That theme came through loud and clear at our recent roundtable with Google Cloud, where measurement emerged as one of the biggest challenges (and opportunities) for teams trying to build credibility and unlock momentum.
What gets measured gets funded
Ed Rees, Google Cloud Director at Valtech, says value lives in two domains:
-
Frontstage (customer-facing). Measure value with metrics like click-through rates, conversions and average order value.
-
Backstage (internal efficiency). Look for gains in productivity, time saved and process accuracy.
The challenge? Frontstage metrics are easier to track and sell. Backstage value can be harder to define and even harder to prove.
Ed’s advice: Define what matters and how you’ll measure it before implementation, or risk misalignment from day one.
Embed measurement from the start
Matthew Hildon, Valtech’s European Retail Director, backed this up with real-world examples:
-
At Matalan, AI supported the copywriting team, improving both output and on-page SEO performance.
-
At Not On The High Street, AI-powered search using Google Cloud tools led to more relevant results and better conversion.
These weren’t speculative pilots. They were A/B tested, performance-tracked and refined in production. Measurement was baked in from the beginning.
Measurement doesn’t end at go-live
Alvaro Silva-Santisteban, AI Solution Lead at Google Cloud, shared a powerful post-launch story. A financial services chatbot went live with 45% containment. Through continuous testing — including small tweaks like adding delays to make responses feel more “human” — the team pushed that number to 65%.
The takeaway: Go-live isn’t the end of measurement. It’s the beginning of iteration. Great systems evolve through experimentation.
Build evaluation into the architecture
Alvaro also shared how more mature organizations are building “evaluation harnesses” directly into their AI systems. These include:
-
Performance monitoring: uptime, throughput, grounding accuracy
-
Quality metrics: customer satisfaction, resolution rates
-
Safety checks: hallucination filters, human-in-the-loop processes
This kind of infrastructure ensures AI systems are measurable, trackable and trustworthy from day one.
Measurement builds confidence
A recurring theme from the group: Measurement is about building trust for different stakeholders.
-
For teams, it demystifies AI and shows what’s working.
-
For leadership, it supports prioritization and investment.
-
For customers, it leads to better, more consistent experiences.
Matt also reminded us that not all data is quantitative. When a content team says, “this changed our lives,” that’s a signal, too — and likely a sign you’re moving in the right direction.
If you can’t prove it, you can’t scale it
For AI to move beyond the innovation bubble, it must speak the language of business: results. That means measuring what matters, early and often.
Start small. Prove it. Share it. Scale it. That’s the playbook.
Contact us to learn how we can help you put it into action.
