Guide
Where AI is genuinely useful in promo, and where it is not
Two jobs in this trade suit a model well: putting a mark on a product photograph, and proposing where the mark goes. Almost everything else people put in a pitch deck does not.
- Good fit
- Decorating a product photo; proposing an imprint placement
- Poor fit
- Deciding what a client should buy, and promising a date
- The part that matters
- A person accepting or rejecting the output before it ships
Good fit
Rendering a mark onto a real product photograph
This is a well-shaped problem: there is a source image, a defined imprint area, and an artwork file, and the output is judged by eye in seconds. It replaces work that was previously done by hand for every colorway of every product in every collection, which is exactly the kind of repetition worth removing.
- One product, one imprint area, one artwork file — a bounded problem
- Wrong output is obvious immediately, so review is cheap
- The saving scales with colorways, which is where the manual cost was
Good fit, with a catch
Proposing where the decoration goes
A model can propose imprint regions on a product's photography far faster than a person can draw them. The catch is that a proposal is not a decision: the region is edited and approved by a person, and the approved one is what every later render uses. Skipping that step is how a whole batch ends up with a chest print on a sleeve.
- Proposals are fast; approval is what makes them safe
- An approved region is reused, so the product stays consistent
- Editing the region by hand has to remain possible, and easy
Poor fit
Choosing the assortment is not a model's job
What a client should buy depends on their budget politics, who is handing it out, what they did last year and what their CFO thinks of branded fleece. A model that generates a plausible shortlist without any of that produces something that reads well and lands badly. Product choice is the distributor's judgment, and it is most of the value you sell.
- Budget and audience context rarely exists in the data a model sees
- A plausible-looking shortlist is harder to argue with than a bad one
- The distributor's taste is the product; do not automate it away
Poor fit
Anything that amounts to a promise
Lead times, stock and delivery dates are commitments backed by suppliers and decorators, not by a model's confidence. Generating them is how a tool makes a promise its owner has to keep. If a system offers you a date, ask what it is reading.
- A generated delivery date is a liability with good grammar
- Stock and lead time belong to the supplier relationship
- Say what is known and stay quiet about what is not
Questions
How does MerchMaker use AI?
To propose imprint regions on a product's own photography, and to render decorated product images. A person approves the region and accepts each render before a buyer sees anything.
Does MerchMaker pick products for me?
No. You assemble the collection. The catalog gives you real listing data to do it against, including the supplier's minimum quantity.
What if a render comes out wrong?
It gets rejected with a reason and does not reach the collection. That review step is the reason generated imagery is usable in front of a client at all.
See the logo on the product first.
MerchMaker is invitation-only while we onboard distributors one at a time. Book a demo and we will set up your catalog with you.