In-House vs Agency vs AI: Content Production Compared

The question I get asked is "should we bring content in-house or keep the agency?" The question that actually matters is "which of our 200 annual assets belongs in which lane?" — because the answer is never all of them, and the teams that treat it as an all-or-nothing decision end up with either an overloaded internal team or an agency invoice full of work a junior could have generated on a Tuesday.

There are three production models available now, not two. In-house teams, external agencies, and AI-assisted production — which is a distinct model, not just a tool the other two use, because it changes who can produce, how fast, and what the unit economics look like.

This compares all three honestly, including where AI production loses, and ends with a routing framework you can apply to your actual asset list.

The three models at a glance

	In-house team	Agency	AI-assisted
Cost structure	Fixed (salaries)	Variable per project, high floor	Variable per asset, very low floor
Cost per asset at low volume	Very high	High	Low
Cost per asset at high volume	Low	High	Very low
Time to first draft	Days	1–3 weeks	Minutes to hours
Revision cycle	Hours	Days	Minutes
Quality ceiling	High, capped by team skill	Highest	High, capped by model and prompt skill
Brand knowledge	Deep, built-in	Shallow to medium, needs briefing	None inherent; enforced through references and workflows
Strategic input	Medium	High — you're partly buying judgment	None
Scales with	Headcount	Budget	Consumption

The line that decides most cases is the fourth one. An agency's revision cycle is measured in days because a human has to schedule the work. AI production's revision cycle is measured in minutes, which changes not just the speed but the kind of creative process you can run — you can test twenty hooks instead of arguing about three.

Where each model genuinely wins

In-house wins on brand knowledge and responsiveness. Someone who sits in your product meetings makes better creative decisions than someone reading a brief, and they can turn a Slack message into a post the same afternoon. The catch is fixed cost: an in-house video team is expensive whether you publish 5 assets or 50 in a month, and most teams publish closer to 5 than they'd like to admit.

Agencies win on the top end and on judgment. The brand film, the campaign concept, the launch that has to land — these benefit from people who have done it fifty times for other brands and will tell you your idea is bad. You are buying taste and accountability alongside production. That's real and it's worth paying for, on the four or five assets a year where it matters.

AI-assisted production wins on volume, iteration, and the long tail. The assets that never got made because they weren't worth a production day. The fifteen variants a performance campaign needs. The localized version for a market that doesn't justify a shoot. The explainer for a feature that might change next quarter.

Where AI production actually loses

Being specific here matters more than the wins, because this is where budget gets wasted.

• Anything featuring your real people or premises. Recruiting videos, office culture content, the founder's face — the value is that it's real. Digital twins exist and are good, but they need consent and setup, and they're a supplement, not a substitute.
• Genuine customer testimonials. Do not generate these. This is a brand-safety line, not a capability question, and the brand safety checklist treats it as non-negotiable.
• Campaign strategy. Models generate assets, not positioning. An agency's strategic contribution doesn't have an AI equivalent yet.
• Very long continuous takes. Multi-scene films work well through chaining and combining. A single unbroken three-minute shot does not.
• Anything a regulator signs off on. The production speed is irrelevant when the bottleneck is compliance review.

The hybrid that most teams land on

After a year or so of adjustment, the stable configuration for a mid-size marketing team looks like this:

• Agency retained, but scoped down to 4–6 flagship assets a year plus campaign strategy. The retainer shrinks; the relationship stays. Agencies that adapt to this do fine; the ones billing for volume production don't.
• One or two in-house owners who run the AI production line and handle the responsive, brand-critical work. Not a full studio — an owner plus reviewers.
• AI production carrying the volume: paid variants, explainers, b-roll, localizations, always-on social. Typically 80–90% of asset count and a small share of total spend.

The thing that makes this work is not the tooling; it's the routing rule. Without one, every asset defaults to whichever lane the requester likes best, and you get an agency invoice for social filler.

A routing framework for your actual asset list

Score each recurring asset type on three questions:

1. Does it need to be verifiably real? (Real people, real premises, real customers.) If yes → in-house shoot or agency. Stop here.
2. Does it carry disproportionate brand or revenue risk if it's merely fine? (Flagship launch, brand film, investor-facing.) If yes → agency, or in-house with agency oversight.
3. Everything else → AI production.

Applied to a typical B2B marketing calendar, that routes something like:

Asset type	Annual volume	Lane
Brand film	1	Agency
Campaign hero video	3–4	Agency or in-house with agency concept
Customer testimonials	6–10	Real footage, in-house or agency
Product/feature explainers	20–30	AI production
Paid social variants	200+	AI production
Localized versions	30–60	AI production, via dubbing and lipsync
Always-on organic social	150+	AI production, via reusable workflows
Recruiting and culture	4–6	Real footage, in-house

Note the shape: agency work is a small number of high-consequence assets, AI production is the overwhelming majority of asset count. That's the correct distribution, and it's roughly the inverse of how most budgets are currently allocated.

Transitioning without breaking anything

Three practical notes from teams that have done this well:

• Move one category, not everything. Start with the highest-volume, lowest-risk lane — usually paid variants or organic social. Prove it for a quarter before touching anything the agency considers core.
• Renegotiate the retainer at renewal, not mid-term. Going back mid-contract to cut scope damages a relationship you'll still want for the flagship work.
• Build the review gate before the volume arrives. The bottleneck moves from production to approval almost immediately. If one person has to approve 200 assets a month individually, you've recreated the agency queue internally. Approve formats, not clips — the mechanics are in AI content governance.

For the cost-side comparison in more depth, cost per creative: AI vs agency breaks down the unit economics, and AI video cost savings vs agency covers what teams actually recovered once the retainer was rescoped.

FAQ

Is AI content production cheaper than an agency?

Per asset, dramatically — the gap widens with volume because AI production has almost no floor cost while agencies have a high one. But the comparison is only fair for assets both lanes could produce. For a brand film or campaign strategy, the agency isn't more expensive; it's doing something else.

Should we fire our agency and bring everything in-house?

Almost never. The pattern that works is scoping the agency down to flagship assets and strategy while volume production moves in-house with AI assistance. That usually shrinks the retainer by half or more without losing the capability you're actually paying for.

How many people do we need in-house to run AI production?

One dedicated owner handles roughly 50 assets a month comfortably, with reviewers who don't generate. Below 20 assets a month it can be a part-time responsibility. The scaling constraint is review capacity, not generation capacity.

What happens to our agency relationship?

Agencies doing strategic and flagship work generally welcome the change, because volume production was low-margin for them anyway. Agencies whose revenue is mostly volume deliverables will resist it, and that's useful information about what you were buying.

Can AI production match agency quality?

On individual hero assets, usually not — an agency's best work has a ceiling you won't reach with prompts. Across a portfolio of 200 assets, AI production wins easily, because the agency would never have made 190 of them.

If you want to test the routing framework rather than argue about it, take the highest-volume row from your own asset list and run one month of it through the AI video generator alongside your existing process.