---
title: "In-House, Agency, or Do Nothing: Deciding Your AI Upgrade"
seoTitle: "In-House, Agency, or Wait: Your AI Upgrade"
description: "Three honest options face a small business on AI: build in-house, hire help, or wait. How to decide which one fits your situation, budget, and risk tolerance."
datePublished: "2026-07-14T10:31:00Z"
dateModified: "2026-07-14T10:31:00Z"
category: business
imageAlt: "Iron Goo blog featured image on choosing between building an AI upgrade in-house, hiring an agency, or waiting."
tags: [ai-strategy, ai-adoption, build-vs-buy, smb-ai, hiring]
faq: true
---
Most owners size up an AI upgrade as inhouse vs agency, two doors with a wall between them: do it ourselves with the people we have, or pay someone outside to do it for us. The whole decision gets spent arguing over those two. There is a third door in the same wall, and owners walk past it because it does not feel like a decision at all. The third door is wait. Not "never," not a stall, but a deliberate "not yet" that is sometimes the most honest call on the table. A binary that hides one of its own options is not a real choice; it is a coin flip with the edge case painted over. The first move is to put all three doors back on the wall.
Naming the third option changes the question you are actually answering. It stops being "which way do we do this" and becomes "should we do this now, and if so, how." Those are different questions, and the second one is the one your money and your calendar care about. Plenty of upgrades that failed in-house and plenty that burned a five-figure engagement were never build-or-buy problems. They were timing problems wearing a build-or-buy costume.
:::callout{type="key" title="Three doors, not two"}
A small business facing an AI upgrade has three real options, not two. Build it in-house with your own people. Hire an agency or contractor to execute it. Or wait, on purpose, until the business or the job is ready. The binary feels complete because waiting does not announce itself as a choice. It is. Treat it as the third door and the decision gets honest.
:::
## Why "wait" is a real option and not a cop-out
Waiting gets dismissed because it looks like the absence of a decision, the thing you do when you cannot decide. That is the lazy version. The deliberate version is different. It says the upgrade is worth doing, and the right conditions for doing it well are not here yet, so spending now buys a worse result than spending in three months. An agency will usually take that job today, because they sell execution and you are asking them to execute. The job still goes badly, because the problem was never who held the keyboard. The problem was that the thing being automated was not defined, or the team had no time to adopt it, or the data it needed to run on did not exist in any usable form.
Real reasons to choose the third door: the job is still a vague wish ("use AI to help with customers") rather than a specific, repeating task someone could write down. The business has no slack to absorb a change right now, because a busy season or a staffing gap means a new process would land on people with no room for it. Or the upside is real but the downside of a fumbled rollout is worse than the cost of waiting a quarter to do it cleanly. None of those is fixed by hiring harder. They are fixed by time, or by a small bit of prep, and then the same decision gets made again from firmer ground.
The honest test for this door is readiness, and readiness is a thing you can actually judge rather than guess at. If you want a structured way to tell "not yet" from "now," the rundown of [how to tell whether your business is ready for an AI upgrade](/blog/ai-readiness) gives you the signals to read, and the longer [readiness walkthrough for small businesses](/guides/ai-automation/ai-readiness-for-smbs) turns those signals into a check you can run before you spend a cent. The point of the wait door is not to stall. It is to come back to the same fork standing on better ground.
## The factors that actually decide it
A decision framework is only useful if its inputs are things you already know about your own business. These are. Six of them carry most of the weight, and each one leans toward a different door depending on where you sit.
**Internal capacity and spare time.** Not "do we have smart people," almost every small business does, but do those people have hours that are genuinely free, not borrowed from work that still has to happen. An AI upgrade done in-house is not free; it is paid in your team's attention. If that attention is already spoken for, "in-house" is a polite way of saying "it will start, stall, and quietly die."
**How well-understood and complex the job is.** A task you can describe precisely, that runs the same way every time, is a job either you or a contractor can ship cleanly. A task you can only gesture at is not ready for either, no matter who you point at it. The clearer and simpler the job, the more in-house becomes plausible. The more it needs specialist execution you do not have on staff, the more an agency earns its fee. And if you cannot define it at all yet, that is the wait door telling you to define it first. Picking the one small job worth doing first is its own skill; the breakdown of [the first jobs to hand to AI](/blog/first-jobs-for-ai) sorts candidate tasks by exactly this kind of clarity.
**Whether there is a deadline.** A real, external deadline (a season, a launch, a contract) changes the math. In-house work moves at the speed of whatever spare capacity you have, which is slow and lumpy. If the clock is fixed and the in-house lane cannot reliably hit it, that is a strong push toward hiring, because buying speed is one of the few things money straightforwardly buys.
**Whether the business is ready.** The readiness factor is the gatekeeper for the whole framework. If the answer here is no, the other factors barely matter, because none of the three doors leads anywhere good until readiness is fixed, and only the wait door admits that.
**Risk tolerance.** How bad is a fumbled rollout. For a low-stakes internal task, a rough first attempt is fine and in-house learning-by-doing is cheap. For anything that touches customers or money directly, the cost of getting it visibly wrong is high, which argues for either specialist execution or waiting until you can do it carefully, not a learn-on-the-job in-house experiment in public.
**Budget.** Last on the list on purpose, because it caps the options rather than choosing between them. In-house spends your people's time; an agency spends cash; waiting spends neither but defers the upside. The honest budget question is not "what does it cost" but "which currency can I afford to spend right now, time or money," and the cost of an AI upgrade and the shape it takes is worth understanding before you answer. The [breakdown of what an AI upgrade actually costs](/blog/ai-upgrade-cost) is the place to ground that number.
::::comparison{title="When each lane fits"}
:::side{label="When in-house fits"}
You have people with genuinely free hours, not borrowed ones. The job is simple and well-understood, the kind you could write down in a paragraph. The stakes of a rough first attempt are low, so learning by doing is cheap. And you want to own the capability long-term rather than rent it. A capable team, a clear small job, and slack to absorb it: that is the in-house case, and it is a real one.
:::
:::side{label="When an agency fits"}
You have no internal capacity, or the people you have are fully spoken for. The job needs specialist execution you do not have on staff and do not want to build for a one-time push. There is a real deadline the in-house lane cannot reliably hit. The case is well-defined enough to hand off cleanly. No capacity, a specialist job, or a fixed clock: that is when hiring earns its fee instead of just spending one.
:::
::::
Notice that nothing in either column is about which option is "better." Both are real, and which one fits is decided entirely by where your six factors land, not by anyone's opinion about agencies.
:::callout{type="warn" title="When the right answer is wait"}
Hold off, on purpose, when the job is still a vague wish rather than a defined repeating task, when the business has no slack right now to adopt a change, or when a fumbled rollout would cost more than a clean one done a quarter later. Waiting here is not indecision. It is refusing to spend time or money on a result you already know would be worse than the one a little patience buys.
:::
## Should I build my AI upgrade in-house or hire an agency?
It depends on three factors: capacity, job complexity, and readiness. In-house fits a capable team with free time and a simple job. An agency fits no capacity, a specialist job, or a deadline. Waiting fits a business or a job that is not yet ready.
That is the short version. The rest of the work is placing your own situation against those factors honestly, which is where most of the value is, because the framework only helps if you are truthful about your own inputs.
## How to place your own case
Run your situation through the factors in order, and the door usually picks itself. Start with readiness, because it is the gate. Be hard about it: is the job a specific, repeating task you could write on an index card, or a hopeful direction. Is there room in the next month for people to actually change how they work, or is every hour already claimed. If either answer is shaky, you are at the wait door, and the right move is to spend a little time defining the job and clearing the runway, then come back to this same fork. That is not failure. It is the cheapest version of success, because the alternative is paying full price for a result you already suspect will disappoint.
If readiness clears, weigh capacity and complexity together, because they trade off. Free hours plus a simple, well-understood job points hard at in-house, and if you also want to own the capability rather than rent it, that settles it. No free hours, or a job that needs execution you do not have on staff, or a deadline you cannot otherwise hit, points at hiring. Then let risk and budget adjust the call at the margin: high stakes argue against a public in-house experiment; a tight cash budget argues against an agency this quarter and may itself be a reason to wait until the money or the time frees up.
Be wary of one tilt in particular. Owners lean toward in-house for the wrong reason all the time, because hiring feels like an admission that the team could not handle it. That is ego, not analysis, and it is expensive. The mirror error is hiring out a job nobody on the inside has bothered to define, then blaming the agency when the undefined thing comes back wrong. The framework exists to take both of those out of the decision and leave only the factors.
:::quote{cite="A composite owner, voicing the common version"}
We almost did it ourselves to prove we could, and we almost hired it out to make it disappear. What we actually needed was two weeks to define the job, and then the answer was obvious.
:::
Work the factors and one of three things is true. The job belongs in-house, and you now know it is because you have the capacity, the clarity, and the appetite to own it, not because hiring felt like losing. The job belongs to the wait door, and you have a short, concrete prep list instead of a vague unease. Or the framework points outward: no capacity, a specialist job, or a clock you have to beat, and the move is to hand it to a team that does this work. If that is where you land, [handing the AI upgrade to a team that executes this kind of project](/services/aio) is the option to weigh, and the case for it is that it was your own factors, not a sales pitch, that sent you there.
So do not answer "build or buy" today. Answer the real question first. Take the one AI upgrade you have been turning over, run it through readiness, then capacity, then complexity, then the rest, and write down which of the three doors your own factors actually point to. If it is wait, you just saved a fortune. If it is in-house, you go in clear-eyed. If it is hire, you now know why, and where to look.