Quick wins or strategic bets? How we score roll-out difficulty and added value
The opportunity matrix plots every eligible solution by how hard it is to roll out and how much value it adds — derived transparently from data we already hold, and gated by confidence.
The catalogue can tell you what exists. The harder question for anyone planning an AI roadmap is where to start. The opportunity matrix on our analytics page answers exactly that: it plots every eligible solution on two axes — how much value it adds, and how hard it is to roll out — so the easy, high-impact moves separate visually from the expensive, speculative ones. Here is precisely how both scores are calculated, so you can trust the placement rather than take it on faith.
Two axes, four quadrants
The horizontal axis is added value (left = low, right = high). The vertical axis is estimated IT roll-out difficulty, and we deliberately flip it so that low difficulty sits at the top — the easiest things to adopt rise to the surface. That gives four quadrants:
- Quick wins (top-left): easy to roll out, modest value. Good for building momentum.
- Sweet spots (top-right): easy and high value. Start here.
- Value drivers (bottom-right): high value but harder to roll out. Worth the investment, with eyes open.
- The bottom-left — low value, high difficulty — is intentionally left unlabelled. If something lands there, reconsider.
Why we derive the scores instead of asking for them
There is a manual "estimated IT difficulty" field in our schema, but only a fraction of entries ever carry it — and a chart that plots a handful of solutions is not a market view. So instead of depending on a field that is mostly empty, we derive both scores from data we already hold for almost every entry: the development stage, the capability and adoption scores from our maturity model, the business criticality, the value-chain footprint, and more. The result is that most of the catalogue is plotted, not a lucky few.
How we score roll-out difficulty (the vertical axis)
Difficulty is a 1–5 score built mainly from three signals that nearly every entry carries, then refined by sharper fields when they exist. The logic: a proven, productised solution is easy to adopt; an unproven proof of concept is not — and more advanced, more business-critical AI demands more integration, governance and testing.
| Signal | What it tells us | Relative weight |
|---|---|---|
| Development stage | Scaled = proven & easy; MVP = mid; proof of concept = highest risk | high |
| Capability maturity (CMS) | More advanced AI capability = more to integrate | medium |
| Business criticality | Mission-critical systems need more oversight to deploy | medium |
| Manual IT-difficulty estimate | A human's direct read, when present | high |
| EU AI Act risk class | Higher-risk use cases carry more compliance burden | low |
| Automation level | Deeper automation = deeper system integration | low |
| Time to value | A longer road to value signals a harder roll-out | low |
We don't require all of these. We take a weighted average of whichever signals are present — and usable, see the confidence gate below — so a solution scored only by stage and maturity still gets a fair number, and one with a human estimate leans on it.
How we score added value (the horizontal axis)
Value is also a 1–5 score, anchored on the Business Adoption Score (BAS) — one of the indices from our AI capability maturity model — and refined by the automation level, the breadth of value categories and value-chain steps a solution touches, and any recorded quality or competitive-advantage gain.
We won't re-derive the whole value story here, because the underlying indices — the capability hierarchy, the Capability Maturity Score and the Business Adoption Score — are already explained in depth in the maturity model itself. For the full picture of what "added value" rests on, see From AI use cases to AI capabilities: a maturity model.
The confidence gate: no value from a shaky field
A score is only as honest as the data under it. So every field is filtered by its confidence rating before it is allowed to contribute. We use the same three-level confidence that runs through the rest of the catalogue:
- Green / orange (verified or interpreted) — the field may be used.
- Red (uncertain, or never assessed) — the field is ignored entirely.
And a solution is only plotted when each axis has at least one usable field. If everything determining its value, or everything determining its difficulty, is red, the solution stays off the chart rather than being placed on a guess. Those confidence levels themselves come from how we rate the underlying sources — explained in How we rate our sources: the Admiralty Code.
What the matrix shows today
With this approach the chart plots the large majority of published solutions, up from the handful that carried a manual estimate. You'll notice the cloud leans toward the top-right — easy, valuable "sweet spots." That is not a quirk of the formula; it reflects reality. Most of the AI solutions documented in the Swiss insurance market today are scaled, productised deployments — proven, and therefore genuinely low-risk to adopt. The few proofs of concept correctly sink toward the harder, lower rows.
Because the scores are derived on read, not frozen into the database, the picture sharpens automatically as entries gain better data and firmer confidence ratings — no migration, no manual recompute. The matrix is a living read on where the easy wins and the strategic bets actually are.