Our approach
The AI Value Stack
Thousands of AI combinations. One Matrix to choose. For boards, CEOs, and the CTOs who advise them, we choose hardware, models, and applications together for each business decision, by the value it creates and the risk it carries.
Too many paths, and one easy default.
Every AI system rests on four layers: the hardware it runs on, the model that does the reasoning, the application that puts the model to work, and the business process it serves. Each layer now offers dozens of credible options. Our own public sample alone gives more than ten thousand combinations: 13 accelerators, 19 models, 11 application roles, and 4 Matrix zones, before counting cloud providers, versions, or the more than 100 models in our full register.
Faced with that many choices, most firms take the simpler path: one default at every layer, often the combination a single provider offers end to end. Providers naturally recommend what they sell, and the simpler path can be the right answer. But it is rarely the right answer for every process in a firm, and its costs tend to arrive later, when changing course has become expensive.
We map the options at every layer and find the combination that fits each decision, independently. We guide the choice. Your teams, or the partners you select, build it, and we stay independent of all of them: we do not resell any chip, model, or cloud.
Guided by the Vector & Value Matrix.
Each business process is placed in a Matrix zone first. The zone sets what every layer below it should prioritize, so one firm usually has several optimal paths, not one.
| Zone | The question it asks | What it means for the stack |
|---|---|---|
| Amplify | What happens to human expertise? | The strongest models, assisting people; quality matters more than cost. |
| Automate | Who captures the value? | Proven, low-cost models at volume, often open-weight, on hardware sized for throughput. |
| Assure | Who is responsible when AI fails? | A checker from a different model family, full decision records, and a person who signs off. |
| Authorize | Who owns the inputs and outputs? | Permissive licenses and hardware under the firm’s control, so data, rights, and records stay with the firm. |
Read more about the zones on the Vector & Value Matrix page.
The economics behind the method.
Choosing an AI stack is an old economic problem: interdependent choices, made under uncertainty, where testing is costly and some decisions are hard to reverse. Our method follows the economics.
- 1 · Frame
Choose the layers together.
Complementarities · Milgrom and Roberts
- 2 · Shortlist
Test a few, in the right order.
Optimal search · Weitzman
- 3 · Choose
Most value for the risk carried.
Efficient frontier · Markowitz
- 4 · Protect
Keep the right to switch.
Real options and modularity · Dixit and Pindyck; Baldwin and Clark
- 5 · Improve
Keep testing a small share.
Explore and exploit · Gittins
Why the simpler path tempts. With thousands of options, firms settle for good enough (Simon), and its costs arrive later, as lock-in (Klemperer; Arthur; Williamson).
Detailed references, scoring indices, and the full method are shared with clients during consulting engagements.
Our work is decision support and does not constitute legal, regulatory, financial, or investment advice. Every engagement begins with a mutual non-disclosure agreement.
Find the right path for your firm.
An AI Value Stack review ends with a board-ready recommendation: the path for each process, the fallback, and the evidence behind them.