Companies are missing the mark with AI projects. The headlines are not great.
- Gartner Predicts 30% of Generative AI Projects Will Be Abandoned After Proof of Concept By End of 2025
- Research by the RAND Corporation estimates more than 80% of AI projects fail, which is twice the rate of failure of non-AI projects.
- Deloitte shows some benefits accruing such as uncovering ideas and insights with ROI generally positive, and yet, barriers due to mistakes with real-world consequences.
- BCG’s 2025 survey indicates only “25% of executives are seeing significant value from AI”, with projects vulnerable due to unclear use cases and poor data readiness.
What’s going on? And more to the point; can product people help improve these outcomes?
Perhaps there are efforts to solve tech problems, but not always customer problems. From a Product person’s perspective, it’s time to re-visit basics and use cases. In the add AI scramble, there’s confusion about our shiny new toys. You see it in the drive to do anything with them. You see ridiculous job requirements posted from tech to product asking 5 – 10 years experience deploying AI. (For traditional ML, that may be fair. But they usually mean GPTs/LLMs, which have only been around a handful of years.)
Let’s reconsider our new capabilities from customer use case perspectives with focus on LLMs/GPTs. While also growing, we’ll skip traditional ML as it’s a separate and better understood category. (Things like regression models, decision trees, or clustering algorithms used in fields like finance, logistics, or medical diagnostics, which have established frameworks and decades of practical application.)
Why should we try a look from use case perspectives? Because for those of us that work in Product, instead of panicking figuring out how to check off the “We’re doing AI” into a checkbox in an Annual Report or a marketing piece, it might be useful to have clarity of purpose. We’ll focus on Generative Pre-trained Transformers (GPTs) and the Large Language Models (LLMs) over traditional ML or multi-modal GPTs.
Strategic Arenas for AI Impact
Here’s the general business domains for AI use.
- AI Industry Itself: This one needs to be included for completeness, but it’s not where most businesspeople live. This is building core ML/AI tools or supporting products.
- Internal Operations: We’re working on internal efficiencies. This could be manufacturing quality, safety and efficiency, to loan approvals, financial risk assessments, and other operations.
- Marketing: We use AI tools from top to bottom of the marketing stack; customer research through Go To Market assistance and automation.
- Product: We’re building products incorporating AI of some sort. This is a broad category. The obvious includes better chatbots, customer service analysis and response, answer engines into our FAQs and Knowledge bases, search tools, agent tools like product and service recommendations and more. This is the place where we can create value with new and different offerings.
Excepting the AI industry itself, we should see where there’s a strategic or tactical fit for AI. In some areas, they’re tooling; we’re accomplishing existing goals in new ways. When it comes to product though, they’re more than simple tools and we have to think more strategically. What goals and objectives might this tech help accomplish for our customers beyond our current offers? It’s not the kind of technical question it might be from an operations perspective. It gets back to a more basic customer WIFM. (What’s In It For Me.)
When thinking about where to start, consider using an idea like an “AI Value Ladder.”
- Low-Value Quick wins: i.e., automating FAQs and knowledge bases for efficiency boosts.
- Mid-tier Enhancements: Such as process improvements on existing workflows, content personalization for marketing or automated report generation.
- High-value Disruptors: New applications that redefine business models; agentic AI in supply chain management that could drastically revolutionize the whole process.
Considering Our Customers
Let’s go over customer use case categories. You should be thinking, “What aspects of these values can enhance our offering(s)?” The traditional questions apply: Are there customer pain points the new tools can solve? Maybe it’s time to revisit long-term problems for which we just had no viable solutions. Can you map AI capabilities to any of these gaps? As you look at each area, consider how solutions might be used by autonomy levels; as a Tool (a passive utility), Copilot (context-aware helper), or Teammate (semi-autonomous partner).
Here’s general use case categories from a customer perspective.
- Search: Search traditionally has had a few use cases: navigation, seeking transactions, or seeking information. That last one gets interesting. You could be seeking just a data point, but more likely intending to use that info as part of some other task. A 2024 Gartner report predicts that by 2026, 25% of search queries will be handled by answer engines, reducing decision-making time by up to 40% in data-intensive industries like finance and logistics.
- LLM/GPT Answer Engine: Here, LLMs and GPTs evolve beyond information retrieval to deliver synthesized, context-aware responses. Unlike a list of links, an answer engine provides summaries and explanations, and concise analysis, increasingly also including cited sources for verification. (Though citations may come after the fact, not part of provenance for the original answer.) This shifts the paradigm from “find and filter” to “ask and apply,” reducing cognitive load and accelerating workflows.
- LLM/GPT as Copilot: Code creation, Writing, etc. Acting as a collaborative assistant, LLMs/GPTs serve as copilots across domains; whether generating code for developers, drafting reports for analysts, or brainstorming ideas for marketers. The key value is augmentation. They can handle repetitive tasks, suggest optimizations, and iterate based on feedback.
- GPT as Creative Generator: GPTs can work as ideation tools with content creation, producing everything from marketing copy and visual concepts to innovative product features, even whole websites and more. This use case taps into their generative capabilities either to spark originality from basic ideas or to produce what’s being asked for specifically. (Challenges remain, as many mistakes still occur and prompt iteration for refinement can be frustrating, but the tools are getting better all the time.)
- LLM/GPT as Trainer: Beyond being used an “Answer Engine” for a one-off type of challenge, they can be used as educational guides for upskilling and in some cases they may serve as simulation engines.
- LLM/GPT as Agentic Coordinator: This area is emergent. As of this writing, the Meme of the Year would be “Agentic.” AI models can be at various stages along a workflow that incorporates agents. LLMs can serve as either (or both) decision points to launch agents towards a sub-task or as processing agents to handle returned outcomes. This takes traditional workflows to more context-aware and autonomous.
Mind the Gap
Are there any new gaps between what your customers need and what’s now possible? We spend a lot of time looking at things. Maybe not enough at the relationship(s) between and among them.
Is your product idea a wholly new capability? Or a copilot adjunct to what you’ve got? i.e., is it revolutionary or evolutionary? Some like to claim that strategically, unless you’re truly different, you have no defensible value proposition. As Alex M H Smith emphasizes in his strategy frameworks, such as in “How to Create Unique Value and Outsmart Competitors,” (YouTube link), where he argues defining your category uniquely to build lasting competitive edges. This is a great goal, but it’s demonstrably not necessarily required. Being better or cheaper are valid strategies. They’re neither great nor the best long-term strategies. And they might not allow the same command of pricing power as more differentiated products. But if you look at any anything from the supermarket shelf to any digital product that’s survived for more than a handful of years, they’re typically not alone in their space. (When studying business strategy, it still pays to study more rigorous views, such as Michael Porter, “On Competition” which is an update beyond his Five Forces concepts, Richard Rumelt, “Good Strategy/Bad Strategy: The difference and why it matters” and similar. And oh yeah… don’t forget Blue Ocean Strategy, which is kind of like Alex Smith’s perspective, but with a more formal framework.)
Does your customer base or your product have longstanding issues or opportunities that were not technically solvable until now? That’s a gap. Examples include enhancing e-commerce platforms with GPT-powered recommendation engines that not only suggest products but explain why they fit the user’s lifestyle, closing the gap in personalized shopping. In healthtech, copilots could simulate patient interactions to train staff, bridging gaps in empathy-driven care. The business insight? Start with customer perspectives to map these gaps, quantify the value (time saved, revenue) to justify investment, and prioritize features that deliver quick wins while building toward differentiation.
When we start working with new capabilities, there’s likely going to be some waste. At this point though, we should be getting a better read on the general use case areas where we can consider solutions and some of the product pricing and cost models. There’s still going to be some newsworthy failures and experimental risk, but I think the next couple of years are going to be about trying to work harder on the win/loss rate, not just jam in AI somewhere.