enable automation to impact these functions. One of the main trends, is that more and more of the communications and knowledge necessary to perform these functions can be found in text format which is readily available for developers to mine in increasingly effective ways to train AI to perform the same functions (e.g. emails, text messages, chats, documents and much more). IBM’s Ross, the “Super Intelligent Attorney” based on their Watson AI Computing platform is one example of this type of automation. See my talk on the Future of Messaging for more thoughts around this.
I believe there are also aspects of these roles that have little variation from case to case, customer to customer. I’ll go into more detail in a future post on why I think these functions are prime candidates for the next wave of automation. As in the case of marketing, automation will be effective only at specific aspects of these roles. For example, in the case of customer service I believe AI can be effective at routing and resolving many customer problems, and could potentially communicate and solve customer problems as effectively as humans. For more high-touch companies, and for specific customer service cases, humans will still be very much required to provide the sort of service that customers expect to receive.
Product ( Design & Engineering ), Business Development
I’ll lastly touch on these departments, because often we think of these as activities that are particularly difficult to automate. What would automating business development look like? This brings up the interesting point of what exactly the business development role does (“What do you do?”).
One of the core aspects of business development is finding new partnerships and acquisition opportunities that support growth of the company, either through increased distribution or through enabling an enhanced product offering. Building partnerships is something that is considered a very high-touch human activity. But, it is conceivable that an algorithm could be very effective in at least the first stages of partnership building: identifying useful partnership opportunities, formulating reasons why it believes it’s a good partnership, and suggesting how the partnership could be structured based on historical precedents. This could be one starting point of what automating business development may look like. From there, Biz Dev AI could eventually try to actually reach out and try to set up the partnership, and maybe one day even negotiate the detailed terms of a partnership agreement.
I give this as an example, just to illustrate something which may seem obvious: that all roles will gradually experience some degrees of automation, and that more and more aspects of those roles which at first seem intractable for software today, will eventually be partially fulfilled and assisted by automation software.
I’ll leave the thought experiment of imagining an automated product department to you.
Human Shareholders in an AI Driven Company
I started this post by asking what would it look like to have a company that was fully autonomous: run by AI, and performing the activities we commonly consider a company to perform. This line of thinking raises interesting follow up questions such as: what is the true purpose of a company: to make things people want? to generate profits for its shareholders? to provide meaning for its employees? To do good in the world? Will people one day be shareholders in a company where no humans work?
These types of questions, and the general trend of automation, also touch on an idea that seems to be in the zeitgeist of late: the notion that automation is gradually reducing the number of jobs available in the economy. I’ll hold off on diving into that topic, but what I will suggest is that it is peculiar that corporations have looked much the same way for hundreds of years, and that we have yet to find better forms of organization. I do believe that we will see a shift in how companies are structured over time, and that automation will have a big impact on that. Looking forward to sharing a few more thoughts on the topic in the coming weeks! Stay tuned :)
Follow me on Twitter @nicolaerusan and on Medium @nicolaerusan for more thoughts on product design, product management, entrepreneurship & futurism.