Choosing a partner
How to choose an AI production partner
Every deck in this category looks identical. Seven questions cut through all of them.
Brief five agencies on AI content production and you will hear five nearly identical pitches: the same model names, the same efficiency claims, the same demo reel of generated footage. The decks have converged because the tools have converged - everyone can license the same software. What has not converged is the operating discipline behind the tools, and that is the thing you are actually hiring. These seven questions expose it. We answer them for clients weekly, so we are showing our own workings as much as anyone else's.
1. Show me a system that ran at volume
Not a demo. Not a pilot. A production system that shipped real volume for a real brand, and what it moved. Ours are on the table: 100,000+ personalized films for one launch and a 250-film always-on engine. Whoever you brief, demand the equivalent - numbers, duration, outcome. In this category the gap between a demo and a delivery is where programmes go to die.
2. Where do the humans sit, and what do they protect?
The wrong answers are "everywhere" (you are paying agency margins on manual work relabelled as AI) and "nowhere" (nobody owns the brand). The right answer names the split: humans lock the idea, the script logic and the brand system; the machine multiplies executions inside those rails; humans review at the edges. If a partner cannot draw that line crisply on a whiteboard, it does not exist in their pipeline.
3. What happens between asset one and asset ten thousand?
Volume flattens quality by default - every honest producer knows it. So ask precisely: what is the review system, what gets sampled, what triggers a human pass, and what does the ten-thousandth asset look like next to the first. Craft at scale is a solvable problem, but only for partners who treat it as a discipline rather than assume the tooling handles it.
4. How do languages actually work?
Every AI tool now claims localization; almost all of it is translation with better voices. Persuasive work that must land in Hindi, Telugu, Bahasa or Arabic needs transcreation - the idea rebuilt natively, not the words swapped. Ask who writes each language, whether voice is native or synthetic-translated, and who signs off that the joke still lands. For multi-market brands this single question eliminates half the field.
5. What data does personalization need - and where does it live?
One-to-one video runs on customer data: names, languages, regions, purchase history. Ask what the programme minimally needs, how it is secured and processed, and what happens to it after the campaign. A partner who has run personalization at six-figure scale answers instantly, because a real deployment forces the discipline. Marketing-in-regulated-categories experience is a strong signal here.
6. Does the price reflect the economics?
AI production's whole point is that the marginal cost of the next version, language or personalization collapses once the system is built. If the quote still scales linearly per deliverable, the efficiency exists - it is just not being passed to you. Expect system-based scoping: pay to build the engine, then a fraction per asset it produces. Anything else is old pricing wearing new vocabulary.
7. What do you keep if you part ways?
Templates, prompts, variation logic, trained voice models, the brand system the engine runs on - who owns them? The honest answer should be contractual and boring. A partner confident in their judgement does not need to hold your production system hostage; the ongoing value is the thinking, not the lock-in.
Tools are rented by everyone. What you are hiring is the discipline around them - and discipline only shows up under questioning.
Scoring the answers
Any serious partner clears four of the seven instantly. The last three - quality at volume, language depth, and pricing that reflects the real economics - are where the field thins out fast. If you want the market context before you shortlist, our honest map of the AI content production landscape covers who is genuinely for what, including where we are the wrong choice. And when the brief is ready, this is how we run AI production - bring the seven questions; we enjoy them.
Questions, answered
What should I ask before hiring an AI content production agency?
Seven things: system proof at volume, where humans sit in the pipeline, how quality holds to asset ten thousand, how languages are handled, what data is needed and how it is protected, whether pricing reflects collapsing marginal cost, and what you keep if you part ways.
What is the biggest red flag?
A pitch that leads with tools instead of outcomes. Everyone licenses the same software; a partner whose proof is a tool list has no proof. Look for a shipped system - real volume, a real brand, a result they will state and stand behind.
Should AI production cost less than traditional production?
Per asset, dramatically less - that is the point. If pricing still scales linearly with deliverables, the efficiency is not being passed to you. Expect to pay for the system, then a fraction per asset it produces.