You sit through an AI demo. It summarizes a customer call in seconds, drafts the follow-up email, updates the CRM, and produces a report that used to take someone half a morning. Then somebody in the room asks the question everyone was already thinking: do we still need the person doing this job?
The honest answer is that you cannot know yet. You have only seen a few tasks performed well. A job is bigger than that, and the gap between what a demo shows you and what a role actually requires is where most automation decisions go wrong.
The employee who updates the CRM may also calm down an upset customer, notice when a sales opportunity is quietly dying, train a new hire, spot an exception to company policy, and make judgment calls using context that never shows up in any software.
AI can be excellent at one part of that role and unreliable at another. That is why the smarter starting point is the work itself. Break the role apart first, and the decisions that follow, about automation, hiring, capacity, and where your people create the most value, get much easier to make well.
A Job Title Hides the Work Inside It
Consider a sales manager. That sounds like a single job. Now look at what typically happens during the week.
- Review pipeline reports and prepare forecasts.
- Summarize sales activity for leadership.
- Coach underperforming team members.
- Approve discounts and resolve disputes between reps.
- Join important customer calls and handle escalations.
- Review account histories and identify stalled opportunities.
- Decide where the team should focus its energy.
- Interview and evaluate potential hires.
Those activities sit inside one position, but they do not require the same capabilities. AI can likely prepare a pipeline summary faster than the manager and may spot patterns across account activity that a person would overlook. It can help draft a forecast and organize notes before a coaching conversation. But deciding why a strong salesperson has suddenly stopped performing, or whether your biggest customer deserves an exception to policy, calls on judgment that lives outside the data.
A job is an organizational label. A task is a unit of work. AI operates much closer to the second, and that distinction changes the entire conversation about what to automate.
Start With Workload Before You Talk Headcount
There is an understandable pull toward turning every AI discussion into payroll math. If technology can complete work that currently takes an employee ten hours a week, the arithmetic feels obvious: ten hours disappeared, so part of the job disappeared with it. Business rarely works that cleanly.
Picture someone who spends twelve hours a week gathering information, entering data, formatting reports, and scheduling, another twelve hours analyzing problems and reviewing work, and the remainder of the week on customers, exceptions, and team coordination. Suppose AI removes eight of the administrative hours. The job has changed shape, but the person is still needed. You now have eight hours of freed-up capacity, and what happens to it is a leadership decision, not an automatic outcome.
That employee might manage more accounts, spend more time on retention, take work off an overloaded manager’s plate, or help build a process that has been neglected for two years because nobody had the time. You might delay the next hire. Eventually the role itself could shrink or disappear, but that should be a conclusion you reach after watching how the capacity gets used, not the assumption you start with.
The Research Backs Up This Task-Level View
This is not just a useful management exercise. It reflects where the research on AI and employment has landed. A 2025 joint study from the International Labour Organization and Poland’s NASK evaluated close to 30,000 occupational tasks rather than assigning a single automation score to each job title. The researchers found that roughly a quarter of global employment sits within occupations with meaningful exposure to generative AI, and their central conclusion is that transformation of work is far more common than clean replacement of entire occupations.
That finding lines up with what shows up inside almost any professional role once you look closely. A job rarely contains one kind of work. It contains a mix, and the mix is what determines the outcome.
Real Estate Makes the Task Differences Easy to See
Take residential real estate. People often talk about the profession as though “real estate agent” describes a single activity. During a transaction, an agent collects information from a client, researches properties, coordinates appointments, prepares marketing material, reviews market evidence, visits homes, negotiates terms, monitors deadlines, and handles the unexpected problems that show up before closing.
Compare two of those activities directly. Scheduling a showing can be structured around calendars, availability, defined rules, and confirmations, so it fits an automated workflow well. Negotiating an offer involves timing, incomplete information, competing objectives, and strategy, which calls for a person who can read the situation as it changes. Or compare organizing a client’s stated property criteria with recognizing that the client has quietly started reconsidering those criteria after touring several homes. The first is largely about information. The second is about understanding a person.
Once a transaction gets broken into its individual tasks, each with its own requirements for data, judgment, accountability, and human presence, asking whether AI can replace a real estate agent becomes the wrong question. The more useful questions are smaller: which tasks are predictable, which depend on dependable data, which can be verified quickly, which require physical observation, which involve professional judgment, and who remains accountable when something goes sideways.
The same lens applies to accountants, consultants, marketers, recruiters, attorneys, project managers, and nearly every knowledge-based role your business depends on. Map the work before you touch the title.
Sort Every Task Into Four Categories
Once the work is visible, move past the simple yes-or-no question of whether a task is automatable. Real businesses need more range than two options. I sort tasks into four categories, and most professional roles contain a mix of all four.
Automate the Predictable Work
Some work should run with minimal human involvement. Look for tasks that are repetitive, structured, rule-based, supported by dependable information, easy to verify, and low-consequence or reversible when something goes wrong. Routine scheduling often fits here, along with predefined reminders, moving approved information between systems, formatting standard reports, and routing straightforward requests.
Tools like Zapier or Make can connect your CRM, calendar, and inbox to handle these handoffs without a person touching them, and a scheduling assistant like Calendly removes an entire category of back-and-forth email. If a capable employee spends hours every week moving information from one system to another by hand, the issue is rarely the employee. It is usually a workflow that was never fixed.
Let AI Assist While a Person Owns the Result
This will likely be the largest category for most professional businesses. AI does part of the work, and a person owns the result. Think about research, drafting, data analysis, document review, meeting preparation, summarization, forecast preparation, and first drafts of customer communications.
Tools such as Claude or ChatGPT can turn a blank page into a working draft, and a call-recording tool like Otter.ai or Fireflies can turn a client meeting into a searchable summary before the follow-up email even gets written. A two-hour preparation task becomes a thirty-minute review, and a pile of raw notes becomes an organized starting point. The value is not that AI becomes the employee. The value is that the employee stops starting from zero. There is a condition attached to all of this, though: someone still has to know whether the output is good, because a faster wrong answer is still wrong.
Keep the Highest-Stakes Work Human-Led
Some work earns its value precisely because a person is doing it. Negotiating an important agreement, coaching an employee, managing conflict, delivering difficult news, giving consequential advice, reading a room, and making a decision when the information is incomplete all belong here.
AI can still help by organizing information before the conversation, surfacing relevant history, and helping someone prepare. That is support, and it is a different thing than ownership. This distinction matters more as the technology gets more capable. The question worth asking is whether you are comfortable letting software carry the responsibility attached to a decision, not whether it can produce something that resembles one.
Build an Escalation Path for the Exceptions
This is the category businesses forget most often. A task can be safe to automate ninety percent of the time, and the other ten percent can be exactly where you need your sharpest people paying attention. Picture a customer refund. Most requests fit a clear policy, but then one comes from your largest account, the contract language is ambiguous, the customer is already frustrated, and a renewal negotiation is sitting in the background.
Same task on the workflow chart, entirely different business situation. Good automation needs a defined way to recognize when the normal rule stops applying. That means naming your exceptions, stop conditions, confidence thresholds, high-value cases, and escalation paths in advance, along with who owns the decision once it escalates. Maturity in automation sometimes means knowing when to stop, and building that stopping point in deliberately is a sign of a well-run process rather than a gap in one.
The Hardest Fifth of the Work Holds Most of the Value
A vendor tells you their system can automate eighty percent of a workflow. That sounds impressive until you ask what is hiding in the remaining twenty percent. It often contains the unusual customer, the difficult negotiation, the incomplete information, the high-dollar decision, the exception to policy, and the judgment call nobody can reduce to a clean rule.
Eighty percent of the activity does not automatically represent eighty percent of the value, and it certainly does not represent eighty percent of the responsibility. In professional services, the routine work tends to consume most of the time while the difficult work creates most of the value, so technology can shrink the first without touching the second. As routine work gets cheaper, the human work that remains becomes more concentrated around exactly the situations where judgment matters most.
This shows up in the broader adoption data too. McKinsey’s November 2025 State of AI survey found that only about a third of organizations report scaling AI across the enterprise, and companies under $100 million in revenue lag well behind larger firms in reaching that stage. McKinsey’s researchers point to workflow redesign, not the underlying model, as the difference between organizations that see measurable value and those stuck running endless pilots. That finding backs up a point worth repeating to your leadership team: buying the tool is the easy part. Redesigning the work around it is where the value actually gets built.
Automating a Broken Process Just Speeds Up Confusion
There is a second problem many businesses underestimate. AI exposes a weak process fast. Picture three employees handling the same customer request three different ways, with nobody able to clearly explain who approves an exception, information scattered across a CRM, two spreadsheets, an inbox, and someone’s memory, and a handoff between departments that depends on one person remembering to send a message.
Automate that today, and speed is the only thing that improves. Clarity stays exactly where it was, and AI can now move through a broken workflow faster than your employees ever could, which tends to make the underlying problem more visible rather than less.
This is why automation needs clear boundaries, defined inputs, named ownership, and escalation rules before you let it run on its own. Before automating an important process, your team should be able to answer what starts the process, what information it requires and where that information comes from, which decisions happen along the way, who owns the outcome, what a normal case looks like versus an abnormal one, and who takes over when the system stops.
A simple way to pressure-test this is to ask five people on the team to answer those questions independently. If you get five different answers, you have found a process problem before you spent a dollar automating it.
Some Tasks Deserve Elimination Before Automation
A task audit does more than surface work for AI to handle. It also reveals work that nobody should be doing at all. Every business accumulates a few of these over time: the weekly report nobody reads, the duplicate approval added three years ago because of one mistake, the spreadsheet maintained because the CRM was never configured correctly, the meeting that exists only to pass information from one person to another, and the manual handoff between systems that could have been connected years ago.
Not every task deserves to be automated. Some deserve to be cut. That gives you a more useful sequence to run through as you evaluate the work: eliminate, simplify, automate, assist. The alternative, automating everything you currently do, risks using genuinely impressive technology to preserve work that never needed to exist in the first place.
Run a One-Role Audit This Month
You do not need a company-wide transformation project to get started. Pick one role where the workload is significant enough that better use of AI could actually move the needle, and list what that person does during a normal month. Not the responsibilities written in the HR file. The specific tasks. “Marketing” is not a task; “turn a customer webinar into three social posts” is. “Sales management” is not a task; “review stalled opportunities before the Monday pipeline meeting” is.
A simple way to organize this is a task-scoring worksheet in a tool like Airtable or Notion, with one row per task and columns for each of the questions below. Score every task against these seven questions.
- Is the task repetitive? Repetition creates the clearest automation opportunity.
- Are the inputs dependable? A sophisticated system working from bad information just produces more sophisticated-looking mistakes.
- Is the output easy to verify? If checking the result takes longer than doing the work yourself, you have not created much leverage.
- What happens if it is wrong? An incorrect internal summary and an incorrect wire instruction belong in very different risk categories.
- Does it require context that is hard to capture in a database or form?
- Does it require judgment, meaning the normal rule stops being enough?
- Who owns the result? “The AI did it” should never become an acceptable answer inside your company.
Once each task is scored, sort it into one of the four buckets from earlier: automate, assist, human-led, or escalate. Then step back and look at the role as a whole. That is the point where the staffing conversation finally becomes useful, because now it is grounded in the work rather than a guess based on a demo.
Bring Your Team Into the Audit With You
Here is a gap that trips up a lot of well-intentioned leaders: running this audit entirely in a leadership meeting, without the people actually doing the work in the room. Your employees usually know exactly which parts of their job are mindless and which parts require their judgment, and they will tell you if you ask them directly and make it clear the goal is to remove drudgery, not headcount.
Frame the conversation honestly. Tell your team you are examining the tasks inside the role to figure out where AI can take work off their plate, and that the goal is to free up their time for the parts of the job that matter more, not to shrink the team by default. A manager I coached said it well during a session: “My team was terrified the first time I brought up AI, until they realized I was asking them to help me find the parts of their job they hated doing.” That kind of transparency tends to surface tasks leadership never would have spotted on its own, and it turns the audit into something your people participate in rather than something that happens to them.
Track the Capacity You Create Every Quarter
There is one more step that businesses skip constantly, and it is the one that determines whether any of this actually pays off. Imagine automation gives a team member back six hours a week. What happens to those six hours matters as much as anything else in this article. Left undirected, the calendar simply expands, meetings creep back in, low-value work fills the space, and everyone becomes busy again without the business changing at all.
Set a simple quarterly checkpoint for every role you redesign. Note the hours you expected to free up, where you decided that capacity should go, whether it actually went there, and what changed as a result, whether that is more accounts managed, faster response times, or a project that finally got finished. A basic tracker in a spreadsheet or Notion is enough. The point is to make the decision about reclaimed time deliberate instead of accidental, because AI creates available time. It does not decide what that time is worth. That call belongs to you.
Redesign the Role Before You Touch the Org Chart
Suppose a quarter of a person’s current workload disappears and another third gets meaningfully faster. What should happen to the role has no universal answer. You might increase the number of accounts the employee manages, centralize work that several people currently perform separately, move a coordinator toward customer success, give a manager more time for coaching, shift an experienced employee toward the harder cases, or delay a planned hire. Sometimes you will conclude that technology has removed enough of the work that the position no longer makes economic sense, and that conclusion is legitimate once you have done the work to reach it honestly.
The sequence matters more than any single decision inside it. Understand the work first. Decide what should be eliminated. Identify what can be automated. Determine what should be assisted. Protect the work that still requires human judgment. Define the exceptions. Measure the capacity you created. Redesign the role. Only after all of that should you redraw the organization chart. That process might result in fewer people on the team. It might also result in the same number of people producing considerably more, with support roles becoming more specialized and managers able to oversee larger teams because less of their time goes to coordination.
Ask the Better Question About Your Business
“Which jobs will AI replace?” makes a good headline, but it is a weak operating question. Your company does not run on job titles. It runs on thousands of individual activities that eventually get bundled into roles, departments, processes, and customer experiences. Some of those activities should disappear. Some should be automated aggressively. Some should get faster with AI assistance. Some should stay firmly human-led. Some should run automatically until the situation stops looking normal, at which point a person needs to step back in.
Your job as a leader is knowing the difference, task by task, before you buy another tool or redesign a team around what a software demo appeared capable of doing. Choose one important role. Break it apart. Look at the tasks, the information behind them, the consequences of getting them wrong, and where a customer or employee genuinely needs a person on the other end. You may find that AI can eliminate part of a job, or even the whole thing. More often, you will find something more valuable: the technology removed the work that was quietly preventing a good employee from spending time on the part of the job that mattered most.
If you want a second set of eyes on which role to start with, or help scoring the tasks inside it, that is exactly the kind of work we do with clients at Deliberate Directions. Schedule a complimentary strategy session with one of our business coaches, and we will help you turn this audit into a plan your team can actually run with.
Not sure if you're the bottleneck? Find out in 2 minutes →
