The key to predicting AI’s impact on your career is understanding the economics behind your role. Part 2 explores cost-center, revenue-expanding, and demand-limited jobs to help you assess the real level of risk to your position.
In my recent articles “MIT’s Project Iceberg Reveals AI’s Job Impact Is Far Bigger Than It Appears” and “The Canary in the Code Mine: What Tech’s Job Slump Means for the Rest of Us” I’ve repeatedly discussed AI-driven job loss. This led to many questions about which jobs are safe and why some jobs will be lost but other roles will be less affected or even increased despite performance improvements.
In “How to Know If Your Job Is Safe from AI — Part 1: What History Shows Us About Job Loss and Job Growth” we examined the impact of technological change in past industries. Looking at different cases helped us isolate why technology improvements for a specific job destroyed some jobs in the past but increased the number of jobs (not new types of jobs but the specific job it made more efficient), in other cases.
Now, in Part 2, we apply these concepts to industries today to understand who is most at risk. Many jobs generally fall into three categories, and we’ll explore what drives that division and the implications for those jobs.
Note that I’m focusing just on knowledge worker jobs. Many jobs that are primarily physical (e.g., plumbing) are not discussed. Likewise, some jobs can benefit from AI but are unlikely to be displaced. Teaching, for example, may get more efficient as teachers could potentially create individualized learning programs for each student, but we certainly couldn't cut the number of second grade teachers due to efficiency. The supervision part of the job cannot easily be automated away. These jobs, too, are not part of the analysis.
It comes down to the economics of a particular role to determine the possible impact of the cost savings. For a given industry, a specific job creates a certain amount of value. As the ROI on the job changes, it becomes a question of whether it should be applied to increasing revenue, reducing cost, or both. There are roughly three categories of job: cost-center jobs that don’t impact revenue, revenue-expanding jobs where additional capacity means more revenue, and demand-capped jobs that add value but only to a point. After exploring these categories, we’ll look at how to apply them to your own job to prepare for what’s to come.
It should be noted that a given job may have different ROI in different industries. An accountant at a manufacturing company is needed to keep the books but doesn’t create revenue; an accountant at an accounting firm generates value for the company. We’ll explore this after the three categories.
Certain jobs don’t provide any revenue directly. Consider customer support; while a bad customer experience will hurt revenue and good experience may help, they aren’t directly driving revenue the way sales, marketing, product, or R&D teams do. These are cost-center jobs.
Call centers and other customer service-type jobs are a pure cost for a company, driving little to no revenue themselves. Even if they do bookings, the bookings are driven by marketing, and the call center reps themselves do little upselling. In most industries these jobs are filled by low-cost workers with limited training. Companies do need call centers to support products and services, but typically do the minimum they need in terms of investment given the limited ROI. (Zeynep Ton’s book The Case for Good Jobs: How Great Companies Bring Dignity, Pay, and Meaning to Everyone's Work shows how better investment can lead to better results, but they are still primarily cost centers.) Every dollar saved is just that, a dollar saved. As such, companies will look to replace as many call center reps with AI as possible. Those dollars can be reinvested into other areas with a better ROI (e.g., marketing) or given to shareholders. In short, the jobs will simply be lost.
Similar jobs include data entry and processing, lower-end business processing outsourcing, accounts payable / accounts receivable, administration, insurance claims processing, IT support, benefits processing, loan and deed processing, medical coding, underwriting (for common cases), fraud dispute, some property inspection, permitting office clerks, hotel front desk (for lower end-properties), and asset recovery. Consider, no insurance company says, “Oh good, we can now process claims in half the time, so let’s hire more people to process twice as many claims!” The number of claims processed is a function of how many people are insured, which is driven by sales and marketing, not claims processing speed.
We’ve seen similar jobs replaced by automation with online ordering, ticketing, and check-out kiosks. Those didn’t need LLMs to be replaced; earlier technologies were sufficient. The jobs themselves were primarily cost centers and consequently removed.
In short, when a job is primarily a business cost, technology will always be used to make it more efficient. If it’s efficient enough to remove some or all of the workers, the company will do that. Paperboys simply went away when newspapers got delivered online.
The key is that those jobs do not create revenue or other value. If the call center reps performed upselling and generated revenue, then it may be a different case. If the AI couldn’t generate the revenue, then it may still make sense to keep the reps (if the AI can do the upselling as well as the humans, they’d still be in trouble).
These are jobs whose function does create revenue. Consider doctors. A medical practice or hospital is typically paid one of two ways. Traditionally, they are paid in a fee-for-services model; consequently, the more services that are performed, the more items the facility can bill for. Suppose a doctor orders one test or service on average per patient visit. In theory, AI can help the doctor better process medical history, review complaints, and recommend courses of action. This means the doctor can see more patients per day, thereby ordering more services per day, and getting more revenue per day. Unlike the call center reps, the doctors generate revenue.
The other model is capitation, where the facility gets a fixed fee per person (“per capita”) regardless of the number of services ordered. Again, faster analysis per patient means higher throughput which means the same facility can service more patients. Some models even offer benefits for the overall health of the population, in which case any help AI provides is further profit.
A facility could choose instead to reduce the number of doctors (and we can assume other staff would be pro-rated as well) and see the same number of patients in the first case, but with less staff and therefore less cost, so better margins. But assuming there’s a net positive profit margin per doctor, why would any company walk away from profit? A small private practice may have a doctor preferring extra time on the golf course, but a for-profit medical practice owned by a PE firm or a hospital chain would want to maximize earnings.
This only breaks down if either we run out of patients to see, or insurance companies cap costs. Since neither currently exists, AI will simply improve margins, and hopefully medical outcomes. (Healthcare is a complex industry and impacted by regulation so these assumptions can change over time.)
The general case here is when a service has a net profit margin. The limiting factor now becomes how fast and efficiently the service can be performed. If AI can speed up the process, more services can be performed (or more customers serviced), increasing revenue and profit. In this case a company won’t eliminate people, it will simply increase their productivity.
Importantly, there is no perceived limit to demand. Because most people would see more doctors if there were no marginal cost to them (cost being not only dollars but effort, say to get a referral); demand outpaces supply even with the technological gains. In many rural areas there is a lack of specialists so when ones do come in (e.g., some specialists travel to remote areas once a month to see patients) they can now see more patients per time spent traveling.
This includes doctors, nurses, and mental health counselors. Lab technicians, whether medical or other areas, also likely fall into this category. Sales and business development almost certainly are revenue-expanding jobs. Customer success could be if it drives revenue or retention in high-churn businesses, but it’s not always such a direct cause and effect. Data science and R&D roles are generally in this category, at least in the near future.
Consider lawyers. Legal services might seem similar to software at first (software was discussed in “How to Know If Your Job Is Safe from AI — Part 1: What History Shows Us About Job Loss and Job Growth”). Lawyers are expensive; if we could make the cost of legal services cheaper with AI, more people may avail themselves of legal support. Consider how many small business owners sign contracts without a lawyer reviewing them? It’s not worth spending $2,000 and many hours on a review. But if that cost was $200 and only an hour, it may be worth it. Cost goes down; demand goes up.
But there’s a limit to this need. Even if people can afford to have more contracts reviewed, there’s still a limit to how many contracts they will have in the first place. That limit is independent of the cost of legal review; it's a function of how many sales they can close. Consider a big corporation; for them the cost of a big contract, partnership, or M&A deal goes down because AI makes it more efficient. Will they suddenly do more deals?
A mid-sized company may take some of that legal savings and use it to do higher level legal strategy and planning (that is a nice-to-have, but it wasn't feasible in the budget before). They may, for example, ask for a review to determine legal risk and to be proactive on legal protection. But there’s a limit to this need; there’s only so many proactive reviews and so much legal strategy a company can consume. If legal work is only 10% more efficient, it may simply shift those dollars to this additional work. If it’s 500% more efficient, then a significant portion of those dollars may be reallocated to other areas or corporate profits. Legal work, by demand, is likely closer to web development than software development. Whether the line is closer to 10% or 500% I don’t know, but there does exist some limit.
Jobs in this category are at risk. The initial automation may lead to services simply moving up the value chain, that is, doing more strategic or complex work that wasn’t previously affordable. However, after a certain point, it will lead to layoffs as the amount of value the customer can get is capped. Similar jobs include accounting, financial analysis, compliance and risk analysts, curriculum developers, many HR jobs, safety inspectors, some types of research assistants, and architects.
Your mileage may vary for any given job and company. While the economics of some jobs may suggest keeping the same staff, some companies may simply opt to cut headcount. Presumably though, their competitors would not, so that would be company-specific and not industry-specific. It also assumes a company can capture additional market share with its newly freed up resources.
Regulated industries have another limitation. Legal work, for example, cannot be 100% automated. By this I mean if there’s a contract with 400 paragraphs and the LLM gets 399 right that’s not sufficient. A company can’t say “well, we avoided most of the legal risk.” 100% of the contract needs human review (at least today with probabilistic AI and the current standards and liability for legal services). Medicine, finance and accounting, aviation and aerospace, energy, and other industries with heavy regulations will have some countervailing limiting factors, at least in the short term.
A particular job may also be different depending on the industry. Creative roles, for example, may be demand-limited or revenue-expanding. If they’re creating corporate branding, or editorial, that is limited, since a company can only put out so much. But if they’re doing in-house creative work for a movie studio, now that studio can have many more options to choose from or do more revisions of a draft (e.g., movie storyboards) to perfect the movie and create more value.
Likewise with management consulting companies. If they can get more clients, then they’ll keep their staff and simply support more clients for more profit. On the other hand, if they can’t add additional clients, then they would be better off reducing staff to keep their own margins higher.
Finally, governments and non-profits are not profit-driven, but mission-based. Similar analysis can be done focusing on how much of a job supports the mission directly, rather than profit. An accountant may get automated away, while a social worker may become more efficient and able to deliver more services to more people. As long as there is more demand for social work than available workers, keeping the staffing the same as productivity increases delivery of the mission. (Of course, legislative policy such as budget setting may choose alternative options.)
Again, this doesn’t cover all jobs (and we’ve ignored jobs requiring significant physical activity), but it works as a starting point for understanding how your job may be affected as automation, from AI or something else, starts to handle tasks for which you had been responsible.
Product managers, project managers, and middle management can fall into any bucket. Pure project management jobs and middle management jobs in general may get eliminated as coordination gets easier. Of course, we’ve heard promises before of how technology will make things easier and it rarely seems to help as much as the vendors would claim. Product managers could do more and better analysis and find higher value areas for the teams to focus on. But if they get very efficient and the limiting factor becomes engineering or other groups, they could see a reduction in headcount.
Software is still an open question. In “How to Know If Your Job Is Safe from AI — Part 1: What History Shows Us About Job Loss and Job Growth” we saw that Jevons Paradox led to increased demand for software engineers. But, like law, there’s some limit. If we get close to the limit, wages will drop. If we are far from the limit, the paradox still holds and demand for the even cheaper software will continue to increase, causing wages to continue to stay high.
Ultimately, it depends on how the changing cost structure impacts demand. Jevons Paradox may create additional demand in an industry, but then any given company must also be able to capture that new demand.
You can likely find similar, perhaps more detailed analysis, in one of the thousands of reports put out by governments and think tanks these days that may go into more detail. But the principle remains the same in all cases.
Creative destruction in capitalism means some jobs are lost and new ones get created by technological processes. The risk today is in the gap between the old and new. Toll booth collectors could move to Amazon warehouses because both were low skilled jobs. It's a bigger challenge moving highly skilled professionals like accountants and lawyers. If the new jobs are skilled, there's a cost to training them up so that they can perform the new jobs (e.g., we can’t suddenly turn them into doctors). On the other hand, if they are unskilled then the jobs may not pay as well since anyone can do them. “Unskilled” is the key term since with AI many more jobs may become “unskilled” or low-skilled, meaning that significant training is not needed. Those website developer jobs that required advanced knowledge in the 1990s are now unskilled given the tooling advances in the past thirty years.
It’s not that there won't be jobs, but that the job displacement will be harder to address than in the past. This will be due to the size and speed of the displacement. The size is a double whammy. When print media jobs died, online media jobs grew, and it was a matter of shifting people from print to online jobs. Here the problem of the large size of the impact hits us on both sides. The large size means more people’s jobs will be impacted than by prior changes over a similar time-period. It also means there will be fewer unaffected industries that can absorb the displaced workers in that same time-period. Until the newer jobs appear, many will have a tough time.
For any individual, understanding how your company may respond to efficiency improvements in your role can help you better plan for what may happen to your job or industry in the next few years. Ultimately, you want to make sure that the nature of your role continues to add value to the company. For most of us, not every job is purely cost-saving or revenue-generating. Looking at the task level, make sure your job has enough value-generating tasks so that as the rote, cost-overhead parts of the job get automated away you will still be needed to drive more value to the company.
As I wrote in “How to Future Proof Your Job Against AI” you need to think about moving up the value chain. Think about each task you do in your job. There are likely dozens of them at least. For each task, think about how likely it is to be automated and if that particular task falls into the cost-center, revenue-expanding, or demand-limited categories. After much of your job has been automated away, what’s left? If it’s still cost-center tasks, then it's just a matter of time until you get replaced. If it’s demand-limited, think through what’s limiting the demand and how much headroom there will be in your company or industry. If it's revenue-expanding, you’re positioned well; focus more on those tasks. If it's either of the first two cases, start looking to shift your current job into more revenue-expanding opportunities or look for a new role that offers it. Good luck.
It’s critical to learn about corporate culture before you accept a job offer but it can be awkward to raise such questions. Learn what to ask and how to ask it to avoid landing yourself in a bad situation.
Investing just a few hours per year will help you focus and advance in your career.
Groups with a high barrier to entry and high trust are often the most valuable groups to join.
