We've now completed thirteen documented client engagements — a mix of AI agent builds, e-commerce platforms, websites, and performance rebuilds. The work spans industries: industrial B2B, luxury real estate, cannabis beverages, beauty brands, community DTC. The problem statements were different. The budgets were different. The teams were different.
But the engagements that worked had something in common. And so did the ones that struggled.
The problem is usually not what they said it was.
A client who says "we need better SEO" often has a slow site. A client who says "our AI chatbot isn't working" often has no evaluation framework. A client who says "we need a new website" often has a conversion rate problem that a new design alone won't fix.
The presenting problem is real — it's just not always the primary constraint. In almost every engagement, the first two weeks involve discovering what's actually going on. That discovery shapes everything downstream. The engagements that skip it — that go straight from brief to execution — usually end up solving the right problem poorly because they defined it too early.
We've learned to treat the first discovery call as a diagnostic, not a sales conversation. The question we're trying to answer isn't "can we do this project?" but "what would actually move the needle here, and is that what's in the brief?"
Speed wins trust before anything else.
Across every engagement with a web component, the single metric that drew the most immediate client recognition was page load time. Not design. Not content. Not even conversion rate — though that followed.
There's something visceral about a fast site. Clients who'd been living with a 6-second mobile load time for years could feel the difference the moment the rebuilt version loaded in under 2 seconds. It sounds obvious, but most clients have normalised the slowness to the point where they'd stopped noticing it. The improvement is the first signal that the engagement is working.
This matters strategically, not just aesthetically. Speed is table stakes for everything downstream — SEO, paid acquisition, email click-to-purchase. You can't fix conversion until the site loads. You can't scale paid efficiently until the landing page works. Speed is the prerequisite.
AI works when the process is clear. It fails when it isn't.
The strongest AI agent deployments in our work — the ones that triaged thousands of leads, deflected the majority of support tickets, reviewed tens of thousands of listings overnight — all shared a common property: the process they were automating was already well-defined before the agent was built.
When a sales team knows exactly how they qualify a lead (criteria, scoring, handoff triggers), an agent can learn to replicate that process. When the qualification criteria are ambiguous and shift with the rep, an agent amplifies the inconsistency. The agent is not better at deciding what a qualified lead looks like — the team is. The agent is better at applying a well-defined decision tree, consistently, at scale, without sleeping.
This means the most valuable thing we do in many AI engagements isn't the engineering — it's forcing the client to articulate what "good" looks like before we start building. That articulation is the hardest work. The agent is the easy part once it exists.
Clients who have metrics before the engagement starts get more from it.
We've scoped projects with two types of clients. The first type arrives with a metric: "we need our checkout conversion rate to go from 1.2% to 2%." The second type arrives with a shape: "we want a better website."
The first type extracts more value from every conversation. They know when we've succeeded. They can make resource allocation decisions: if we spend an extra two weeks on checkout optimisation and recover 0.3 percentage points of conversion, is that worth it? With a number, that's a real question. Without one, it's a vibe.
We now ask every client to define their success criteria before we start. Not as a contractual formality — as a diagnostic. If they struggle to define what "done" looks like, that tells us something important about what we're walking into.
The best clients have a healthy scepticism about technology.
The clients who expected AI to be magic — who thought the agent would just figure it out — consistently needed more hand-holding and produced worse outcomes than clients who arrived knowing that AI is a tool, not an oracle.
The same pattern held for e-commerce. Clients who thought a new Shopify theme would fix their conversion rate needed to understand the infrastructure work that actually drives conversion before they'd make good decisions about where to invest. The ones who arrived with that understanding moved faster and made better calls.
This doesn't mean we turn away clients who don't know the technology. It means part of the engagement is always building that shared understanding — and the clients who are open to it, rather than resistant to it, get more done.
Tools don't fix broken processes.
An e-commerce platform doesn't fix a fulfilment operation that can't scale. An AI agent doesn't fix a sales process where nobody follows up. A new website doesn't fix a positioning problem that makes it unclear who the brand is for.
We've had to have this conversation with clients. Sometimes the technology is not the constraint. The constraint is the process, the team, the product, or the market. In those cases, we'll say so — and decline to build a system that won't move the metric they care about.
This is the "honest assessment" we keep coming back to: if we don't see a clear path from the build to the outcome, we say it before the contract is signed, not after the project is finished.
The one pattern.
If I had to distil thirteen engagements into one observation, it would be this: the projects that worked started with clarity about what success meant, and the ones that struggled started with a solution instead of a problem.
"We need an AI chatbot" is a solution. "We're spending $40k/month on support tickets and 60% of them are order status queries that don't need a human" is a problem. The second brief produces a better engagement, a more focused build, and a measurable outcome. The first produces a demo that looks impressive and performs mediocrely.
We're getting better at surfacing the real problem early. Every discovery call now starts with some version of: "Forget what you've built or what you want to build. What's the outcome you're trying to move, and what's currently stopping you from moving it?" The answer to that question determines everything else.