28 APR 2025 | 12 min read

Not all projects contribute to value in the same way. If you have a proven business model and make relatively few assumptions, you can comfortably anticipate the value which will be created for given levels of investment. This is why so many AI projects today don’t aim at radical results. They are centered on efficiency or taking cost out, but the business model remains relatively stable and AI is merely an incremental improvement. Think of law firms using AI to respond to RFP’s, or public relations firms using the technology to write press releases. The fundamental structure of how they deliver work for clients doesn’t change, nor does their business model — at least for now.

Far greater ROI comes from the high-uncertainty spaces that represent new business models, entirely new capabilities, and potentially disruptive innovations. You can’t calculate value in the same way that you can for existing lines of business. With these, you are instead building option value. A real option is a bet you place today that buys you the right, but not the obligation, to make a follow-on investment. Simply put, if your bet is in the money, you can make the future investment. If your option didn’t work out, you can stop. All you lose at that point is the (hopefully) modest amount that you spent to create the option. You only need to commit to the next step in the learning process and can re-assess the prospects of the business at each step.

Research has shown that the ROI on investing in options is greater per unit of investment than investing in the core business — but they require very different approaches. It’s the way venture capitalists think — they’re willing to tolerate high failure rates because they manage the cost of failure, and the few winners win really big.

An example of AI making a game-changing difference is the way Kraft-Heinz manages its entire supply chain using digital twin technology. When an event happens in real life — say a port is closed or a supplier doesn’t deliver on time — the real data can be brought into the simulated supply chain. The system then computationally runs through dozens, even hundreds of scenarios in real time and recommends the course of action that best meets the company’s objectives — a task that would be impossible to do with human brains alone.

Does this sound familiar? Your strategy around AI is trying to pull you into the future. The budgets for it are holding you back as existing lines of business resist giving up their resources. There isn’t a clear process for governing AI projects, with the result that the portfolio may not be connected to either strategy or budgets. And your people are still rewarded for delivering near-term results. The result is that the management of your AI project portfolio can be a bit of a mess. A good first step is to figure out what your organization already has going on.

Imagine two kinds of uncertainty, uncertainty about future markets and uncertainty about technology and capabilities.

Figure 1: A Portfolio Framework Incorporating Uncertainty

Juxtaposing these, as in Figure 1, gives us a quick visual way to see where our projects are. Projects in the core area have low uncertainty on either dimension. Real applications with real benefits have been developed. In the case of AI, many applications from fraud prevention, analyzing medical imagery, gaming, analytics and more already exist and are being used today.

Projects in the platform area are being prepared to become part of tomorrow’s core business. Alphabet’s Waymo autonomous car project is this phase – it isn’t yet large enough to make a significant difference to that company’s bottom line, but it is in-market and demonstrating that it can add value. Indeed, a recent study done in conjunction with Swiss Reinsurance found that Waymo vehicles were far safer than human drivers, a vitally important point to its value proposition. Waymo is partnering with Uber as the user platform, potentially creating a new platform for both companies.

When we get to the outer edges of the model, now we are in the world of options, where you are placing small bets on future hypotheses or experiments. Most of the time they won’t work out, but you can think of these as keeping you in the game — quite literally keeping your options open — so that you aren’t taken by surprise as the technology evolves.

Positioning options are investments when you believe many customers have a need, but it is unclear what the winning solution will be. Consider the future of remote communication — will it be holograms? Flexible surfaces? Chatbots that go to our boring meetings in our stead? Nobody knows, but it’s smart to plant a few seeds and see which might flourish, realizing that most of them will need to be culled.

Scouting options involve the major uncertainty of taking something you think you know how to do into a new market. A lot of AI applications are in this space right now — we’ve invented some whiz bang app or solution and now we have to find out if anyone wants it. Key to success in scouting options is to have an experimental mindset, bearing in mind that an experiment that doesn’t go the way we hoped is not a failure — just an unsupported hypothesis.

Stepping stone options are the most uncertain of all. The idea is to find a real group of customers that has a problem that can’t be solved without AI. The introduction of autonomous driving technology in the military is an example. By replacing human drivers in trucks that can play follow-the-leader, the AI technology reduces risk for the humans, saves time and allows for more precise management of transport fleets than is possible in an analog way.

The general rule of prudent investing is that 70% of the available resources (people and money) go into the core, 20% into platforms and no more than about 10% into options. Ironically, research shows that ROI goes the other way — a dollar invested in the core returns about 10%, in platforms around 20% and in options, if they are properly managed, 70%.

What a map like this allows you to do quickly is figure out what your organization is actually working on. In far too many cases, the portfolio is a ragtag assortment of people’s passions, the pet bunny from two CEO’s ago and stuff that just isn’t making progress. If your portfolio analysis ends up looking something like the one in Figure 2, you really can’t claim to be investing the capital in AI strategically.

Figure 2: A Portfolio Framework Incorporating Uncertainty

How to score the levels of uncertainty? Here are some ideas.

How confident are you about the following?

1 = certain

2 = relatively uncertain,

3 = highly uncertain

Market Uncertainty

• Clarity of AI value proposition (how clearly the AI solution addresses specific customer pain points)
• Customer willingness to pay for AI-generated outputs versus traditional solutions
• Data rights and ownership model (who owns the training data, inference data, and outputs)
• Competitive moat (barriers preventing competitors from replicating your AI capabilities)
• Revenue model sustainability (subscription, usage-based, outcome-based, or freemium)
• Regulatory landscape for your specific AI application domain
• Go-to-market strategy (direct sales, platform partnerships, API ecosystem)
• Customer trust and adoption readiness for AI in your specific domain
• Ability to scale the business model without proportional cost increases
• Potential for leveraging network effects through data accumulation

Score responses as follows:

10-16: Low market uncertainty

17-23: Medium market uncertainty

24-30: High market uncertainty

Technical Uncertainty

• Data acquisition strategy and data quality assessment
• Compute requirements and scaling strategy as usage grows
• Model performance benchmarks versus customer expectations
• Your firm’s ability to efficiently develop, deploy and maintain models in production
• Technical debt management approach for ML systems
• Ability to detect and mitigate model drift over time
• Feedback loop implementation for continuous improvement
• AI security and vulnerability management
• Integration complexity with existing customer systems
• Technical team composition and critical expertise gaps

Score responses as follows:

10-16: Low technical uncertainty

17-23: 17-23: Medium technical uncertainty

24-30: High technical uncertainty

With the picture of your “as is” portfolio in hand, you can now start to evaluate both its health and what that implies for other investments you might be asked to make. To prioritize, see if the portfolio you are developing will support your strategy.

AI necessitates a major re-evaluation of existing strategies. Consider the unfortunate impact it has had on Edtech pioneer Chegg. The company was — oddly enough — famous for “helping” students improve their homework grades by relying on armies of workers in India. It believed that AI would not harm their business because of hallucination problems. It turns out that students are more than willing to deal with a little hallucination when the service is free. Chegg’s stock dropped over 90% between 2024 and 2025 and there are concerns the company could fail to be viable as AI does the “job” students used to hire Chegg to do for them.

A pragmatic way to inform what is worth investing in and what isn’t is to create a scorecard around the dimensions that reflect your strategic intent. A scorecard lets people see the logic behind your strategic choices throughout the organization and act in accordance. The screening statements implicit in the scorecard make it crystal clear which opportunities are desirable and which aren’t, and allow ideas to be mapped against the same set of criteria. The magic is not the scores — it’s the thought process behind them.

So what would make a proposed investment in AI more or less desirable? A scorecard might look something like this:

The creation of scorecards like this is often a combination of experience with the past and insight about the future. In a workshop setting, you might ask what kinds of projects have worked for us in the past? What kind have not, and we wish we hadn’t invested in them? What criteria made the difference? You also want to think about the future — what dimensions reflect your aspirations about where and how you would like to play going forward.

Next, go back to your inventory of projects and initiatives and score them. This could be a quick-and-dirty analysis by your strategy group or executive team — the goal here is not perfection but directional correctness. Once each major initiative has a score, rank order them. The top ranked ones should get the highest priority. The second group can be maintained if you still have resources. The lowest group are candidates for discontinuation, and should go through a disengagement process.

AI is certainly a development that should be on the radar for most organizations. That doesn’t mean you have to throw massive amounts of capital into the game. Participating smartly and with discipline will allow you to stay on top of the new developments without betting the company on them.