This is Part 2 of a 2-part discussion where we explored why most innovation fails before it starts, not from lack of ideas, but from lack of time to test them properly. We looked at the discipline required to pause operational processes, avoid premature builds, and create the conditions where experimentation can actually happen. Now, we’re looking at what happens when you actually build something small, and how to know whether it’s worth scaling… but first, see part 1: Making Space to Innovate.
Leon Calverley, Door4 founder, describes what happened: “The GA4 agent prototype – built in Retool – turned into a mini game-changer. We thought it’d be a nice internal demo. Instead, clients started asking if it could answer their own reporting queries. It unlocked a whole ‘agentic analytics’ narrative, now embedded in our positioning.”
“Show something real that saves minutes (or an hour) every time, suddenly they want it,” Leon says. “We’ve called it Fidelity.”
When you test small and solve real friction, the signal is unmistakable. People use it, ask what’s next, and change how they work.
Leon breaks down Door4’s approach: “Crawl: Retool prototype or basic chat-connected agent. One user. Manual input. Not much risk. No live data to overwrite! Walk: Lightweight integration. More users. Partial automation. Training and feedback loops from the Door4 team. Run: Full adoption, more UI polish, data feedback loop, maybe productisation.”
The crawl phase is unglamorous and scrappy. But it proves whether the idea works without requiring months of development.
“We use this everywhere – even internal tools follow that arc,” Leon explains. “If a thing doesn’t survive ‘crawl’, we don’t ‘push it uphill’.”
John O’Rourke, founder of GetConnect, recognises this caution:
“When you ask AI for content, you’re effectively asking it to create something based on thousands of other peoples’ press releases, so unless you teach it your own tone of voice, describing in detail just the way you’d teach an employee or freelancer, you won’t get authentic-sounding results.”
The effort to teach the system properly is the same effort required at crawl. Skip it, and you won’t get results worth scaling.
Leon identifies three signals that matter more than traditional ROI:
Saved time is easiest to spot. Changed behaviour is stronger—when someone adjusts their workflow, you know it’s solving a real problem. Commercial appetite is validation.
These signals emerge at crawl, exactly when you need them.
John’s warning extends beyond tone of voice: “Accessibility is something I’d specifically like to highlight when it comes to social media posts. Since AI trains on existing content, it will pick up on all the bad habits people have—for example, there’s a trend of using various emojis instead of bullet points, or even mathematical symbols instead of letters, which looks very fancy, but for a visually impaired user using a screen reader, might make your brand sound utterly confusing.”
“We need to care enough about our brand to look carefully at what trends and bad habits our AI assistants are learning from other people,” John says. “It’s your responsibility to ensure all your staff, robot or not, represent you well.”
Small experiments still need to meet the same standards as finished products.
At crawl: Does this solve the problem? If yes, move to walk. If not, stop.
At walk: Does it work for more people? Are they asking for more, or just tolerating it?
At run: You’re scaling something already proven.
Most businesses commit to run before validating at crawl. They build the polished version before knowing if anyone wants the rough one.
John’s approach offers a parallel: “Thinking about customer relationships, it’s easy to talk about relationships when you’re face to face, but developing this idea further, we can talk about where our customers go, what they’re interested in, and what work they do.”
Understanding context before committing resources is what makes small experiments effective.
Leon’s observation: “We’ve been working with GenAI since it basically became usable. And I’d say we’re past the novelty phase now. Most of the smart principals we deal with — boards, owners, senior leads — they get that the limitations aren’t really there anymore.”
The technology works. The constraint is whether businesses will create the conditions to test it properly.
“Content, analysis, admin, support — the shape of that work is different now,” Leon says. “Whether businesses adapt to that or not… well, that’s the choice.”
John’s counterbalance: technology changes, but the fundamentals don’t. Understanding your customers. Defining your outcomes. Representing your brand consistently.
The businesses getting innovation right are the ones willing to crawl first, measure what matters, and stop when the signal isn’t there.
Want to explore what innovation could look like for your business? See how we help clients test small, learn fast, and scale what works.
Thank you to our contributors for their time and insights.
Our latest articles, features and ramblings.
We explore performance marketing, AI, communications and optimisation.
“When you ask AI for content, you’re effectively asking it to create something based on thousands of other peoples’ press releases, so unless you teach it your own tone of voice, describing in detail just the way you’d teach an employee or freelancer, you won’t get authentic-sounding results.”
“Saved time – measurable friction removed, positive feedback – especially where we can spend MORE time with clients rather than in their spreadsheets. Changed behaviour – are people actually using it because it helps them? Commercial appetite – does the client ask ‘what’s next?’ with this thing.”