The Changing Nature of Performance Metrics With AI

Performance Management in the Age of AI: the new 3‑Dimensional Model For decades, the 9‑box grid shaped how organizations assessed talent—mapping individuals along two familiar axes: ✔ Business performance (“what”) ✔ Behaviors or potential (“how”) Over time, many companies moved away from this model, concluding it oversimplified the complexity of human performance and sometimes reinforced bias more than it reduced it. AI is fundamentally reshaping work, shortening the lifecycle of skills and creating new capability demands at a pace conventional frameworks were never designed to keep up with. As a result, a new paradigm for performance management is emerging. Organizations are starting to consider a three‑dimensional approach to performance—one that integrates not just what people deliver and how they behave, but also how they grow. The new 3D model consists of three axis: 1. Business Results: Measures impact, delivery, and contribution to outcomes. 2. Behaviors / Ways of Working: Captures collaboration, leadership etc. and.. 3. Skills Development: Assesses capability building, learning velocity, and readiness for future roles. The third axis reflects a simple reality: In an AI‑driven workforce, continuous skills development is no longer optional—it’s strategic. IBM has begun to formalize this multidimensional view in its talent and rewards model. Their approach includes: 1. Integrating skills into pay: Base pay and equity linked to skill progression. 2. Balancing objectives: Business and skills goals carry equal weight 3. Future skills visibility: Regular communication on evolving skill requirements see: https://lnkd.in/eTDE-XmE Not every organization can replicate this model at scale, but it illustrates where performance management is heading. The central questions are shifting. Not just: “Did someone deliver results?” But also: “Are they developing the skills the organization will need next?” and “Are they learning at the speed the environment requires?” The move from a 2D grid to a 3D, capability‑driven framework may become one of the most consequential shifts in performance management in the age of AI—signaling a future where growth, adaptability, and skill relevance stand on equal footing with results.

Over the last year, I’ve seen many people fall into the same trap: They launch an AI-powered agent (chatbot, assistant, support tool, etc.)… But only track surface-level KPIs — like response time or number of users. That’s not enough. To create AI systems that actually deliver value, we need 𝗵𝗼𝗹𝗶𝘀𝘁𝗶𝗰, 𝗵𝘂𝗺𝗮𝗻-𝗰𝗲𝗻𝘁𝗿𝗶𝗰 𝗺𝗲𝘁𝗿𝗶𝗰𝘀 that reflect: • User trust • Task success • Business impact • Experience quality    This infographic highlights 15 𝘦𝘴𝘴𝘦𝘯𝘵𝘪𝘢𝘭 dimensions to consider: ↳ 𝗥𝗲𝘀𝗽𝗼𝗻𝘀𝗲 𝗔𝗰𝗰𝘂𝗿𝗮𝗰𝘆 — Are your AI answers actually useful and correct? ↳ 𝗧𝗮𝘀𝗸 𝗖𝗼𝗺𝗽𝗹𝗲𝘁𝗶𝗼𝗻 𝗥𝗮𝘁𝗲 — Can the agent complete full workflows, not just answer trivia? ↳ 𝗟𝗮𝘁𝗲𝗻𝗰𝘆 — Response speed still matters, especially in production. ↳ 𝗨𝘀𝗲𝗿 𝗘𝗻𝗴𝗮𝗴𝗲𝗺𝗲𝗻𝘁 — How often are users returning or interacting meaningfully? ↳ 𝗦𝘂𝗰𝗰𝗲𝘀𝘀 𝗥𝗮𝘁𝗲 — Did the user achieve their goal? This is your north star. ↳ 𝗘𝗿𝗿𝗼𝗿 𝗥𝗮𝘁𝗲 — Irrelevant or wrong responses? That’s friction. ↳ 𝗦𝗲𝘀𝘀𝗶𝗼𝗻 𝗗𝘂𝗿𝗮𝘁𝗶𝗼𝗻 — Longer isn’t always better — it depends on the goal. ↳ 𝗨𝘀𝗲𝗿 𝗥𝗲𝘁𝗲𝗻𝘁𝗶𝗼𝗻 — Are users coming back 𝘢𝘧𝘵𝘦𝘳 the first experience? ↳ 𝗖𝗼𝘀𝘁 𝗽𝗲𝗿 𝗜𝗻𝘁𝗲𝗿𝗮𝗰𝘁𝗶𝗼𝗻 — Especially critical at scale. Budget-wise agents win. ↳ 𝗖𝗼𝗻𝘃𝗲𝗿𝘀𝗮𝘁𝗶𝗼𝗻 𝗗𝗲𝗽𝘁𝗵 — Can the agent handle follow-ups and multi-turn dialogue? ↳ 𝗨𝘀𝗲𝗿 𝗦𝗮𝘁𝗶𝘀𝗳𝗮𝗰𝘁𝗶𝗼𝗻 𝗦𝗰𝗼𝗿𝗲 — Feedback from actual users is gold. ↳ 𝗖𝗼𝗻𝘁𝗲𝘅𝘁𝘂𝗮𝗹 𝗨𝗻𝗱𝗲𝗿𝘀𝘁𝗮𝗻𝗱𝗶𝗻𝗴 — Can your AI 𝘳𝘦𝘮𝘦𝘮𝘣𝘦𝘳 𝘢𝘯𝘥 𝘳𝘦𝘧𝘦𝘳 to earlier inputs? ↳ 𝗦𝗰𝗮𝗹𝗮𝗯𝗶𝗹𝗶𝘁𝘆 — Can it handle volume 𝘸𝘪𝘵𝘩𝘰𝘶𝘵 degrading performance? ↳ 𝗞𝗻𝗼𝘄𝗹𝗲𝗱𝗴𝗲 𝗥𝗲𝘁𝗿𝗶𝗲𝘃𝗮𝗹 𝗘𝗳𝗳𝗶𝗰𝗶𝗲𝗻𝗰𝘆 — This is key for RAG-based agents. ↳ 𝗔𝗱𝗮𝗽𝘁𝗮𝗯𝗶𝗹𝗶𝘁𝘆 𝗦𝗰𝗼𝗿𝗲 — Is your AI learning and improving over time? If you're building or managing AI agents — bookmark this. Whether it's a support bot, GenAI assistant, or a multi-agent system — these are the metrics that will shape real-world success. 𝗗𝗶𝗱 𝗜 𝗺𝗶𝘀𝘀 𝗮𝗻𝘆 𝗰𝗿𝗶𝘁𝗶𝗰𝗮𝗹 𝗼𝗻𝗲𝘀 𝘆𝗼𝘂 𝘂𝘀𝗲 𝗶𝗻 𝘆𝗼𝘂𝗿 𝗽𝗿𝗼𝗷𝗲𝗰𝘁𝘀? Let’s make this list even stronger — drop your thoughts 👇

AI field note: my word of the year is 𝔼𝕍𝔸𝕃: celebrating the art and science of rigorous measurement of AI performance, progress and purpose. (1 of 3) This year delivered a wealth of new AI models, architectures, and use cases - all united by one thread: evaluation. Model benchmarking, evaluation, or just "eval" has evolved from a simple, singular measure to a more complex blend of stats, metrics, and measurement techniques. Today's evals help discerning practitioners make pragmatic, informed technology decisions and measures improvements as AI systems are tuned. With AI innovation accelerating, staying up to date on evals ensures informed trade-offs when building intelligent systems, agents, and applications. Let's start by looking at measuring "performance"; the best way we know how to compare model behaviors, and find the right fit-for-purpose. Defining 'good performance' now involves a sophisticated suite of metrics across diverse dimensions. ⚙️ Task eval - beyond raw performance numbers. Today's evals measure how models perform across diverse scenarios - from basic comprehension to complex reasoning, reliability, consistency, and nuanced evaluation of reasoning paths, output quality, and edge case handling. 👛 Token economics - balancing cost, efficiency, and operation. Understanding token costs - both input and output - was essential last year, but evals have evolved beyond raw price per token, to understanding efficiency patterns, batching strategies, and the total cost of operation. ⏲️ Time-to-first-token. Speed is a feature, as they say, and while streaming responses have improved user experiences, this metric has become particularly crucial as models are deployed in production environments where user experience directly impacts adoption. 🔥 Inference compute: The amount of compute used for prediction shapes what problems a model can solve. More compute enables greater complexity but increases costs and latency - making it a pivotal benchmark for 2024. For some light holiday reading to explore this further: Service cards (OpenAI, Amazon), Meta's Llama 3 paper, and Anthropic's evaluation sampling research (links below).

When machine learning started moving into enterprise production, monitoring became non-negotiable. That was the moment when the industry began talking seriously about “drift” because models that worked in the lab often failed quietly in the real world. That led to well-known categories like data drift, model drift, and concept drift. They provide statistical signals that something in the input, model, or world had shifted. This framing worked because ML systems were largely predictive, stateless, and single-step. Drift was about “accuracy.” Fix the model, retrain, and move on. This idea of “drift” has fundamentally changed as we moved from machine learning models to AI agents. Fast forward to today, agentic systems break that assumption. Today, a new LLM version can change the performance of your prompts without any data shift. An agent can behave differently based on memory, tools, or execution context even when the model weights are unchanged. Drift now shows up as behavioral divergence. We see goal drift, where agents optimize for speed or completion instead of intent; environment drift, where API schemas, MCP edits, or tool behavior change under the agent’s feet; policy or reasoning drift, often caused by long-term memory contamination and reinforced shortcuts; and memory drift, where stale or one-off facts persist and quietly shape future actions. None of these are easily captured by traditional drift dashboards, yet all of them directly impact quality, safety, and trust. That’s why monitoring in the agentic era cannot remain model-centric. Drift management will move upstream: from statistics to intent, constraints, and control. We will need explicit goal anchoring, memory governance, environment contracts, and reasoning audits as first-class primitives in AgenticOps. The most important question will no longer be “Is the model still accurate?” but rather “Is the system still behaving as designed?” It’s a subtle shift from model mindset to system mindset. In systems that can act, learn, and adapt, drift is no longer just a performance issue; it is a governance and trust problem. PS: Experimenting to convert my posts into AI videos. Love to get the feedback if you feel it's adding value or not. #ExperienceFromTheField #WrittenByHuman #RecordedByAI

I’ve been interviewing more than a dozen of the top AI-native founders. The metrics they obsess over are shifting *away* from old SaaS metrics like LTV:CAC or MAUs. Seven alternative metrics that feel more relevant & urgent in 2026: Full deep dive in Growth Unhinged here: https://lnkd.in/ei9tgEgx 1. Monthly active users → Token consumption Perhaps the most unifying thing about AI companies is that they consume an immense amount of tokens. Tokens are a metric that’s hard to hide from. They show whether people are deeply using your AI product or if being AI-first is mere marketing jargon. 2. ARR → Gross profit per token Tokens are essentially fuel consumption. But gross profit shows how many miles you’ve driven with that fuel. 3. Product activation → AI quality Everyone can produce AI outputs for customers. That doesn’t mean these outcomes are any good, or that they’re improving over time. 4. Customer health score → Outcomes generated AI quality is a leading indicator. Outcomes prove whether customers are actually getting value out of the product. 5. Magic Number → Burn multiple You can’t hide from this with creative accounting or by grouping forward-deployed engineers into R&D spend. The more incremental ARR you generate for every $1 you burn, the better. 6. ARR per FTE → ARR per headcount $ AI companies have been extremely impressive about ARR per employee. But this is often paired with exorbitant compensation and/or six-figures a year in token spend per engineer. A better metric, in my opinion, is ARR per dollar spent on headcount. 7. Lifetime value (LTV) → First year value In 2026 a new Claude release can trigger a selloff across the entirety of enterprise software. I simply don’t trust the LTV of software products right now. The AI-native equivalent of 3x LTV:CAC: aim for the expected first year value to be >$0 (the higher, the better). --- SaaS metrics have never been set in stone. There’s not even a consistent definition for net dollar retention (NDR) among publicly traded companies. I hope this helps start a conversation about how to evolve. And drop into the comments if I missed your favorite next-era metric.

Everyone’s excited to launch AI agents. Almost no one knows how to measure if they’re actually working. Over the last year, we’ve seen brands launch everything from GenAI assistants to support bots to creative copilots but the post-launch metrics often look like this: • Number of chats • Average latency • Session duration • Daily active users Useful? Yes. But sufficient? Not even close. At ALTRD, we’ve worked on AI agents for enterprises and if there’s one lesson it’s this: Speed and usage mean nothing if the agent isn’t solving the actual problem. The real performance indicators are far more nuanced. Here’s what we’ve learned to track instead: 🔹 Task Completion Rate — Can the AI go beyond answering a question and actually complete a workflow? 🔹 User Trust — Do people come back? Do they feel confident relying on the agent again? 🔹 Conversation Depth — Is the agent handling complex, multi-turn exchanges with consistency? 🔹 Context Retention — Can it remember prior interactions and respond accordingly? 🔹 Cost per Successful Interaction — Not just cost per query, but cost per outcome. Massive difference. One of our clients initially celebrated their bot’s 1 million+ sessions - until we uncovered that less than 8% of users actually got what they came for. That 8% wasn’t a usage issue. It was a design and evaluation issue. They had optimized for traffic. Not trust. Not success. Not satisfaction. So we rebuilt the evaluation framework - adding feedback loops, success markers, and goal-completion metrics. The results? CSAT up by 34% Drop-off down by 40% Same infra cost, 3x more value delivered The takeaway: Don’t just measure what’s easy. Measure what matters. AI agents aren’t just tools - they’re touchpoints. They represent your brand, shape user experience, and influence business outcomes. P.S. What’s one underrated metric you’ve used to evaluate AI performance? Curious to learn what others are tracking.

🔍 Excited to share my latest Fortune column on why businesses need to stop chasing AI benchmarks and create their own evaluation frameworks instead. Here are the key insights for executives looking to make smarter AI investments: Misalignment of metrics: The celebrated benchmarks (like graduate-level reasoning and abstract math) rarely reflect actual business needs or represent truly novel AI frontiers Going deeper than performance averages: measure the spread of models’ performance—how often they get things wrong, and how badly. Building business-specific evaluations: Test potential models in the environments where they'll actually be deployed, using realistic data and scenarios Looking beyond accuracy: Consider the full spectrum of business requirements including speed, cost efficiency, flexibility, and regulatory compliance Creating a culture of continuous evaluation: Implement ongoing assessment to ensure AI solutions remain optimal while maintaining alignment with business objectives In my view, the path to AI success lies in developing evaluation frameworks tailored to your specific business objectives—essentially "a leaderboard for every user." The companies that master this approach will identify solutions optimized for their needs without paying premium prices for capabilities they don't require. https://lnkd.in/ezq6FD2R Always a pleasure collaborating with Theodoros Evgeniou from INSEAD and Maxwell Struever and David Zuluaga Martínez from BCG Henderson Institute on these insights into effective AI implementation. #AI #BusinessStrategy #AIEvaluation #Innovation #Leadership #DataDrivenDecisions  

AI has made velocity almost irrelevant as a performance measure. A piece of work can now be built in 20 minutes, but still take 25 days to reach production. Not because teams are slow, but because most of that time is not spent building anything at all. It’s spent waiting, for approvals, for dependencies, for prioritization decisions, for someone to stop starting new initiatives so existing work can actually move. And that waiting has real consequences. It burns money while teams are blocked but still staffed. It erodes trust when delivery slips for reasons no one can explain. It drains motivation when people are ready to work but unable to move. It creates a sense of constant activity without real impact in outcomes. Velocity no longer tells the story. Execution is no longer the constraint. Decision latency is. Approval friction is. Uncoordinated dependencies are. Too much work in progress is. Velocity is now the smallest part of delivery. Flow metrics don’t exist to optimize teams. They exist to make leadership delay visible, and to show where decisions, policies, and priorities are slowing the entire system down. That’s why this matters #NavigateYourFlow #FlowMetrics #Leadership #Delivery #AgileTransformation

Managing Performance in the Age of AI The way we measure performance has always been tied to the nature of work — and that nature has changed dramatically over time. First era → Operational work Tasks were repetitive, well-defined, and measurable. Quantity mattered more than anything else (with a baseline of quality). It was simple: assign the task, check if it’s done, measure output. Performance was visible almost instantly. Second era → Knowledge work Things got more complex — expertise and judgment started to matter. Here, quality became far more important than quantity (though both still played a role). Businesses and clients were willing to pay a premium for quality. Timelines stretched, but that was fine if the final project met the standard. Performance measurement was still manageable: look at the outcome, assess quality, and you knew whether the person delivered. Third era → Novel work (today, in the age of AI) AI has changed the game — quality and quantity can now be generated at scale, on demand, 24×7. The differentiator isn’t how much you produce, or even how polished it looks — machines already excel at that. The differentiator is creativity: solving new problems in new ways. And that’s far trickier to measure. Creativity doesn’t run on predictable timelines. You can’t guarantee novelty just by working harder or longer. A person could put in endless hours and still not create something truly new. Which means performance measurement has become slower and riskier. Businesses may have to wait much longer before they know whether someone can deliver in this “novel work” world. So what’s the answer? I believe we need to shift focus away from lagging indicators (outputs, outcomes) and start looking at leading indicators — early signs that a person has the mindset and skills to thrive in this new environment. Here are some early indicators: 1. Work ethics – discipline, consistency, ownership of outcomes. 2. Contribution in meetings – do they bring valuable perspectives, even if you need to draw it out of them? 3. Comfort with conflicting thoughts – can they handle ambiguity and contradictions without stalling? 4. Contextual understanding – do they grasp what’s happening around them and shape solutions that actually fit the situation? 5. Articulation – can they communicate ideas with clarity? In tomorrow’s world, articulation will matter more than knowledge, because knowledge is already democratized. In short → managing performance in the AI era means measuring what machines cannot. Not just IQ, but also EQ and SQ. Not just end results, but the early signals of creativity, adaptability, and alignment. This is harder — no doubt. But it’s also necessary. Because without evolving our performance measurement, we’ll keep applying old metrics to a new world — and miss what really drives value today. What do you think — are leaders and organizations ready to rewire performance management for this reality?

Your AI spend per employee keeps climbing. That does not mean AI is failing. It means the economics are changing. For the past 2 years, the business case was framed too simply: AI replaces headcount. Headcount is expensive. Therefore, AI saves money. That logic is now being tested. Not because AI cannot create value. But because many companies are replacing a cost they understand, labor, with a cost they do not yet fully manage: AI consumption. CFOs, CEOs, and boards are asking a harder question: Are we reducing the cost of work? Or just moving it to a different line of the P&L? Here is what is becoming clear: 1/ AI spend is now workforce planning AI budgets are no longer just software budgets. → They affect hiring → They affect productivity targets → They affect vendor spend → They affect operating leverage → They affect future headcount The question is now “What work are we redesigning, which roles are changing, and what outcome should AI produce?” 2/ Usage growth is not the problem More token usage is not automatically bad. If AI helps teams write better code, accelerate analysis, or reduce manual work, usage should rise. The issue is invisible token consumption. → Usage without measurement → Automation without workflow redesign → Agent activity without cost controls → AI adoption without unit economics 3/ The real metric is cost per outcome Leaders need to know what the spend is buying. → Cost per qualified opportunity → Cost per campaign launched → Cost per engineering task completedd → Cost per manual hour avoided Without that visibility, AI spend can look productive while becoming operational drag. 4/ AI is financially different than labor A salary is relatively predictable. AI consumption is variable. It scales with: → Usage → Task complexity → Retries → Context length → Model choice → Data volume → Agent autonomy It means AI has to be managed differently. 5/ Savings only happen when the work changes Cutting a role does not automatically eliminate the work. → Sometimes the work disappears. → Sometimes it gets automated. → Sometimes it reappears as AI spend, oversight, QA, exception handling, vendor management, or technical debt. The companies that win with AI will redesign the work. 6/ The people who remain become more valuable AI raises the value of people who know how to use it well. → People who design workflows → People who evaluate outputs → People who manage agents → People who connect AI activity to business outcomes AI strategy is also talent strategy. 7/ Token budgets need governance, not fear The answer is not to discourage AI usage. The answer is to manage AI consumption like a strategic operating expense: → Visibility → Accountability → Benchmarks → Controls → Clear business outcomes The question is “Are we using AI to reduce the cost, time, or complexity of valuable work?” Because if the work does not change, the savings may never show up. Save for future reference.