AI vs Hiring a Designer for POP Display Concepts

The question I get asked most often by owners of display shops is some version of whether they can stop paying a designer. The honest answer is no, and I say that as somebody with a commercial reason to say yes. I build AI POP Displays and I have worked in point-of-purchase for close to four years, so read all of this knowing where I stand.

What has actually changed is narrower, and more useful, than the headline. One stage of a designer's week got fast. The stages around it did not move at all, and several of them cannot move, because they are not drawing problems.

What does a display designer actually spend the week on?

Take a normal custom display job. A client email arrives with a paragraph of requirements and maybe a photo. Somebody reads it and works out what was actually meant. Somebody picks a format that suits the product and the retailer, and a material the client can afford at that volume. Somebody makes a concept the client can look at. Then the client changes something, and something else, and that loop runs three to six times. Then the approved concept becomes structural CAD, a dieline, a white sample, a full-color comp, and a quote with real tooling in it.

Of all of that, exactly one stage is a picture-making problem. That is the one AI took.

Everything else is judgment about a physical object that has to survive a pallet, a store associate assembling it without tools, and eight weeks of shoppers leaning on it. No image model has an opinion about any of that.

Which part does AI genuinely compress?

The brief-to-concept leg, and it compresses it hard.

Before, showing a client three directions meant three speculative builds in 3D, or three careful hand visuals, and a designer who had to decide up front which one to bet the hours on. That decision is where a lot of display work quietly goes wrong. You pick a direction, the client sees it, and their reply tells you they meant something else entirely. You have now spent a day finding out what the brief said.

With a model in the loop, you answer the ambiguity by showing it. Same brief, three readings of it, in the time the first one used to take. The client points at one, and the conversation starts from an image instead of from adjectives.

Two things make that work in practice, and neither one is the image model. The first is knowing which questions decide a display before you ask for anything, meaning sector, format, material, style, retail environment, product load and framing. The second is feeding real assets in as references, labeled by what each one is for, so the packshot gets used as product and the sketch gets used as structure. The mechanics are in how to prompt AI for retail displays, and the reason a bare prompt falls apart is in why generic image AI fails at POP displays.

If your real bottleneck is that you can only afford to show a client one direction per brief, that is the specific thing our free plan exists to unblock.

I should be straight about the engine underneath, because it matters to the argument. We run on Google's Nano Banana Pro, the model Google ships as gemini-3-pro-image. You can rent the same one. Google's own documentation puts it at 1K, 2K and 4K output, up to six high-fidelity object references in a single request, and a SynthID watermark on every image it produces. None of that is ours. What is ours is four years of knowing what to ask it for, plus the commitment that nothing you upload or render trains AI models, on any plan.

What does a concept actually cost today?

This is where most AI-versus-designer articles go wrong, including the ones currently ranking for this search. They put a per-render AI price next to a designer's hourly rate and declare a winner. In point-of-purchase that comparison barely applies, because the render is frequently already free.

Look at how display manufacturers sell. Plenty of them bundle 3D rendering into the quote package against a production order, and say so on their own websites. The buyer never sees a line item for the concept. That does not mean it was free to make. It means the cost is buried inside a salary and, more painfully, inside the pitches that never convert. Spec work for a pitch is unbilled, and most pitches lose.

For contrast, when a render is bought as a standalone deliverable outside this industry, the studios that publish prices are in a different world. RealSpace 3D's own pricing guide, updated in April 2026, lists product rendering at $250 to $900 per image with a two to four day turnaround. That is a billed deliverable with modeling, revisions, overhead and margin inside it.

Salary is a third number and should not be set against either. The US Bureau of Labor Statistics put the May 2025 median wage for commercial and industrial designers at $40.34 an hour, or $83,910 a year, and for graphic designers at $30.27 an hour, or $62,960. Those are employee wages, not what anyone bills a client, and quoting them as a price would be dishonest in both directions.

So the useful question is not what a concept costs. It is how many concepts one person can put in front of a client per week, and how many losing pitches you can afford to enter. That is the number AI moves.

For what it's worth, that is the arithmetic behind our own pricing, with 150 concept generations a month on the Pro plan at a founding $49 against a $69 list price.

What stays with the designer, permanently?

Four things, and none of them look like they are about to shift.

Manufacturability. A concept image is a proposal, not a structure. Somebody has to decide whether that cantilever holds a full facing of product, whether the base ships flat, whether the board grade survives the retailer's handling. Get this wrong and the factory quietly redesigns the display after the client has already fallen in love with the picture. That is the failure mode that costs real money, and a model has no way to catch it.

Production files. No concept tool on the market outputs a dieline, a board spec or a cutting layout. That work lives in dedicated structural CAD, run by people who build displays for a living.

The client relationship. Somebody takes the call where the retailer changes the footprint two days before the deadline. Somebody knows that this brand's marketing director hates rounded corners and always has. Somebody pushes back when a brief asks for a premium acrylic read on a promo budget. None of that is a rendering task.

Brand guardianship across time. A campaign is not one fixture. It is a fixture, a shelf tray, a totem and a window piece that all have to read as the same family, across markets, against guidelines somebody wrote two years ago. Consistency across a program is a human job.

If the concept stage is not the part slowing you down, the more useful read is how to write a POP display brief, because a thin brief poisons every stage after it.

Are companies actually buying less design?

Not according to the only recent buyer-side data I could verify. Clutch surveyed businesses that buy graphic design in March 2026 and published the results that April. In it, 88 percent said they use AI design tools, and only 18 percent said AI had reduced their need for designers. Forty-seven percent had increased their design budgets over the previous year. Asked what they weigh when choosing a designer, buyers put creativity first at 39 percent and strategic thinking second at 19 percent, with speed and affordability last, tied at 7 percent each.

Two caveats, because I would rather flag them than lean on the number. That survey covers graphic design broadly, not point-of-purchase, and the published release does not state its sample size or its respondent profile. Read it as a direction of travel, not as a measurement.

The direction it points is the one I hear on calls. Nobody is firing designers. They are asking those designers to answer more briefs.

Where has AI actually landed in the production layer?

It is worth looking at what the structural CAD vendors did, because they build for exactly the people this post is about, and their choices are informative.

Arden Software shipped Impact 2026 on 17 August 2026 with what it calls its first AI-powered capability, Design Similarity, which uses machine learning to find and rank similar past designs and manufacturing layouts inside your own database. That is a retrieval tool for a converter checking whether suitable tooling already exists, and it runs against your own data. Esko's ArtiosCAD, on release 26.07 from July 2026, markets itself explicitly for POP, POS and FSDU work and carries no AI feature of its own, though Esko has put AI into neighbouring products. EngView's Package and Display Designer shipped its 2026 version in January with an expanded parametric library and no AI claim, and it is the only one of the four publishing prices openly, at 760 euros a year for Lite up to 3,620 for Professional. AG/CAD's KASEMAKE does not mention AI anywhere on its site.

Notice what none of them attempted. Nobody in structural CAD is generating concepts, and nobody generating concepts is producing cutting layouts. The two halves of the job are converging from opposite ends and have not met. The designer sits in that gap, and it is not closing this year.

How should you actually split the work?

• Reading the client's email and deciding what the brief really is. Designer. This is where experience pays and no tool helps.
• Producing three or four concept directions to resolve the ambiguity. AI, with the designer choosing which ones are worth showing.
• Judging whether any of them can be built and quoted. Designer, every time.
• Revision rounds two through six. AI carries the visual iteration, the designer decides what actually changes.
• CAD, dielines, sample, tooling. Your engineering team, in their own software.
• The conversation where the client signs. Designer.

The way I would put it to a shop owner is that AI buys your designer back the hours they were losing to speculative rendering, and does nothing whatsoever about the hours they spend being a designer. If that trade is worth testing on the brief sitting on your desk, signing up costs nothing and comes with 15 credits.

For the wider picture of which tool fits which job, the tools roundup covers the category, and sketch to render for POP displays covers the case where your client already drew you something.