A portfolio shows what an agency wants you to notice. Your job is to discover what it leaves out: awkward mobile layouts, inaccessible controls, slow templates, unclear responsibilities or concepts that never reached production.
When comparing web design agencies, treat each case study as an evidence file rather than a mood board. The strongest candidate is not necessarily the one whose work looks most like your preferred aesthetic. It is the one that can explain its decisions and demonstrate a working result under constraints similar to yours.
Start by establishing what actually shipped
Open the client website from the case study. Look beyond the homepage: visit a service page, an article, a product listing and whichever form or transaction matters to that business.
Ask the agency to distinguish between:
- A concept created for a pitch or internal exploration.
- Approved designs that another supplier implemented.
- A website the agency designed and built.
- A live site subsequently changed by the client or another team.
These are different kinds of evidence, not automatic reasons to reject an agency. A design-only practice may be exactly right if you already have engineering capacity. But it should not receive credit for implementation it did not perform.
Request the launch context, the agency’s scope and the names of any delivery partners. If the current site differs from the portfolio, ask for launch screenshots, a dated walkthrough or an archived version. These can explain the discrepancy, although a static image cannot prove that an interaction worked.
If a launch date, delivery role or result is absent from the agency’s case study, record **“not published”** and identify the case-study URL as the source checked. Do not convert missing evidence into an assumption.
Judge craft through ordinary tasks
Visual craft includes hierarchy, typography, spacing, imagery and consistency. Evaluate those qualities through a customer’s task, not just an attractive composition.
Try to understand the offer, compare alternatives and locate the next action without the agency narrating the experience. Then repeat the journey on a phone.
Look where polished mockups rarely linger
- **Content hierarchy:** Can you distinguish essential information from decoration?
- **Navigation:** Do labels make sense without knowing the company’s internal structure?
- **Responsive behaviour:** Do menus, tables, cards and forms remain usable as the viewport narrows?
- **Interaction states:** Are loading, errors, empty results and successful submissions understandable?
- **Content resilience:** What happens with long headings, missing images or dense information?
- **Consistency:** Do similar controls behave similarly across templates?
Ask why a design decision was made and what alternatives were rejected. An answer about audience needs or technical constraints is more useful than “the client wanted something modern.”
If your project involves complicated account journeys or product interfaces, include UX and UI agencies in the comparison. An attractive marketing homepage alone does not establish capability in those areas.
Check accessibility beyond the colour palette
Accessibility should be visible in the finished experience and in the agency’s delivery process. W3C describes its Easy Checks accessibility review as an initial assessment, not a comprehensive evaluation. Use a portfolio check to identify questions, not to certify compliance.
Navigate without a mouse. Check whether focus is visible, menus can be operated and controls have meaningful labels. Zoom the page and inspect whether content remains readable. Examine form instructions and error feedback without sending unnecessary submissions to a live business.
Use contrast requirements accurately
WCAG distinguishes minimum and enhanced text contrast requirements:
- Normal text at the minimum level requires a contrast ratio of at least **4.5:1**, subject to the criterion’s exceptions: W3C minimum contrast guidance.
- Large text at the minimum level requires at least **3:1**: W3C minimum contrast guidance.
- Normal text at the enhanced level requires at least **7:1**: W3C enhanced contrast guidance.
- Large text at the enhanced level requires at least **4.5:1**: W3C enhanced contrast guidance.
Use the linked guidance’s size and weight definition of large text rather than judging by appearance. Enhanced contrast is a different conformance target, not the default minimum. Passing these checks does not establish accessibility across the site.
Ask what accessibility target was agreed, who tested the implementation, what manual testing occurred and whether unresolved issues were documented. A useful response separates an automated report from a broader audit and explains who owns remediation.
Test performance without turning it into a scoreboard
Run representative live pages through PageSpeed Insights, including a content-heavy template and an important conversion journey. Record the tested URL and conditions so comparisons are interpretable.
Google explains that PageSpeed Insights combines lab and real-world field data. Lab testing helps diagnose issues in a controlled setting; field data reflects actual user experiences where sufficient data is available. Do not treat them as interchangeable.
Read the findings, not just the headline score. Ask the agency to explain oversized media, delayed interactions, shifting layouts and third-party scripts. If field data is unavailable, record that limitation rather than treating absence as a pass.
Responsibility matters here too. A client may have added tracking tools, chat software or unoptimised content after launch. Ask for launch evidence and the maintenance arrangement before assigning blame. Agencies offering web development services should be able to discuss implementation trade-offs, not merely say that the design files were fast.
Separate delivery proof from outcome claims
A shipped website proves delivery. It does not, by itself, prove increased sales or better lead quality.
For every claimed improvement, ask for the metric definition, comparison period, data source and relevant changes elsewhere in the business. Traffic acquisition, promotions, pricing and tracking changes can complicate attribution.
If evidence is confidential, request a redacted report or a reference conversation with permission. If results are absent from the case study, mark them “not published” and identify that source; do not estimate them from the client’s apparent success.
When commercial improvement is central to the brief, compare conversion optimization agencies as well. Ask whether research, measurement and post-launch experimentation are included or separately scoped.
Turn the portfolio review into a hiring decision
Keep a simple evidence ledger for each candidate: criterion, observation, supporting URL or document, agency explanation and unresolved question. Use labels such as **demonstrated**, **partially demonstrated** and **not published**, rather than inventing a precise score from incomplete information.
Request a walkthrough led by someone who would work on your project. Have them explain a difficult trade-off, show a less glamorous template and describe what the client can maintain independently.
For platform-specific work, ask about handover as well as appearance. When considering Webflow agencies, for example, request a demonstration of content editing, reusable components and the boundaries of routine client changes.
Finally, put the evidence into the scope: named responsibilities, accessibility acceptance criteria, performance testing conditions, supported templates, content ownership and post-launch fixes. Hire for demonstrated capability and a credible delivery process—not for the screenshot you remember most.