Feasibility questions to ask before you quote
Unprofitable jobs are visible on the discovery call. Fourteen questions that surface specific action, exact geometry and likeness risk before you price.
Go back through the jobs that lost money and the pattern is boringly consistent. Almost none went wrong because the client was difficult or the deadline slipped. They went wrong because something in the brief was much harder to generate than it sounded, nobody tested it before the price was agreed, and by then the number was already in an email.
That pattern is detectable early. Three shapes account for most of it, and each has a question that surfaces it in about fifteen seconds on a call.
The three shapes that blow up a quote
Specific character action. Not "a woman in a kitchen" — "she picks up the jar, twists the lid off, and pours". Attempt counts rise sharply here. Community-reported figures put general usable output at roughly one clip in four, or three to five attempts per usable clip, and the same reporting notes the number climbs for specific character action and complex motion. Those are aggregated anecdote rather than measured data, but the failure mode is specific: contact, weight and object handling are where generated motion falls apart. Physics failures in generated motion is the diagnostic detail.
Exact product geometry. A hero shot of a bottle you have a reference for is fine. A hero shot where the label text has to be legible, the barcode has to scan, the dieline has to match the real pack, or the logo has to be the actual logo rather than a plausible cousin — that is a different job. What models draw well and what needs a designer and barcodes and legal copy on generated packaging both cover where the line falls.
Guaranteed likeness. A recognisable person, appearing consistently, doing specified things, cleared for the usage. Every word there adds cost. Consistency mechanisms exist but fail in different conditions, and "guaranteed" is a word you should not put in a quote — four consistency mechanisms and when each fails is why.
The question list
Run these on the discovery call, in this order, before any number is discussed. Most take one exchange. The whole list is about twelve minutes.
On action and motion
- Walk me through what happens on screen, second by second. (Vague answers are good news. Precise ones are the expensive ones.)
- Does anyone need to pick up, hold, hand over or operate anything?
- Is there a specific gesture, expression or piece of choreography that has to read?
- Does anything need to happen in a single unbroken take, or can it be cut?
On products and geometry
- Does the real product appear, or a representation of it?
- Does any text have to be legible on screen — label copy, packaging, a UI?
- Is there a barcode, regulatory mark, ingredient panel or legal line that has to be correct?
- Do you have reference images from every angle we need, or only the pack shot?
On people and likeness
- Is there a named or recognisable person in this?
- Does the same person need to appear across multiple shots, or multiple assets?
- Who holds their consent, and does it cover synthetic depiction and the territories you are running in?
- If we cast a generated character instead, is that acceptable to the brand?
On the frame around the job
- Where does this run, in which territories, for how long, and is there paid spend behind it?
- Who signs off, and is there anyone above them who will see it for the first time at the end?
Question 14 is the highest-yield one and the one most people forget. A job with an invisible final approver is one where the direction gets reopened at the worst possible moment. It belongs in the decision ladder you scope against.
Scoring the answers
You do not need a model. You need to know, before quoting, whether any answer is red.
| Signal | Green | Amber | Red |
|---|---|---|---|
| Action | Ambient movement, no object handling | One clear action, cuttable | Precise multi-step handling, single take |
| Geometry | Product suggested or stylised | Real product, references from key angles | Legible label copy, scannable codes, exact dieline |
| Text on screen | Added in the edit | Short, large, generated | Paragraph of legal copy, generated |
| Likeness | Generated character, no real person | Real person, one shot, consent held | Named talent, multi-shot, consent unclear |
| Consistency | Single asset | One character across a few shots | Same face and product across a campaign |
| Approval | One named signer | Two, both on the call | Unknown final approver |
| Exposure | Organic social | Paid social | Broadcast, regulated category, packaging that goes to print |
Anything amber, buffer against it. Anything red, do not quote yet.
Two ambers interacting is quietly worse than one red. Real product plus real person plus multiple shots is the combination that eats studios, because the failure modes compound: fixing the product angle breaks the face, fixing the face breaks the product. Multi-reference editing to lock product, face and set at once is the technique for it, and knowing you will need it before you price is the point of asking.
What to do with a red
Three moves, in order of preference.
Test before quoting. This is the one that actually resolves it. Take the hardest single shot, spend a modest number of credits, and run it. Not the whole piece — the shot that scared you. Twenty minutes of generation tells you more than an hour of estimating, and you can compare several models in one pass since the agent chat can fan a single prompt across named models in a single request. A/B testing AI models with one prompt is the method, and estimating credit cost before you dispatch a batch keeps the test from being a surprise.
Bill the test if it is substantial. A paid feasibility sprint sells easily when framed honestly: "there's one shot here I want to prove before I put a number on the whole thing, and I'd rather find out now than in week three."
Reshape the brief. Most reds have a green-adjacent version that satisfies the actual objective. The jar-pouring shot becomes a cut: hand approaches jar, cut, contents in the bowl. Legible label copy becomes an overlay added in the edit rather than generated into the frame. A named face becomes a consistent generated character — stock avatar, digital twin or generated character covers that trade. Clients usually accept these when offered as a choice with reasoning attached, because they care about the objective, not the method.
Price the uncertainty into the structure, not the number. If the red survives, quote a base plus a change-order rate rather than a padded fixed price. A fat fixed price loses the job to someone who did not do the analysis; a base with a stated trigger wins it and stays defensible. Fixed-price or cost-plus quoting for video covers where each structure fits.
Where the numbers to buffer against come from
The published hit-rate figures are not benchmarks and you should not price off them. They are community anecdote — useful for direction, useless for arithmetic on a specific job.
What is usable is your own log. Record attempts and keepers per shot type, and inside a couple of months you have a real number for your briefs, your models and your standards, which is the only version that predicts anything. Usable rate is the benchmark nobody publishes makes the case for logging it, and reroll rates and the shots you throw away is how the number turns into a buffer. Nobody has published measured hit rates, so instrumenting your own is the difference between feasibility judgement and feasibility guessing.
One habit that pays for itself: after every job, add one question based on whatever surprised you. The list is supposed to grow. A studio's intake questionnaire is the compressed record of everything it has been burned by, and it is the cheapest asset you own.
FAQ
Won't this many questions make me look like I can't do the work?
The opposite, in practice. Most clients have been through a project where nobody asked and it went badly, and a supplier who probes the hard parts on the first call reads as someone who has done this before. The tone that works is curious rather than defensive: you are working out how to make it good, not building a case for why it is hard.
What if the client doesn't know the answers?
That is a finding, not a dead end, and it usually means they have not yet had the internal conversation that produces the answers. Send the list, ask them to come back on it, and note that the quote follows. A brief that cannot answer question 11 or question 14 is not ready to be priced, and saying so early is far cheaper than discovering it in week three.
Should I run the feasibility test for free?
For a small job, yes, treat it as sales cost. For anything with real production behind it, quote a paid feasibility sprint with a fixed fee and a written output: which shots are proven, which are risky, what the recommended reshape is. A client who will not pay a small amount to de-risk a large amount has told you something useful about the rest of the engagement.
Does this list change for images rather than video?
Drop the action questions and weight the geometry ones harder, because a still holds up to inspection in a way a moving frame does not. Add one: will this go to print or a marketplace? Both are real constraints — marketplace rules on generated product shots covers the second, and it disqualifies more jobs than people expect.