Published
July 15, 2026
Most founders evaluate design agencies the same way. They open the portfolio, look for a product in their industry, and shortlist whoever has the best screenshots.
Six months later the design system cracks the first time their own team designs a new feature. Knowing how to choose a SaaS product design agency means testing for what a portfolio cannot show you. This is what it looks like from the other side of those calls.
Every evaluation call runs on the same two filters. Does the agency have work in our industry, and do the screenshots look good.
Both feel like risk reduction. Both are aesthetic proxies for questions nobody asks out loud.
Domain match feels safest. If they have built a procurement tool, surely they understand procurement.
In practice, the agency that has built four tools in your category will build you a fifth that looks like the other four. You get pattern reuse sold as expertise. You launch into a market where your product is visually indistinguishable from the competitor you are trying to displace.
A screenshot is the output of a hundred decisions you cannot see, and it is the only part of the work you are allowed to evaluate.
Portfolio aesthetics fail differently. A screenshot is a static frame with clean, invented data. It says nothing about what happens when a table holds 4,000 rows, when a field errors, or when a role sees half the interface.
Those states are where SaaS products actually live. They are never in the case study.
Some buyers add a third filter and ask to see design options. That feels like getting more for the money.
It is the opposite signal. Three directions means the agency has no reasoned position and is handing the decision back to you.
None of this is dishonesty. Buyers optimize for the artifacts that are easy to compare, and agencies produce more of the artifacts that win work.
Andy Budd, co-founder of the design agency Clearleft, described what teams get when they hire off aesthetic showcases. The work arrives slick, polished, and fast.
"The eye of the stylist tends not to focus on deeper problems."
He said that more than a decade ago. The showcases changed. The hiring pattern did not.
When a founder says they want domain expertise, they mean industry. Fintech, healthcare, logistics, procurement. It is the wrong axis.
What transfers between products is not vocabulary. It is structure. How many roles see the same object differently, how deep the permission model runs, how dense the data is, how many states a single record can occupy.
Call that the complexity class. An agency that has designed a multi-role approval chain in procurement has learned something that carries into claims adjudication, grant disbursement, or trade settlement.
An agency that has designed four procurement tools has learned procurement.
Industry knowledge is the cheapest input a SaaS product design agency brings to the engagement. Your team already has it, and any competent designer absorbs enough in two weeks of user interviews. Structural experience takes years and does not come from reading your docs.
This changes what you look for in a portfolio. Stop scanning for your logo's neighbors. Start looking for products where multiple roles operated on shared data under conflicting mental models, whatever the industry on the label.
Then ask the agency to explain a structural decision from that work. Not what they designed. Why the alternative lost.
Here is the pattern we see most often, told as the pattern it is rather than any one client's story.
A funded SaaS team comes to us months after a previous engagement ended. They are not unhappy with what they were given. They will open the Figma file with something close to pride.
There is a color ramp, a type scale, spacing tokens, a component library with dozens of entries, a documentation page. It looks complete.
Then their team designed one new feature, and it broke.
Not visually. Structurally. The components had been drawn, not built.
No auto layout, so a button with a longer label overflows its container instead of growing to fit. No responsive constraints, so a card that is perfect at 1440 collapses at 1280. No state definitions, so there is one component for the default and nothing for hover, disabled, loading, error, or empty.
What they were handed was a sticker sheet. It looked exactly like a design system in every screenshot and it functioned like a folder of pictures.
"We design for the happy path, and society pays the price." - Jesse Weaver, designer and writer
Every missing state is a decision nobody made, deferred to whoever designs next.
So their designers did the only thing available. They detached instances and drew the new feature by hand.
Every detached instance is a fork. Six months on there are four button styles in production, three date pickers, and the team is asking whether to rebuild from scratch. That is what the previous agency's beautiful case study bought them.
The lesson is uncomfortable for buyers. The failure was fully present at handoff and completely invisible in the deliverable.
Every warning sign lived inside the component construction, one layer beneath a surface that looked finished.
None of this required more design talent. It required someone treating the file as a system a different team would extend, rather than as an artifact to be approved.
Ask for edit access to a real file from a shipped project, not a presentation. Select a button and check for auto layout.
Resize a card and see if the constraints hold. Open the variant panel and count the states.
Good looks like every component resizing and every state defined. What usually gets delivered is a sticker sheet that photographs well.
Every portfolio shot has four perfectly sized rows. Ask what the table does at 4,000 rows, what the chart does when a metric returns null, what the empty state says on day one, and what the error copy reads on a timeout.
An agency that has shipped real products answers immediately. One that ships mockups asks you to send a spec.
Give a candidate agency a flow and see whether they map the failure scenarios before anyone asks. Who sees this in a rejected state. What happens on a partial save.
If your product manager has to hand them the exhaustion list, you have hired hands, not a partner.
"There are no tiny features when you're doing things properly." - Des Traynor, co-founder, Intercom
Ask any agency to walk through a structural decision on past work and explain what they rejected and why.
Good looks like one recommended direction, the alternatives they evaluated, and the tradeoff they knowingly took. What usually gets delivered is three concepts and a request that you choose. If they hand you options, they are transferring the decision back to you.
When conversion is flat, most agencies propose a visual refresh, because a refresh is what gets bought. Describe your core user task and ask how many clicks it takes today.
The agency worth hiring starts asking about roles, navigation, and information architecture before it says a word about the interface. Low conversion is usually a structure problem wearing a visual problem's clothes.
Teams with dense platforms often ask for the entire workflow on a phone. Ask a candidate agency what they would cut.
The right answer names the two or three things that belong on mobile, approvals, alerts, a summary snapshot, and argues for leaving the rest on desktop. An agency that agrees to port the full workflow is agreeing to bill you, not to design for you.
The clearest confirmation shows up in the calls we lose.
A founder tells us conversion is down and points at a competitor's interface as evidence that theirs looks dated. We ask what the core task is and how many clicks it takes. Often the answer is five or six clicks for something a user does forty times a day.
That is not a visual problem. Fixing it means restructuring navigation and information architecture, which is slower, less photogenic, and nearly impossible to sell as a before and after. Some founders hear that and hire whoever promised the refresh instead.
Nielsen Norman Group has made a version of this point for years. Users fail on structure long before they fail on aesthetics, and teams misdiagnose the former as the latter with remarkable consistency.
The tell is what happens next. Teams under pressure bolt on a high-end feature, an AI assistant, a new dashboard, and ship it onto journeys that were already broken.
The new feature inherits the broken structure. Six months later conversion has not moved and the roadmap has one more thing to maintain.
Send this list ahead of the call rather than improvising during it.
Bottom line. When you choose a SaaS product design agency, the criteria that are easiest to compare are the ones that predict least. The work either survives real data and the next feature or it does not, and every question on that list exists to find that out before you sign.
If you want a design partner who expects you to open the file and check the components, that is how we work. See how we approach SaaS product design for funded SaaS and AI teams (/saas-product-design-agency).
