Most GenAI portfolios have at least two initiatives that should have been killed months ago. Everyone knows which ones. Nobody wants to say it out loud.

This is the framework I use with leadership teams once the “honeymoon” phase is over. It is not about general adoption or whether people use ChatGPT at their desk. It is about funded initiatives with a business case attached that should be delivering measurable economic impact.

The matrix classifies initiatives on two axes: whether there is a real P&L effect, and whether the initiative runs at meaningful scale. That combination determines what to do next.

Kill: pilots that never had an economic spine

Low scale, no defensible economic mechanism. Think the dreaded “CompanyGPT” initiative built by IT to get acquainted with LLMs. Nobody is using it and people reverted back to ChatGPT or Claude months ago.

These are rarely killed on their own, because nobody is harmed by letting them sit there. The cost is the licence, the attention in every steering meeting, and the impression that something is being done.

Rescope: rolled out broadly, value stays diffuse

Scale is real, but the economics are not. A GenAI proposal automation tool approved to “increase sales productivity,” now deployed across multiple teams. But win rates have not changed, external bid support is still in place, and hiring plans were never adjusted. Re-anchor it to a bottleneck with an actual cost line.

The tell is that everyone likes it and nobody can name what changed. Usage is healthy, feedback is positive, and no number in the plan moved because of it.

Alter: it works, but it is not the default path yet

The economics are clear. Uber has described a GenAI invoice processing system in finance ops that reduced manual handling time and improved processing efficiency. But the question is whether AI owns the default path across the enterprise. If it is limited to one geography or still requires full human review on the happy path, it sits in Alter.

Moving out of Alter is usually an organisational job rather than a technical one. The system already works. What is missing is the decision to make it the normal route and to stop maintaining the old one alongside it.

Scale: embedded in operations with measurable displacement

Deployed across real volume and tied to metrics that directly affect cost-to-serve. ING built a GenAI customer service chatbot that served 20% more customers within seven weeks, with a blueprint to scale across ten markets and 37 million customers. When the economic effect is visible in cost structure and the initiative runs as infrastructure, it sits here.

Three things that cut across the quadrants

Three considerations cut across all quadrants: whether the initiative creates competitive advantage, whether to buy or build, and whether the organization should invest in a dedicated GenAI team:

• An initiative with weak economics might still warrant investment if it builds a defensible capability.

• A custom-built solution in Kill might belong there because a buy option would have delivered the same result at lower risk.

• And if most of your portfolio sits in Kill, the case for an internal GenAI team looks very different than if you have Alter and Scale candidates.

That last one is worth sitting with before you hire. A team brought in to run a portfolio that is mostly Kill will spend its first year rescuing work that should have been stopped, and you will conclude that the team was the problem.

Finishing what you started

The question is whether you have the discipline to finish what you started, rather than looking for new ways to do more with AI.

A useful habit is to place every funded initiative in a quadrant once a quarter, with the same people in the room. The movement between quarters tells you more than the position does.

Curious where your portfolio sits.

Originally posted on LinkedIn