It doesn't have to take months or millions, but you have to approach things the right way..
AI has been called everything from a civilization-changing technology that will steal our jobs to the most overhyped and useless technology since Web3 (no shade, but I’m still waiting for my blockchain investments to pay out).
Everyone seems to be asking the same question right now: “Where’s the ROI in AI?” Actually, two questions sometimes get unfairly merged. The first is when the massive capital investments in foundational companies like OpenAI and Anthropic will generate returns. The second is, can my company use AI to deliver real value? I can’t answer the first question, but I’ve basically built my company to answer the second.
The TLDR; is that a thoughtful strategy can build your confidence, clarify your path, and start showing results within weeks, not months. Since there is no proven playbook on how to build AI projects, the name of the game right now is incremental progress, not monumental projects.
Why AI projects fail to deliver ROI
According to random studies I looked up hastily on the web, 30% of AI projects never pass the proof-of-concept stage. Who knows if this is right, but it actually seems low to me and that’s not a bad thing. Why? There isn’t a simple answer. But there is a complicated one.
- Bad framing. Leaders think that because AI is technology, integrating it must be a technical problem. That is a problem. But it’s not THE problem. The problem is understanding how AI will enhance and disrupt your workflows and processes. Where can AI be transformative versus augmentative versus a waste of time and money? This is a business problem not a technical problem. Lacking this understanding then leads to the next factor…
- Delegation. Because it’s seen as a technical problem it gets handed off to technical leaders. You may have an awesome CTO, but that doesn’t mean he or she knows anything about best practices in AI. The tooling is amazing, so pretty much anybody with a modicum of Python knowledge can build an impressive seeming demo. And that’s mostly what happens. It sounds dumb, and I hope your company is an exception, but oftentimes, a technical leader will miss the forest for the trees, build a “cool” app, talk about how advanced it is, and then nobody will use it because it’s dog shit. Why does this happen? See the next factor…
- No strategy. We are strategy nerds in the most insufferable way. Our patron saint, Richard Rummelt showed us the light and now we’re rarely invited to parties. It is a very expensive mistake to start integrating AI without a proper strategy. The landscape today is too complicated and moving too quickly to be winging it. Having a strategy means understanding the state of play and building alignment around an executable and measurable plan. Easy to say, but it rarely happens, because… hard.
- Impatience. I think leaders feel pressure to move quickly in order to fend off competitors, feel innovative, or worse, impress the board. None of those things are bad, but they are if they are the motivation rather than the outcome. If ever there were a time to go slow to go fast, it’s now. In fact, laying down a proper foundation and establishing a solid, transparent development process will undoubtedly put you ahead of the curve when things really get heated up.
- Ambition. It’s easy to get excited by a proof-of-concept and ask, how quickly can we get this integrated into our systems? But traditional software architectures don’t always play nice with AI workflows. Additionally, developers tend to get nervous when coding with AI because it can be unpredictable and unreliable, which wreaks havoc on well-tested, buttoned up applications. Moving too quickly to integration usually drags the process into molasses making every decision fraught. Ambition is good. It just needs to be tied to a strategy that establishes the right time for integration. We recommend building AI systems in isolation, on different stacks, and coupling them loosely to existing systems.
AI strategy FTW
I’m not going to get into how we help develop a full AI strategy, but it’s something any decent product strategist would find sensible. I mention it here because the simplest things often get forgotten: How does your AI strategy connect to your overall business strategy? How, specifically, do we hypothesize that it will deliver on your goals? This kind of analysis sets the stage for the techniques we describe below.
Good ROI is about choosing the right opportunities.
We have two frameworks that we use to help us calculate and evaluate the projected (at the feasibility stage of the project) and actual (during and after the project) return on investment. They aren’t big secrets – they’re on our website, and I talk about them every chance I get: they’re called DEW and VVV. I’m going to touch on them briefly today, but we’ll come back to each in depth at a later date.
What is their purpose? (DEW) helps us identify valuable use cases up front, and (VVV) is used to score the size of their impact, ensuring every AI initiative is tied to measurable outcomes.
Looking for DEW
When we look for promising use cases, we’re really looking for DEW. It ensures we don't waste time exploring dead ends and focus on areas where AI can make an impact. Once you understand these areas, you can measure the ROI like any business enhancement or optimization project (e.g., through KPIs and leading indicators).
- Data: AI thrives on data. Do you have enough of it? The first step is to ensure the accessibility and quality of data, whether tabular, text, images, video, or telemetry. If you don't have the right data accessible now, we will know immediately—no months wasted building something that can't be trained. The nature of your data will have an outsized impact on the success of your project. A lack of data is usually a project killer.
- Expertise: What knowledge does your organisation rely on that’s difficult to scale or replace? If AI can encapsulate it effectively, it can transform that expertise into operational leverage. We can usually get a good idea of whether AI can truly augment your people’s capabilities within days.
- Workflows: Which processes are repetitive, dull, or resource-intensive? Automating these can unlock massive efficiencies. The quickest wins often come from automating existing processes - we can spot these opportunities in week one.
A VVVery cool way to estimate impact
By layering VVV onto our analysis, we focus on impact and ensuring projects are feasible, profitable, and deliver results fast.
- Viability: Can this be built? Factors such as technical complexity, organisational culture, and regulatory hurdles contribute to this. Some projects might seem intuitively easy, and that’s exactly what makes them difficult for an AI. Others might seem terribly complex based on the current approach, and AI can offer an entirely new way without much effort.
- Value: What’s the ROI? How does this project contribute to your bottom line, customer satisfaction, or operational efficiency? Will this show measurable ROI within the first month, first quarter, first year? Sometimes it’s strategically valuable to simply have confidence that an initiative is doable.
- Velocity: How quickly can you deliver results? Small wins build momentum, making velocity a critical part of success. We strive to demonstrate concrete value in the first week of effort because it keeps us focused on what matters instead of building out infrastructure or writing boilerplate code.
Go narrow and get deep.
Remember when I talked about incremental progress? I’m not advocating chipping away at the edges of a problem. It’s exactly the opposite. Once you’ve found a use case that you think could deliver transformational value to your business, zero in on the hardest, must unpredictable, or hard-to-discern aspect, and attack that. If you can start building confidence that an AI-based approach will work, don’t stop! One of the most common mistakes in AI projects is trying to solve everything at once. Instead, we recommend “starting narrow”: focusing on a specific, high-impact problem and then “going deep” to prove that you can solve it well.
Why? Because this approach builds confidence. A tightly scoped project allows you to deliver measurable results quickly with a small team, creating momentum and buy-in for future AI initiatives. You get quick wins, quantifiable results, and momentum for more significant projects down the line. It’s way better than spreading yourself thin and getting stuck working on aspects of the problem that don’t deliver meaningful value.
This approach isn't just about focus—it's about speed. By narrowing the scope, we often get a working prototype that delivers value within the first month. That's not just faster results: it's less draining on your team's time and energy. Working this way allows you to bring in your subject matter experts so they can evaluate results early and take ownership of the solution.
Getting concrete
One of our recent projects exemplifies everything I’ve shared above. I can’t go into the details, but the client’s business is based around data extraction and processing. They had competitors, but disruption was more likely to come from native AI upstarts. They decided to take advantage of their strengths (domain expertise, a huge lake of high-quality existing data, and a strong in-house tech team) to explore how to disrupt themselves. We asked them to imagine starting from scratch, but now with AI. How would they do it? Obviously, they imagined a very different company. We explored several critical use cases, from data extraction to validation to retrieval. After scoring them with VVV, we decided that one particular use case was foundational and, over time would be key in supporting the others. We built a custom AI engine that cut core processing time by 75%. The ramifications of this kind of change are hard to get your head around. It means reallocating people. It increases the speed and lowers the cost so much that they need to consider a new pricing model. And it opens up the possibility of new product offerings altogether. We’ve been working on this project for months, but it was clear in the first 6 weeks that what we wanted to do was possible. Our calculations showed that even a 10% improvement in processing time would be worth the effort. At 75% we’ve opened up possibilities that had never been considered. This is the kind of transformational ROI we know is possible when you approach things thoughtfully, precisely, and with diligence.
ROI for your AI project: proven and rapid
The key to AI ROI isn't mysterious. It's about moving fast, staying focused, and protecting your team's time. At Machine & Partners, we commit to showing value within the first week and generally delivering working solutions within a few months. If you're not seeing results that quickly, something's wrong with the approach, not the technology.
Want to learn more about how we deliver ROI in weeks, not months? Please get in touch. We love helping businesses transform through this exciting new technology.
