The Four Value Plays of AI Strategy. And Why They Require Different Approaches.

This is an article written by Steven Nooijen, Head of Data & AI Strategy at Xebia, and was originally published on Steven’s blog. 

Why Most AI Strategies Create Confusion 

Ask ten executives what their “AI strategy” is and you’ll get ten answers that sound similar but address different things. One is rolling out Copilot licenses. Another is speeding up the software delivery lifecycle. A third is automating an invoicing process. The fourth is evaluating LLM platforms for a customer facing chatbot.  

AI enables four different value plays. If you don’t distinguish between them, the chances are that your AI conversation creates confusion. Each play has its own benefits, challenges, operating model, and hence investment logic. This article describes the four plays, so that you can create a clear and actionable AI Strategy for your organization. 

1. Personal Productivity AI 

Tools that enable individual knowledge workers in their day-to-day jobs through assistants like Microsoft CoPilot, ChatGPT, Gemini or Claude CoWork. AI tasks cover writing emails, summarizing meetings, analyzing documents and spreadsheets, or building decks.  

Questions we hear from leadership: 

What makes it hard. This category looks easy because the tools are cheap and abundant, but extracting real value from them is much harder than buying licenses. In many organizations adoption stays shallow as training and change management are an afterthought. Anxieties about data security and compliance block the use cases that would matter most, with the biggest risk being confidential information leaking into the public domain. 

Investment logic. The economics of this play are about adoption at scale. Licenses are cheap on a per-seat basis, but you’re rolling them out to hundreds of people, which means most of the real spend goes into training, support, and change enablement rather than the technology itself. The return shows up as hours saved across a large population with all kinds of activities, notoriously hard to tie back to revenue. Track success through adoption rates and cost-per-seat. Run the programme as an IT enablement effort, not an engineering one. 

2. Software Development AI 

Code assistants and LLM-powered development frameworks change how your engineers create software. These tools help developers with tasks like generating code, debugging, and creating documentation. This significantly speeds up the software delivery lifecycle (SDLC). 

Questions we hear from leadership: 

What makes it hard. The main metric in this play is speed, but the real challenge is the trade-off between speed and quality. Code that ships faster but fails under heavy load, or that introduces subtle security vulnerabilities, can erase the apparent productivity gains overnight. Security, IP, and compliance risks multiply once a model has access to your proprietary codebase. And there’s a softer challenge undermining adoption underneath all of this: engineers are losing trust in other people’s (AI-generated) code and are worried about lagging behind or even losing their jobs.  

Investment logic. Costs in this category come in two layers: a per-engineer license fee for the tools themselves, and the often-underestimated cost of redesigning engineering workflows around them. The good news is that ROI is genuinely trackable through metrics like cycle time, deployment frequency and defect rate. If engineering leadership commits to measuring this deliberately, the spend becomes visible on the books and the gains become impossible to prove. 

3. Business Process AI 

This is AI to automate or improve internal processes, common use cases include: ticket triage, R&D, invoice processing, demand forecasting or knowledge assistants. We are already doing this for years with “classical” AI techniques like machine learning and neural networks, but Generative AI and Agents are creating new opportunities or can enhance existing solutions.  

Questions we hear from leadership: 

What makes it hard. This is where the loudest “where’s the ROI?” conversations happen, because every use case is expected to pay back within a budget cycle. This becomes especially tricky when cross-department processes are redesigned. These processes hold the biggest ROI promise but are also the most organizationally challenging in terms of ownership and governance. A new technical challenge introduced by agentic systems is that they don’t behave like deterministic software. This means that testing and monitoring them reliably require techniques most teams haven’t developed yet.  

Investment logic. Investment in this category is use-case driven, with each initiative funded against a specific business process. The ROI is also the most concrete of the four plays because every project has a defined scope and a clear baseline to compare against: FTE savings, increased throughput, lower error rates. That same precision is what makes scrutiny so high. If you can’t show proof of value within twelve months, the use case might be killed before it gets a chance to scale. 

4. Customer-Facing AI 

Customer-facing AI is where AI interacts directly with your customers. It’s embedded in your products, apps, and digital channels and owned by teams who are accountable for your brand’s positioning: product, digital, CX, sales or marketing. Applications are chatbots, personalization engines, smart search or AI-generated content.  

Questions we hear from leadership: 

What makes it hard. Security, regulatory, and ethical risks effectively turn every release into a compliance event. Hence, building and testing customer-facing AI is a serious engineering discipline, not something you should prototype over a weekend. As “garbage in, garbage out” still applies, high-quality and reliable data is crucial, but it is hard to acquire since customer data is often fragmented across CRM, product, and support systems.  

Investment logic. The return is measured in commercial outcomes: conversion, retention, NPS, revenue uplift, with brand-reputation risk sitting on the other side of the ledger as a real downside. Payback timelines are longer than for the other plays, as the platform investment is substantial. You should see this type of AI as a product investment rather than IT spend. Hence, the strategic decision sits with your product and commercial leadership. 

How to Build an Effective AI Strategy. The two steps every leadership team should take 

Concluding: "AI strategy" is actually four strategies. How should you proceed? 

Step 1: Be explicit about where AI creates value for you. Which of the four categories is your priority? Which is secondary? Which aren’t yours to play in right now? Vague answers here cause misaligned investment downstream. If you want a structured way to get to a sharp answer, our AI strategy consulting team runs exactly this diagnostic with leadership teams. 

Step 2: Build a modern Data & AI operating model that offers real flexibility across architecture, procurement, deployment, change enablement, and governance. Each of the four plays will stress one or more of these dimensions differently. Read more in our companion piece on how to establish an adaptive and resilient operating model that's built to support all four plays at once. 

Thank you for reading. We hope this gives you a sharper lens for the next AI strategy conversation in your organization. 

If you'd like to discuss which AI priorities make the most sense for your organization, and what operating model is needed to support them, schedule a complimentary strategy consultation with Steven Nooijen, Head of Data & AI Strategy at Xebia (author of this article), or one of our regional Data & AI experts. Book your consultation.