The KPIs worth using to evaluate an agency's past work (and the ones that tell you nothing)

If you're researching agencies for a rebrand, you're about to read a lot of case studies. Almost all of them will look impressive. Very few will tell you whether the work actually did anything.

That's not usually dishonesty. It's that most case studies are written as portfolio pieces, so they optimise for the wrong thing: the before-and-after screenshot, the award badge, the percentage with no baseline attached. None of that tells you what you need to know, which is whether this agency can diagnose a problem like yours and solve it.

The useful KPIs are different depending on whether you're evaluating brand work or website work, because those two things fail in completely different ways. Here's how I'd split it.

Start with the goal, not the number

A KPI without a stated goal is just a number. "Traffic up 260%" means something very different if the goal was investor credibility ahead of a funding round versus product sign-ups.

So the first question to ask about any case study isn't "what were the results." It's "what was this site or brand actually built to do, and who decided that?"

This matters more than it sounds. On every project I start by trying to understand what's currently failing and where the client wants to go, not what solution they think they want. That means asking about their ICP, their offering, and how easily their customers understand what they do. Most website problems turn out to be messaging and navigation problems. A common one: a company serving three distinct audiences tries to serve all of them on a single page, instead of building a clear path for each. That's obvious once you say it out loud, but it only surfaces if someone asks the right questions before anyone opens a design tool.

If an agency's case study can't articulate the original problem in a sentence, the results that follow are decoration.

Website KPIs that actually mean something

Website work is the easier half to measure, so hold it to a higher standard. What I'd look for:

Form fills or qualified enquiries, with a baseline. A percentage on its own is meaningless. Abatable's team reported website form fills up 54% after launch, and, more usefully, said the quality of contacts improved, so more relevant customers were getting in touch. The second half of that sentence is the part that matters. Volume is easy to game with a lower-friction form. Better-fit enquiries mean the messaging is doing work.

Traffic, with the context around it. Permission saw more than 260% growth within months of launch, before they'd fully activated their go-to-market efforts. That last clause is what makes the number usable. It isolates the site from the campaign that usually launches alongside it.

Time-to-publish for the internal team. How long does it take the client's marketing team to ship a new page after handoff? This is rarely in a case study, and it's one of the most honest indicators of whether the build was designed for the client or for the agency's portfolio.

AI visibility and share of voice. This one's new, and it's worth asking about because most agencies aren't tracking it yet. In the two weeks after Uniserve's new site launched, their visibility in AI search results rose 7.4 points and their share of voice rose 16.7 points against their competitor set. For B2B, where buyers increasingly start with an LLM rather than a search box, that's becoming a real acquisition channel.

Worth saying plainly: Uniserve's rebrand is also attracting better-fit clients and more leads than the previous site, but it hasn't yet converted into revenue growth. That's what an honest three-month report looks like. Be suspicious of any agency whose every case study ends in a clean revenue number.

Brand KPIs are mostly internal, and that's fine

Brand work is harder to measure and anyone who pretends otherwise is selling you something.

I treat brand and web as separate phases, because brand is solving perception. How it feels, what's right for that audience, whether the company is memorable and distinct from what already exists. That covers how they look and how they speak. Most of it is genuinely not quantifiable, and an agency claiming a hard number on brand recall for a Series A B2B company is usually inventing it.

What you can look for instead are internal signals, and they're more revealing than they sound:

• Can the whole team explain what the company does, the same way, without reaching for a deck? That usually traces back to clarity on a single mission statement.
• Does the sales team pitch faster and with less hedging?
• Are the leads coming in better qualified, because the positioning is filtering properly?
• Is the team visibly proud of it? Soft, but real. Teams that are embarrassed by their brand don't share it.

When you're on a reference call with an agency's past client, these are the questions to ask. They're harder to coach than a metric.

What to ignore

Awards. They're nice for recognition and some clients value them, but outside a handful of organisations they don't indicate anything about outcomes.

Percentages with no baseline. A 300% increase from four visitors is four more visitors.

Engagement metrics. Time on page, scroll depth, bounce rate. Occasionally diagnostic, never a goal.

Design-only case studies. If the entire case study is a set of beautifully art-directed mockups and no mention of what the client was trying to achieve, you're evaluating an aesthetic, not a capability. That might be all you want. Just be clear that's the trade.

The attribution question nobody asks

Rebrands never ship alone. They land alongside a funding announcement, a conference, a new sales hire, a campaign.

The right way for an agency to handle this is not to claim the whole result. I've built sites to land before an important conference, where the success clearly wasn't just the website. What the work does is make everything else land properly, so the booth, the messaging, the deck, and the site all feel like one coherent thing instead of four departments' guesses.

Ask an agency directly what else was happening when their headline result was recorded. The good answer names the other factors without being prompted. The bad answer is silence, or a number that conveniently arrived in a vacuum.

The KPI that reveals the most about an agency

Here's the one I'd ask every agency on your shortlist:

Six months after launch, what could the client's team do without you?

This is the question most agencies handle badly, because the honest answer is often "not much." And it exposes something bigger than post-launch support. It tells you whether the agency chose the platform for the client or for themselves.

Platform choice should come down to how a client wants to use their site and what resources they actually have. A company with no developers will struggle to own a complex custom build. A company with no designers can now ship things quickly with AI tooling, but risks drifting off-brand within a quarter. So I ask how a team operates day to day, and what will genuinely move the needle for them, before deciding what to build it in. Sometimes that's Webflow. Sometimes it's a custom build on something like Astro with Sanity behind it. The stack follows the team, not the other way around.

Whatever the platform, the deliverable should be a system, not a pile of static pages. A set of components the team can arrange to build new pages themselves and keep scaling within the brand. That's the thing that determines whether the site still works in year two.

So when you're reading through an agency's portfolio, check whether every single client got the same stack. If they did, either they've got a remarkably uniform client base, or the recommendation was made before the discovery call.

What to ask on the call

Pulling it together, the shortlist:

1. What was failing before, and how did you find that out?
2. What was the site or brand built to achieve, and who set that goal?
3. What's the baseline behind each number in this case study?
4. What else was happening in the business when that result landed?
5. What can this client's team do on their own now that they couldn't before?
6. Why this platform for this client, and what would have made you recommend something else?
7. What didn't work, and what did you change?

Question seven is the tell. Every real project has one. An agency that can answer it specifically has been through enough projects to have learned something. An agency that can't either hasn't done the work or isn't being straight with you.

You're not hiring an agency for the work in their portfolio. You're hiring them for the thinking that produced it, and KPIs are just the most reliable way to check whether the thinking was there.

See how I approach the work