Outsource, automate, or eliminate? A framework for deciding what to do with work that shouldn't be on your plate
Quick answer
When a business owner realizes they are drowning in operational work, there are three possible responses for any given task: eliminate it, automate it, or delegate it to someone else. The order matters. Eliminate first; a surprising share of recurring work exists out of habit rather than necessity. Automate second, but only tasks that are rule-based, high-volume, and require no judgment. Outsource or delegate third: the tasks that genuinely need human judgment, communication, or context but do not need yours. Most businesses get this sequence wrong: they hire help for tasks that should not exist, or they try to automate work that needs a human, and conclude the whole approach failed. This post gives you the decision framework — grounded in decades of academic research on exactly this question to get it right.
Why this question is harder than it looks
The question "should we outsource, automate, or just stop doing this?" feels like a modern dilemma born of AI tools and the gig economy. It is actually one of the oldest questions in economics.
The earliest literature on the make-or-buy decision starts with Ronald Coase in 1937, who asked why firms exist at all — why some work happens inside a company and some gets bought from outside. Oliver Williamson won the Nobel Prize in Economics for extending that work into what became transaction cost economics. The key insight of the theory is that make-or-buy decisions are determined not just by the price of the purchased item but by its transaction costs — the hidden costs of searching, negotiating, coordinating, and monitoring that surround every arrangement. ResearchGate
That insight is the reason your gut instinct on this question is often wrong. The visible cost of a solution — the software subscription, the VA's monthly rate, the hours saved — is only part of the equation. The invisible costs — setup time, error correction, coordination overhead, and the ongoing management each option requires — frequently decide which option actually wins.
Modern automation research adds a second principle. Academic frameworks for analyzing work automation emphasize that evaluation must happen at the task level rather than the job level, because depending on the level of specificity, the impact of automation ranges from eliminating entire jobs to eliminating only specific tasks. The same is true of outsourcing and elimination. "Should I get help with operations?" is unanswerable. "Should I outsource, automate, or eliminate my weekly reporting process?" has a clear answer. NBER
So the first move is breaking your work down into tasks specific enough that each one can be evaluated on its own.
Step one: eliminate (the option everyone skips)
Before asking who or what should do a task, ask whether the task should exist at all.
This sounds trivial. It is not. Recurring work accumulates in businesses the way sediment accumulates in rivers — slowly, invisibly, and without anyone deciding it should. The weekly report nobody reads. The approval step that made sense when the team was three people. The meeting that exists because it has always existed. The data entry into a second system that duplicates the first.
The test for elimination is simple: if this task stopped happening tomorrow, who would notice, and what would actually break? If the honest answer is "nobody" or "nothing important," the task is a candidate for elimination — and eliminating it costs nothing, forever.
Elimination has to come first in the sequence for a structural reason: automating or outsourcing an unnecessary task does not remove the waste, it just makes the waste cheaper and more permanent. A pointless report that a VA now produces reliably every Friday is worse than a pointless report that occasionally got skipped, because now it looks like a functioning system and nobody ever questions it again.
Run this filter across your task list first. Most businesses that do this honestly eliminate 10 to 20 percent of their recurring operational work before spending a dollar.
Step two: automate for rule-based, high-volume, judgment-free work
For the tasks that survive elimination, the next question is whether a machine can do them.
The academic literature on human-machine task allocation offers a useful caution here. Studies of automation in practice debunk full automation myths, emphasizing the persistent human roles in AI-assisted work — research on so-called "ghost work" reveals how automation reshapes jobs rather than eliminating them. In other words: most "automated" processes still have humans inside them, handling the exceptions, the edge cases, and the judgment calls the automation cannot. If you automate a task that is 80 percent rule-based, you have not removed the task — you have removed 80 percent of it and made the remaining 20 percent harder to see. arxiv
The tasks that genuinely automate well share three properties. They are rule-based: the correct action can be fully specified in advance, with no interpretation required. They are high-volume: the setup cost of the automation amortizes across many repetitions. And they are low-consequence on failure: when the automation breaks — and it will — the cost of the error is annoying rather than damaging.
Concrete examples that pass this test: invoice reminders on a fixed schedule, data syncing between two systems, appointment confirmation emails, form submissions routing to the right folder, and recurring report generation from structured data.
Examples that fail it, despite looking automatable: customer support responses (judgment about tone and context), inbox triage (interpretation of priority and intent), client follow-up (relationship context determines timing and content), and anything involving an unhappy human on the other end.
There is also a hidden cost the sales pages never mention: automation requires maintenance. Integrations break when tools update. Workflows need adjusting when your process changes. Someone has to own the automation stack — and if that someone is you, you have traded a recurring task for a recurring maintenance obligation.
Step three: outsource or delegate (for judgment work that doesn't need your judgment)
What remains after elimination and automation is work that genuinely requires a human — communication, interpretation, coordination, relationship context, exception handling. The final question is: does it require you specifically?
This is where transaction cost economics earns its Nobel. The framework says the choice between doing work in-house and buying it from outside depends on the nature of the transaction — and for most operational work in a small business, the analysis favors buying. Williamson's basic regularity is that as bilateral dependency builds up, efficient governance moves from simple market exchange toward hierarchy — and the "make" decision, doing it in-house, is viewed from the TCE perspective as the organization form of last resort. Springer
Translated out of economics language: keep work in-house only when it is so specific to your business, so dependent on deep context, or so strategically central that no external arrangement can handle it. Everything else is a candidate for delegation.
The tasks that delegate well are exactly the ones that fail the automation test: inbox and calendar management, client communication, CRM upkeep, follow-up sequences, research, coordination across vendors and team members, and the endless category of "things that need a sensible human for twenty minutes." These require judgment — but ordinary, trainable judgment, not founder judgment.
The transaction costs of delegation are real and worth naming honestly: writing the brief, onboarding, the first month of feedback and correction, and ongoing light management. This is why delegation fails when it is treated as a transaction — hire someone, hand off chaos, hope — and succeeds when treated as a system: documented tasks, defined standards, structured onboarding. The businesses disappointed by outsourcing are overwhelmingly the ones that skipped that preparation, not the ones that chose the wrong option.
Putting the three together
For every recurring task on your plate, run this sequence in order.
First: would anything break if this stopped? If no — eliminate. You are done, at zero cost.
Second: can the correct action be fully specified in advance, does it repeat at high volume, and is a failure cheap? If yes to all three — automate, and assign someone to own the maintenance.
Third: does it require human judgment, but not your judgment? If yes — delegate or outsource, with documented processes and a proper onboarding.
Fourth: does it genuinely require you — your expertise, your relationships, your authority? Then keep it. This is your actual job. Everything the first three filters removed was preventing you from doing it.
In practice, the three options are not competitors — they are layers, and most businesses end up using all three. A typical operational cleanup eliminates 10 to 20 percent of tasks outright, automates another 15 to 25 percent of pure rule-following work, and delegates 30 to 40 percent to a capable assistant — who, notably, often becomes the person who runs and maintains the automations. The question the lead in your position is really asking is not "which one?" but "in what order, and for which tasks?" — and the answer is: eliminate first, automate the rules, delegate the judgment, keep only what is genuinely yours.
For what the delegation layer looks like in practice — which tasks transfer well and what they cost — What tasks should you give a nearshore virtual assistant? covers the task-level breakdown, and How much does a nearshore virtual assistant cost in 2026? covers the budget side. On the relationship between automation and delegation specifically, AI Is Not Your Assistant. Your Assistant Should Use AI makes the case that the highest-leverage setup is not choosing between a human and the tools — it is a human who runs the tools.
Frequently asked questions
Should I automate or outsource a task?
Automate tasks that are rule-based, high-volume, and low-consequence on failure — where the correct action can be fully specified in advance. Outsource or delegate tasks that require human judgment, communication, or context but not your specific expertise. Trying to automate judgment work produces broken workflows; hiring humans for pure rule-following work wastes their capability and your money.
What tasks should a business eliminate rather than delegate?
Any recurring task where nothing meaningful would break if it stopped: reports nobody reads, approval steps that outlived their purpose, meetings held out of habit, and duplicate data entry. Run the elimination filter before automating or outsourcing anything — otherwise you are paying to preserve waste.
What is transaction cost economics and how does it apply to outsourcing decisions?
Transaction cost economics, developed by Ronald Coase and Nobel laureate Oliver Williamson, holds that make-or-buy decisions depend not just on price but on the full costs of the arrangement — searching, negotiating, coordinating, and monitoring. Applied to small business operations, it argues for keeping work in-house only when it is highly specific to your business; for most operational tasks, well-structured external help wins on total cost.
Why do automation projects fail in small businesses?
The most common causes are automating tasks that actually require judgment, underestimating the maintenance burden when tools update or processes change, and automating tasks that should have been eliminated. Academic research on automation consistently finds that humans remain inside "automated" processes handling exceptions — plan for that reality rather than assuming full hands-off operation.
Can one hire cover both automation and delegation?
Yes, and this is often the highest-leverage setup. A capable assistant can execute the judgment work directly and also build, run, and maintain the simple automations for rule-based tasks — giving you one accountable owner for the entire operational layer instead of a patchwork of tools and workflows nobody maintains.
How do I decide which tasks require me personally?
A task requires you if it depends on your expertise, your relationships, or your authority — advising clients, setting strategy, negotiating key deals, making hiring decisions. If a task only requires ordinary competence plus context that can be documented, it does not require you, regardless of how long you have been the one doing it.
Sources: Williamson, O.E., "Outsourcing: Transaction Cost Economics and Supply Chain Management," Journal of Supply Chain Management (2008). Coase, R., "The Nature of the Firm" (1937). Acemoglu, D. & Restrepo, P., "Automation and New Tasks," NBER Working Paper 25684. ScienceDirect, "A framework for research on the automation of work" (2025).
