In-House AI Development Strategies

Every company has an "AI strategy" now. But 90% suck. Here's step-by-step how to build one that doesn't: AI strategy is different from regular product strategy. This is the battle-tested framework Miqdad Jaffer & I use. We've used at Shopify, OpenAI, & Apollo: — 1. SET CLEAR OBJECTIVES At Shopify, Miqdad killed dozens of technically cool AI projects... And doubled down on inventory management. Why? That’s where merchants were losing money. No business impact = no AI initiative. Simple as that. Look for pain points humans consistently fumble, impact their growth, and first solve that with AI. — 2. UNDERSTAND YOUR AI USERS Users don’t adopt AI the same way they adopt a button or a new flow. They don’t JUST use it. They test it, build trust with it, and only then rely on it. So, build something that empowers them throughout their journey with your product. — 3. IDENTIFY YOUR AI SUPERPOWERS Not everyone has access to the same behavior signals... User context, or proprietary data that make outputs smarter over time. That’s your moat, the data nobody else can use. Not the fancy models. Not the MCPs. Not even revolutionary AI agents. Your goal is to build around your moat, not your product or models. — 4. BUILD YOUR AI CAPABILITY STACK In AI, speed beats pride. Think of it this way: A team spends 9 months building their own LLM. Meanwhile, a smaller competitor ships with OpenAI and captures the market. So, did you make the smartest move by trying to build everything yourself? Great PMs lead when to build and when just to leverage. — 5. VISUALIZE YOUR AI VISION In 2016, Airbnb used Pixar-level storyboards to communicate product moments. Today? Tools like Bolt, v0, and Replit make it possible in hours for a fraction of a cost. Create visiontypes that show: → Before vs. after (and make the “after” impossible to do manually) → Progressive learning and smarter experiences → Human + AI collaboration in real workflows — 6. DEFINE YOUR AI PILLARS At this stage, you’re building a portfolio of some safe and some big bets: → Quick wins (1–3 months) → Strategic differentiators (3–12 months) → Exploratory options (R&D, future leverage) And label each one clearly: Offensive = creates new value Defensive = protects from disruption Foundational = unlocks future bets — 7. QUANTIFY AI IMPACT If your AI strategy assumes flat, linear returns - you’re modeling it wrong. AI compounds with usage. Every interaction trains the system, feeds the flywheel, and lifts the entire product. Even Sam Altman shared that just adding a “thank you” feature increased OpenAI’s operational cost by millions.... — 8. ESTABLISH ETHICAL GUARDRAILS One biased result. One hallucination. One misuse. And the entire product feels unsafe. Set guardrails around every part of the process to make it safe... From all the hallucinations that disrupt your trust! — Making a great strategy is still hard. But these steps can help.

If you’re leading AI initiatives, here is a strategic cheat sheet to move from "𝗰𝗼𝗼𝗹 𝗱𝗲𝗺𝗼" to 𝗲𝗻𝘁𝗲𝗿𝗽𝗿𝗶𝘀𝗲 𝘃𝗮𝗹𝘂𝗲. Think Risk, ROI, and Scalability. This strategy moves you from "𝘄𝗲 𝗵𝗮𝘃𝗲 𝗮 𝗺𝗼𝗱𝗲𝗹" to "𝘄𝗲 𝗵𝗮𝘃𝗲 𝗮 𝗯𝘂𝘀𝗶𝗻𝗲𝘀𝘀 𝗮𝘀𝘀𝗲𝘁." 𝟭. 𝗧𝗵𝗲 "𝗪𝗵𝘆" 𝗚𝗮𝘁𝗲 (𝗣𝗿𝗲-𝗣𝗼𝗖) • Don’t build just because you can. Define the Business Problem first • Success: Is the potential value > 10x the estimated cost? • Decision: If the problem can be solved with Regex or SQL, kill the AI project now. 𝟮. 𝗧𝗵𝗲 𝗣𝗿𝗼𝗼𝗳 𝗼𝗳 𝗖𝗼𝗻𝗰𝗲𝗽𝘁 (𝗣𝗼𝗖) • Goal: Prove feasibility, not scalability. • Timebox: 4–6 weeks max. • Team: 1-2 AI Engineers + 1 Domain Expert (Data Scientist alone is not enough). • Metric: Technical feasibility (e.g., "Can the model actually predict X with >80% accuracy on historical data?") 𝟯. 𝗧𝗵𝗲 "𝗠𝗩𝗣" 𝗧𝗿𝗮𝗻𝘀𝗶𝘁𝗶𝗼𝗻 (𝗧𝗵𝗲 𝗩𝗮𝗹𝗹𝗲𝘆 𝗼𝗳 𝗗𝗲𝗮𝘁𝗵) • Shift from "Notebook" to "System." • Infrastructure: Move off local GPUs to a dev cloud environment. Containerize. • Data Pipeline: Replace manual CSV dumps with automated data ingestion. • Decision: Does the model work on new, unseen data? If accuracy drops >10%, halt and investigate "Data Drift." 𝟰. 𝗥𝗶𝘀𝗸 & 𝗚𝗼𝘃𝗲𝗿𝗻𝗮𝗻𝗰𝗲 (𝗧𝗵𝗲 "𝗟𝗮𝘄𝘆𝗲𝗿" 𝗣𝗵𝗮𝘀𝗲) • Compliance is not an afterthought. • Guardrails: Implement checks to prevent hallucination or toxic output (e.g., NeMo Guardrails, Guidance). • Risk Decision: What is the cost of a wrong answer? If high (e.g., medical advice), keep a "Human-in-the-Loop." 𝟱. 𝗣𝗿𝗼𝗱𝘂𝗰𝘁𝗶𝗼𝗻 𝗔𝗿𝗰𝗵𝗶𝘁𝗲𝗰𝘁𝘂𝗿𝗲 • Scalability & Latency: Users won’t wait 10 seconds for a token. • Serving: Use optimized inference engines (vLLM, TGI, Triton) • Cost Control: Implement token limits and caching. "Pay-as-you-go" can bankrupt you overnight if an API loop goes rogue. 𝟲. 𝗘𝘃𝗮𝗹𝘂𝗮𝘁𝗶𝗼𝗻 • Automated Eval: Use "LLM-as-a-Judge" to score outputs against a golden dataset. • Feedback Loops: Build a mechanism for users to Thumbs Up/Down outcomes. Gold for fine-tuning later. 𝟳. 𝗢𝗽𝗲𝗿𝗮𝘁𝗶𝗼𝗻𝘀 (𝗟𝗟𝗠𝗢𝗽𝘀) • Day 2 is harder than Day 1. • Observability: Trace chains and monitor latency/cost per request (LangSmith, Arize). • Retraining: Models rot. Define when to retrain (e.g., "When accuracy drops below 85%" or "Monthly"). 𝗧𝗲𝗮𝗺 𝗘𝘃𝗼𝗹𝘂𝘁𝗶𝗼𝗻 • PoC Phase: AI Engineer + Subject Matter Expert. • MVP Phase: + Data Engineer + Backend Engineer. • Production Phase: + MLOps Engineer + Product Manager + Legal/Compliance. 𝗛𝗼𝘄 𝘁𝗼 𝗺𝗮𝗻𝗮𝗴𝗲 𝗔𝗜 𝗣𝗿𝗼𝗷𝗲𝗰𝘁𝘀 (𝗺𝘆 𝗮𝗱𝘃𝗶𝗰𝗲): → Treat AI as a Product, not a Research Project. → Fail fast: A failed PoC cost $10k; a failed Production rollout costs $1M+. → Cost Modeling: Estimate inference costs at peak scale before you write a line of production code. What decision gates do you use in your AI roadmap? Follow Priyanka for more cloud and AI tips and tools #ai #aiforbusiness #aileadership

Don’t overcomplicate AI for your legal team. Here are 12 initiatives to get started: (Based on conversations about AI with over 300 in-house lawyers): PEOPLE 1. Organise CPD sessions on key legal-specific topics. Examples: 'Gen AI for Legal Practice', 'Under the Hood of an LLM' and 'Prompt Engineering 101.' 2. Create dedicated AI experimentation time each month. Let your team know it's okay to experiment (safely). Set up guardrails and opportunities to share knowledge. 3. Identify Innovation & Technology champions. Peer-to-peer sharing is key. Your champions will drive digital literacy and engagement. GOVERNANCE* 1. Understand privacy and confidentiality requirements across different legal workstreams. Consider segmenting by data-type (e.g., client, company sensitive, company non-sensitive). 2. Consider privacy and confidentiality of different AI approaches. For example, state-of-the-art proprietary services vs. smaller, hosted models. 3. Set up set of rules for using AI to align with privacy and confidentiality requirements. TECHNOLOGY 1. Identify 3 legal work streams that present high potential for automation. 2. Assess the benefits and risks associated with each. 3. Survey the market for legal technology solutions that align with identified opportunities. Consider collaborations with law firms and industry experts to build customised solutions. OPERATIONS 1. Review legal team processes and identify 3 priority areas for optimisation and automation. These might include team meetings, client management, knowledge management, etc. 2. Develop an AI knowledge hub for the legal team. Include a prompt library, use cases, user guides, and lessons learned. 3. Collaborate with other areas of the business. Ensure the legal team is part of organisation-wide AI projects - from both a risk and legal ops perspective. *This assumes a foundational layer of governance and risk management, e.g. AI Guiding Principles, Risk Management Frameworks, etc. -- Here’s the thing: Legal teams won't be first up for new AI initiatives. They could be behind or lost in the shuffle. That's a real shame - because the opportunity for AI in law is huge. AI will help in-house lawyers move up the value chain. Do less boring work. Do more stuff that matters. I really want to see that happen. And these initiatives can help your team get there. Let me know your thoughts below - is your team exploring any of these initiatives? What do you think of this approach? #lawyers #ai #inhousecounsel

Most consumer brands are experimenting with #AI. Very few are building AI operations. That’s the real shift happening right now. I had a call with a leading home-care brand yesterday and spoke with one of their executives who oversees building in-house AI products. Some of the examples she shared with me, and the similarities to other industries they're replicating, were quite impressive. Platforms like Google Vertex AI are quietly becoming the infrastructure layer behind the next generation of enterprise AI systems, what many now call #agenticAI. Not just models. But full operational stacks: • data pipelines • experiment tracking • model training at scale • model registries • deployment endpoints • grounding with enterprise data • AI agents orchestrating workflows In other words: AI moving from analysis→ to execution. Under the hood, platforms like Vertex AI combine foundation models such as Gemini with enterprise MLOps and data infrastructure. Developers can experiment in Vertex AI Studio, train or fine-tune models, register them in the Model Registry, and deploy them through Prediction endpoints for batch or real-time inference. With BigQuery integration, Feature Store, and Vertex AI Pipelines, data scientists can operationalize predictive and generative models across the full ML lifecycle while continuously monitoring drift, skew, and performance. Where things get interesting is agent orchestration. Vertex AI Agent Builder enables companies to build multi-agent systems where specialized agents collaborate using tools like RAG retrieval, vector search, and API connectors to enterprise systems. Using frameworks like the Agent Development Kit (ADK), teams can deploy production agents in under 100 lines of code, connect them to ERP, marketing platforms, and data warehouses, and scale them on a managed runtime while maintaining governance, security, and observability across the entire agent ecosystem. And some of the biggest consumer brands are already moving in this direction: • Mondelēz Internationalēz scaled 20M personalized marketing assets globally. • General Mills is applying AI to supply chain and commercial decision-making. • The Estée Lauder Companies Inc. / Jo Malone London built an AI-powered fragrance advisor to replicate in-store expertise digitally. • Kraft Heinz reduced product content development from 8 weeks to 8 hours using Google AI tools. This is the early stage of agentic enterprise systems that will soon assist and increasingly execute workflows across: • marketing planning • retail media optimization • digital commerce operations • demand sensing • product content creation • retailer joint business planning Over the next 5 years, the brands that win won’t have the most AI pilots. They’ll have AI embedded directly into the decision-making process. The real disruption won’t be AI writing copy. It will be AI running parts of the business. Supply chain and media operations are already leading the pack.

In one of my recent executive coaching sessions, a CEO told me they had built an internal version of ChatGPT which was announced with great fanfare, but the adoption rate has been poor and keeps declining. Employees tried it once, compared it to the tools they use at home, and went back to pasting things into Claude and ChatGPT instead. When digging into the details I saw two issues. First, it was built as a general-purpose chatbot competing with tools that have billions in R&D behind them and ship updates weekly, a race you can't win. Second, it was treated as a classic waterfall project driven by IT, designed, built, shipped, done, while the expectations set by the publicly available tools kept moving further away every month. Building in-house AI capabilities and platforms is essential, and the best ones are built around specific business problems where your own data and processes give them a real edge over anything generic. Involve actual business users from day one, not for requirements sign-off but for their ideas and their buy-in, evolve the product continuously, and treat it like something that's never finished. The people who help shape it are the ones who end up championing it.

Your AI strategy is only as strong as your operating model. Turning vision into execution requires three deliberate shifts. 1/ Design the organization around AI, not beside it In the early stages, it makes sense to centralize AI expertise to establish standards, tooling, and governance. But execution fails when AI remains isolated as a function. To scale, AI must be woven into how the organization actually runs: - Clear interfaces between technical teams and business owners - Defined handoffs between AI systems and human operators - Explicit roles for who designs the system, who monitors it, and who intervenes when it fails If AI lives next to the business instead of inside it, adoption stays superficial and accountability remains unclear. 2/ Make ownership explicit before automation expands Execution breaks down fastest where ownership is assumed rather than assigned. Every AI-enabled workflow needs: - A named owner accountable for outcomes - Clear escalation paths when the system encounters ambiguity - Agreed rules for when AI defers, pauses, or hands control back to humans AI does not eliminate responsibility. It concentrates it. Without clear ownership, organizations gain speed at the cost of trust. 3/ Sequence before you scale One of the most common execution mistakes is layering AI onto unstable workflows. Effective teams move in order: 1. Stabilize the workflow and define exceptions 2. Assign ownership and escalation paths 3. Introduce AI with constrained scope 4. Expand autonomy only after reliability is proven Skipping steps creates systems that perform well in demos but fail under real-world pressure.

Are you building your AI strategy on quicksand? This week on the Product Thinking Podcast, I spoke with Dr. Maryam Ashoori, PhD from IBM Watson X about a critical blind spot in AI product development. Her insight hit me: "The world is changing so fast and by the time that your product gets out, there's a good chance that these underlying technology is already outdated." Think about it. Companies are building entire product strategies around specific AI models or tool vendors. But what happens when GPT-7 launches, or a your chosen tool is discontinued? You're stuck with expensive rebuilds instead of seamless upgrades. Maryam's solution? Build with architectural abstraction. Separate your business logic from the AI technology through layers that treat models and external tools as interchangeable components. This approach prevents technical debt and ensures strategic survival. I've seen enterprises trapped by their own tech choices, unable to leverage breakthrough advances because they didn't plan for change. The cost of switching becomes prohibitive, so they fall behind. The companies winning in AI aren't necessarily picking the best models today. They're building systems that can adapt to whatever comes next. How technology-agnostic is your AI architecture? Are you ready for the next breakthrough, or locked into yesterday's choices?

AI strategy that wins: build outcomes, not just models. Most AI plans are shopping lists. Winning strategy is a connected system miss one link and results stall. Common breakdowns (diagnose in seconds) Direction w/o Demand → elegant solution, quiet pipeline Demand w/o Economics → top line up, runway down Advantage w/o Direction → margin today, misallocated effort Economics w/o Advantage → value created, race to the bottom The four pillars (Breakthroughs happen at the overlap, not in a silo) 🧭 Direction — Where AI plays. How it’s governed. How wins are measured. 🎯 Demand — Problem felt weekly. Named owner/sponsor. 💰 Economics — Unit cost & payback. Capacity redeployed or revenue. 🔑 Advantage — Proprietary data. Domain expertise. Reusable components. Build only when these 4 are true (the overlap): 1. Strategic fit: Only we should build it (our data/mission) 2. Relevance: Felt problem this quarter 3. Viability: Profitable at scale (payback ≤ 12 months) 4. Efficiency: Low run cost; reusable components Board metric stack North star: one outcome people feel Pick one metric: lead time • error rate • time to feedback • cost per run • capacity redeployed Decision gates (go only if) ☑️ Workflow + sponsor named ☑️ Baseline + target set ☑️ Data access + governance cleared ☑️ Payback ≤ 6–12 months ☑️ ≥50% components reusable for next 2 use cases 90-day runbook Days 1–15: select workflow, baseline, risk check, sign charter Days 16–45: ship a thin slice with real users, instrument metrics Days 46–90: prove lift, document reuse, decide: scale / pause / kill Quick heat check Direction ☐ Red ☐ Yellow ☐ Green Demand ☐ Red ☐ Yellow ☐ Green Economics ☐ Red ☐ Yellow ☐ Green Advantage ☐ Red ☐ Yellow ☐ Green Repost to help someone in your network make better AI bets. Follow Gabriel Millien for pragmatic AI and ops insights. Save for your next portfolio or board review. Infographic style inspiration: Justin Wright

Your org structure can hold back your AI strategy Because AI projects do not succeed in isolation. They succeed or fail inside the organizational system around them. → Structure → Governance → Decision rights → Authority → Resource allocation Many AI strategies focus on models, tools, and data, and they skip a harder question. → Is our organization built to execute cross-functional work at scale? AI does not behave like a functional initiative. AI: ✅ Connects data across departments ✅ Reshapes end-to-end processes ✅ Shifts decision rights ✅ Requires shared ownership of outcomes That does not belong to Finance, IT, or Operations alone. It belongs to the system. This is why the same pattern repeats. → Finance launches an AI initiative → Customer service runs a pilot → Operations experiments with automation → IT builds a central platform Each effort looks rational on its own. At enterprise level, priorities diverge, ownership blurs, and decisions slow down. The organization generates activity. The P&L barely moves. The constraint is not the model. The constraint is the structure. Functional structures were built for stability and specialization. AI requires integration and shared authority. That is why organizations that scale AI usually change operating logic before they change the tech stack. → Stronger matrix execution → Outcome-led governance → Authority that follows end-to-end value, not departments If AI is truly strategic, your operating model must reflect that priority. No AI strategy will scale beyond the limits of the organization that has to execute it.

🤯 Eating the AI Elephant: A Friday Thought.. Bill Dobbie, Ian Morrison and I were swapping notes on how AI is impacting the businesses we’re involved with—I shared a mental model I’ve built from various experiences and realised it might be useful to share more broadly. AI can feel like a giant elephant—overwhelming. But breaking it into 3 areas makes it manageable. Each gives you a way to progress and a scorecard to measure it. Break the AI challenge into: ⚡ Product, Service & Growth ⚡ Optimisation & Cost Reduction ⚡ Culture & Adoption (the most important one, in my view) 1. Product, Service & Growth This is where most teams start—how can AI support the customer offering? For service businesses, it might mean better analysis, faster transcription, or smarter chatbots. Tech firms will push into agentic AI and embedded intelligence. 2. Optimisation & Cost Reduction This isn’t about cutting jobs—it’s about avoiding unnecessary hires. Most teams ask for headcount when they hit a 20–50% capacity gap. AI tools can return 20%+ of time—enough to keep moving without hiring. Whether it’s automation, no-code apps like lovable.dev, or better workflows, it’s about helping teams do more with less. 3. Culture & Adoption The real game-changer—yet often overlooked. The best organisations are embedding AI into culture, not just the tech stack. Here are a few approaches I’m seeing: AI Champions: Appoint team members to experiment and build internal knowledge. Forge Holiday Group has done this brilliantly. In a 30–40 person firm, you might have 3–4 champions. Larger firms: 1 per 30–50 people. Hands-On Testing: Give everyone a £40 monthly budget for 3 months—enough to try one or two tools. Let them know in advance, and encourage chats with their AI champion. After testing tools that could help their role, they write a short ROI summary. If solid, they keep the tool. Training & Access: CreateFuture nailed this. They built an internal AI training program. Complete it, and you get a paid Gemini account. They track usage (ethically and transparently) to gauge adoption. That usage becomes part of your scorecard. Policy Development: As people explore, define how tools should be used—but do this after they’ve tested them. Don’t block access early on or you’ll cause delays and stifle adoption. Not easy if you’re regulated—I get that. This cultural approach is vital. Getting comfortable with AI now—even in small ways—will be far easier than playing catch-up later. It’s about building mental flexibility across the business, not just in the innovation team. So where do you start? Build 3 roadmaps: one for product & growth, one for optimisation, one for culture. Assign different lead teams and scorecards. Make space for people to learn, play, and adapt. That’s where real transformation happens. 👍 Worth following Graham Donoghue — an exemplar CEO in this space who shares loads in his feed 👍 Please share any good transformation examples you're seeing below