Evaluate a creative technology project by agreeing the business outcome before work starts, choosing a small set of measures that connect to it, and being honest about which results you can prove and which you can only infer. Hybrid projects, such as an AI-assisted film plus a new landing page plus an automation, rarely have one clean number, so the framework matters more than any single metric.
Start with the outcome, not the asset
Most disappointing evaluations come from measuring what is easy (views, downloads, page loads) instead of what the project was for. Write one sentence per project: “We want [audience] to do [action], which supports [business goal].” If you cannot write it, the project is not ready to be measured, and it may not be ready to be built. The AI return on investment framework covers how to frame benefits and costs without hype, and it applies to hybrid projects too.
A four-layer measurement stack
Hybrid projects touch several layers. Measuring each layer separately stops a strong result in one area hiding a weakness in another.
| Layer | Question | Example measures |
|---|---|---|
| Delivery | Was it built to brief, on budget and on time? | Scope delivered, revision rounds, hours against plan |
| Experience | Does it work well for real people? | Page speed field data, completion of key tasks, accessibility checks |
| Behaviour | Did people do the thing we wanted? | Enquiries, sign-ups, purchases, returning visits |
| Business | Did it move the goal? | Revenue influenced, cost saved, time returned to staff |
Delivery and experience measures are largely within your control. Behaviour and business measures depend on markets, seasons and other activity, so treat them with more caution.
Choose measures you can actually collect
Experience: use field data
For websites, real-user performance data is more representative than a single lab test. Google’s guidance on Core Web Vitals says that only field measurement can capture the complete picture, because performance varies with device, network and how people interact. It sets targets for loading (Largest Contentful Paint within 2.5 seconds), interactivity (Interaction to Next Paint of 200 milliseconds or less) and visual stability (Cumulative Layout Shift of 0.1 or less), assessed at the 75th percentile of page loads. Our Core Web Vitals business guide explains what these mean in plain terms.
Behaviour: define the key action
In Google Analytics 4, a key event is an event that measures an action particularly important to the success of the business, and any event you collect can be marked as one. The discipline is choosing very few: a form submission, a booking, a purchase. If everything is a key event, nothing is. If the project includes a film, decide which action the film should drive (a click to a product page, a form start) so it can be tracked rather than guessed at.
Business: be explicit about attribution
Creative work often influences decisions that happen later or on another channel. Report two columns: results you can directly attribute (a tracked enquiry from a campaign page) and results you can only associate (a rise in branded searches after launch). Our guide to measuring creative campaign performance goes deeper on moving beyond view counts.
Set a baseline and a review date
Without a “before”, you cannot show change. Capture the baseline for each chosen measure before launch, then set a review date that suits the sales cycle: a consumer promotion might be judged in weeks, a considered B2B purchase in quarters. Note anything else that changed in the period, such as a price change or a seasonal peak, so you do not credit the project for it.
Count the whole cost
Include creative production, development, review time inside your own team, tools and subscriptions, and ongoing maintenance. AI-assisted work can lower some costs but add review and rework, which is why the hidden costs of AI content production belong in the denominator.
Checklist before you sign off a project
- One-sentence outcome written and agreed.
- Three to five measures across the four layers, each with an owner.
- Baseline captured and stored.
- Key action tracked and tested before launch.
- Review date set, with a note of other changes in the period.
- A clear statement of what the data cannot tell you.
Limits to accept
Small businesses often lack the traffic for statistically firm conclusions. That is acceptable if you say so and treat results as directional. Qualitative signals, such as the questions prospects ask or sales feedback, are legitimate evidence when labelled as such.
Where Radiance Impact fits
Radiance Impact combines creative production, digital technology and marketing, so measurement planning can be shaped across the whole project rather than bolted on afterwards. To discuss a project, contact us, or see how the Technology division approaches websites and automation. For the wider approach, read about Radiance Impact.
Sources
- Web Vitals, web.dev (Google)
- Key events in Google Analytics 4, Analytics Help (Google)
