# [[Evos MOC]] Evos builds autonomous operator systems for legacy industries. Not co-pilots. Not dashboards. Not chatbots that summarise. Systems that actually do the work, the way a 20-year domain expert does it. The thesis: AI needs expertise, not just data. The unlock: codifying decades of tacit operational knowledge into bespoke AI systems that run 24/7. Founded by Urav Shah (4x founder, $8.7M raised across previous ventures, Investec early-stage tech entrepreneur of the year). Pre-seed: $1.5M SAFE at $12M cap. 23 years old. ## What is an Operator An operator is not a chatbot. It is not a co-pilot. It is not a workflow automation tool. An operator is an autonomous system that owns an entire operational function end-to-end. It doesn't assist a human doing work. It does the work. Humans shift from executing tasks to directing outcomes and handling exceptions. The distinction matters: **Chatbots** (ChatGPT, Claude): general-purpose. You prompt, they respond. They have no memory of how your company works. They don't know your processes, your edge cases, your vendor quirks, your escalation paths. They search and summarise. They don't do the job. **Co-pilots** (GitHub Copilot, Microsoft Copilot): embedded in a single tool. They help you do a task faster inside that tool. Still reactive. Still need a human in the loop for every decision. They augment output, not replace it. **Workflow automation** (Zapier, Make): static rules. If X then Y. No reasoning, no judgment, no adaptation. Still needs consultants to set up and humans to babysit. Breaks when anything changes. **Agents**: goal-driven, can reason and take actions autonomously. But most agents today are generic. They don't understand your domain, your processes, or how your specific business actually works. They're smart in general, dumb in your context. **Operators** (Evos): domain-specific, function-specific, intelligent. They are trained on how your experts actually think and work. They understand the judgment calls, the exceptions, the "we always do it this way because..." institutional knowledge. They sit on your existing stack (ERP, TMS, WMS, CRM) and operate across your real infrastructure. Every operator is bespoke to that company. No two are the same. The gap Evos fills: agents have intelligence but lack domain expertise. ERPs have domain structure but lack intelligence. Operators have both. ## The Macro Shift: Outputs to Outcomes The way enterprises are managed is changing at a fundamental level. For decades, management meant managing outputs: how many reports filed, tickets closed, shipments processed. The unit of work was the task. Humans did tasks. Software helped them do tasks faster. The shift happening now: **Manual → Software-assisted → AI-augmented → Autonomous** We are entering the autonomous era. The unit of work is no longer the task. It is the outcome. You don't manage "process 500 invoices." You manage "accounts payable runs at 99.5% accuracy with 2-day cycle time." The operator owns the outcome. Humans direct the system and handle what the system escalates. This is [[Counter Positioning]] at the management layer. Companies that grew up managing outputs (headcount, hours, tasks completed) struggle to switch to managing outcomes (SLA met, error rate, cost per unit of work). The whole org structure, incentive system, and reporting layer is built around outputs. Evos gives them the system to make the switch without rebuilding everything. ServiceNow calls this the "Autonomous Workforce." Deloitte calls it the "Human-Agentic Workforce." The framing doesn't matter. The structural shift does: roles, skills, and career paths get rebuilt around outcome ownership, not task execution. But here's what the data actually shows: the shift is structurally inevitable yet practically stalling. Anthropic's [[Observed Exposure — Anthropic's AI Penetration Metric]] separates theoretical AI capability from real-world deployment. Office and admin roles sit at 90% theoretical coverage but only 14% observed penetration. Computer and mathematical occupations: 94% theoretical, 33% observed. The gap is enormous. This is not a capability problem. It is an adoption and product problem. The technology can do the work. Enterprises can't figure out how to let it. That gap — between what AI can theoretically handle and what actually gets automated — is the entire Evos market opportunity. Every percentage point of that gap is a function that should be running autonomously but isn't, because nobody has captured the domain expertise needed to make it work. ## The Rise of Operators in the Enterprise Legacy industries (logistics, manufacturing, retail) lose $3T+ annually to manual processes, rework, and downtime. 15 hours per week per employee spent on manual, repetitive tasks. 1.9M unfilled operations jobs in the US alone. They can't hire their way out. The expertise is retiring. The workforce pipeline is dry. And the work is not the kind that generic AI handles well, because it requires deep domain judgment built over decades. The hiring data confirms the structural shift is already biting. Anthropic's observed exposure research shows hiring for 22-25 year olds in AI-exposed occupations has slowed by 14% since 2023. The entry-level pipeline is drying up precisely in the roles where AI penetration is highest. Companies aren't replacing junior staff; they're waiting for automation that actually works. But that automation isn't arriving through chatbots or co-pilots. It requires operators. 95% of enterprise GenAI projects have failed to deliver measurable business impact (MIT 2025). Why? Because they bolted AI onto broken processes. Consultancies spent billions advising on "AI transformation" without changing the underlying operating model. The consultancy model itself is the problem: expensive, slow, outdated by deployment, and org change is their product, not operational improvement. And the market being disrupted is far larger than most investors realise. Per Coatue's thesis in [[AI Eats Services Not Software]], software is a $0.2T market. Services are $5.5T — a 25x delta. AI doesn't just eat software. It eats services. The entire TAM for autonomous operators isn't the SaaS market. It is the management consultancy market, the staffing market, the outsourced operations market. Evos is not competing with Salesforce or SAP. It is competing with Accenture, McKinsey, and the 500,000 junior consultants who fly in on Monday and leave on Thursday. Current solutions don't get close: - **AI consultancies**: advise, don't do. Very expensive, extremely slow, recommendations outdated by deployment. - **AI-enabled software (ERPs)**: inform, don't act. Show you the problem, don't solve it. - **Generic AI co-pilots**: search and summarise. Interface doesn't know how to do the work. - **Workflow automation tools**: static rules. No reasoning, no autonomy. The expertise that allows AI to actually act lives in people, not software. That is the insight. Every other approach tries to work around this. Evos works through it by capturing the expertise directly. ## How an Operator Gets Built Evos compresses what used to take months of consultancy into hours: **Assess** (3 hours): Agent-led workshops capture domain expertise from your team. Maps processes, pain points, exception handling, judgment calls. Connects to your systems live. Not a consultant with a whiteboard. An AI system actively learning how your experts think. **Define**: Identifies highest-value AI use cases specific to your company. Not a generic playbook. Specific to your operations, your stack, your workflows. **Deploy** (~4 hours): Builds autonomous operator systems on your existing infrastructure. 100+ enterprise integrations (ERP, TMS, WMS, CRM). Live in 24 hours. **Optimise** (ongoing): Graduated autonomy. Systems start supervised. Earn trust through performance. Autonomy grows as the operator proves reliability. Think of it like onboarding a new hire: you don't give them full authority on day one. You let them earn it. This is the key difference from consultancy-driven transformation. Consultancies sell a plan. Evos ships a working system. The transformation happens through use, not through a PowerPoint deck and a 6-month change management programme. ## How Transformation Has to Change The old model: hire McKinsey or Accenture. 6-12 month engagement. $2-10M. They interview your people, write a report, recommend "AI transformation," help you buy some software, run change management workshops. By the time it deploys, the recommendations are stale. 95% of these projects fail. The new model has to be experiential, not advisory. You don't transform by being told what to do. You transform by doing it. The software itself is the transformation. This is where [[The Gap of Imagination]] becomes structurally important. Alvaro Higes (CEO of Luzia, 65M users) nails the problem: three things must happen for AI to deliver value — the AI must be able to do the task, the user must discover it can, and the user must decide to try. The industry has solved general AI capability. It has largely solved awareness. It has completely failed at value extraction. Today's AI delivers roughly 70% of what a great executive assistant can do. Most enterprise users get 0% of that value, because they cannot imagine what the system can do for them. They default to asking ChatGPT to summarise an email and stop there. Two product failures compound this. First, companies build for demos, not defaults. The AI can theoretically handle complex operational workflows, but nobody designs the product so it actually does this from day one. Second, nobody closes the imagination gap. The user has to picture a workflow they've never seen and configure the AI to do it. That's an absurd ask for a logistics manager who's been doing the job manually for 20 years. You need what Christensen called a 10x better experience to change a default behaviour. Evos collapses the consultancy-to-deployment pipeline into a single product experience: - Expertise capture (what consultants do in months) happens in 3 hours - System deployment (what IT teams do in quarters) happens in hours - Adoption (what change management programmes do in years) happens through graduated autonomy: the system proves itself, humans learn to trust it, autonomy expands naturally Critically, Evos closes the imagination gap by design. The user does not need to imagine what an autonomous operator looks like. Evos shows them. The 3-hour assess workshop is not just expertise capture — it is a live demonstration of what autonomous operations look like in their specific context, with their data, on their systems. The operator doesn't ask the user to configure it. It configures itself from the expertise capture and starts working. The user's job shifts from "figure out what AI can do" to "tell the AI where to look." That is the 10x experience shift. This is [[Consultancy-to-Platform Transition]] compressed by AI-native architecture. The transformation is not a project. It is a product. You don't buy advice. You buy an operator that works. The consultancy industry sits at a precarious inflection point. Their core economic model (armies of junior staff, billing by the hour, protecting the status quo) is structurally misaligned with a world where AI can capture expertise in 3 hours and deploy a working system in 24. They advise. Evos acts. ## First Principles Three core beliefs: **1. AI needs expertise, not just data.** This is the central insight. Every other enterprise AI approach assumes that if you feed enough data into a model, it will figure out how to do the work. It won't. Operational data tells you what happened. It doesn't tell you why a veteran logistics manager routes shipments a certain way when weather patterns shift, or why a manufacturing floor supervisor adjusts the production sequence on Fridays. That judgment is tacit, learned over decades, and lives in people's heads. Evos captures it through structured conversation, not data pipelines. **2. Bespoke by default.** No two companies operate the same way. The same industry, same function, same ERP, completely different processes. This is why horizontal SaaS fails in operations: it imposes a generic template on a messy, human, path-dependent reality. Evos builds every operator system tailored to that specific company. The system learns how you work, not how companies in general work. **3. Graduated autonomy earns trust.** You don't flip a switch from human to AI. That's why 95% of enterprise AI projects fail: they try to automate everything at once and the org rejects it. Evos starts supervised. The operator proves itself on easy tasks, earns trust, gets promoted to harder ones. Adoption is built into the product, not bolted on as change management. ## Directional Arrow of Progress **Conversation → Deployed Operator → Proven Value → Expanded Systems → Outcome Ownership** The speed is the wedge. Live in 24 hours, not months. First deployment results: 15 hrs/week reduced to 2 hrs/week per employee. ~$20K saved per employee annually. But the arrow points beyond cost savings. The trajectory: Phase 1: **Cost reduction**. Automate repetitive operational tasks. Prove ROI fast. This is where Evos is now. Phase 2: **Operational expansion**. One operator works, deploy more across the same company. 2 of 5 pilot companies already asked to expand pre-launch. Phase 3: **Outcome ownership**. Operators don't just do tasks. They own KPIs. The company manages outcomes, not people doing work. Phase 4: **Institutional intelligence**. The operator network across a company becomes the operating system for how the enterprise runs. The codified expertise is the company's competitive advantage, embedded in software. Long-term, this is a new category: the autonomous operations layer. Sits between the ERP (system of record) and the workforce (system of execution). Evos becomes the system of intelligence. ## Growth Flywheel **Domain-Specific, Function-Specific, Intelligent Operators → Data + Context → Models + Implementation → Transformation x Adoption → back to Operators** Breaking this down: **Operators generate proprietary data + context.** Every operator system running in production produces a stream of domain-specific operational data paired with the context of why decisions were made. This is not generic training data. It is judgment-in-action, tagged with outcomes. **Data + context feeds models + implementation.** The proprietary data improves the underlying models. But more importantly, it improves the implementation patterns: how to capture expertise faster, how to deploy systems more reliably, how to handle edge cases in specific industries. The capability library grows. **Better implementation drives transformation x adoption.** Faster deployment, higher hit rates, more trust, faster graduated autonomy. The product improves and transformation becomes easier. Adoption isn't a separate workstream. It is a byproduct of the product working. Each deployment also narrows [[The Gap of Imagination]] for the next customer in that vertical — the capability library already knows what operators look like in their context, so the 3-hour assess becomes even more of a live demo and less of a blank-slate discovery. **Adoption compounds back into more operators.** Customers expand. New customers in adjacent verticals adopt faster because the capability library already has relevant patterns. PE firms roll Evos across portfolio companies. The flywheel accelerates because the bottleneck in enterprise AI adoption has always been expertise capture and trust-building. Evos compresses both. Each turn of the flywheel makes the next turn faster. This is structurally similar to [[Azraq Data Flywheel]]: proprietary operational knowledge (not just data) creates a compounding advantage that widens over time. [[Wright's Law]] applies: every doubling of deployed operator systems compresses the cost and time of the next deployment. ## Moat and Differentiators Evos stacks multiple [[7 Powers]]: **Cornered Resource**: Codified domain expertise. This is tacit knowledge from real operators, structured as executable AI capabilities. No public dataset contains how a 20-year logistics veteran handles exceptions. Every deployment adds to a proprietary capability library. Competitors cannot replicate this without doing the same painstaking capture work, customer by customer. Evos capability library already scores +11.88pp higher than Claude on verified eval scenarios for domain-specific tasks. **Process Power**: Conversation to deployed autonomous system in under 24 hours. The full pipeline (Assess → Define → Deploy → Optimise) is deeply integrated and hard to copy. The graduated autonomy model builds institutional trust that takes time to earn. This is not a model you can replicate by having better AI. It is a system that compounds through use. **Counter Positioning**: Every incumbent is trapped. Consultancies sell advice and billable hours; switching to a product model cannibalises their revenue. ERPs sell licenses and show you data; giving you an autonomous system that acts on that data undermines the need for their analytics add-ons. Workflow automation sells static rules; admitting that reasoning and autonomy are needed means rebuilding from scratch. Each incumbent would have to destroy their own business model to compete. **Switching Costs**: Once Evos captures your domain expertise and builds operator systems on your stack, switching means losing all that codified institutional knowledge. The operator holds your operational DNA: how your people think, your exception handling logic, your vendor relationships, your seasonal patterns. Starting over with a competitor means re-doing the expertise capture from zero. And the system gets better with every interaction, so the switching cost deepens over time. **Scale Economies**: Software margins (~85%+ gross margin, no model/compute costs to Evos) with service-level outcomes. Unit economics: ~$500 CAC (founder-led), <1 month payback, ~$360K LTV (24-month, no expansion). As the capability library grows across customers in the same vertical, new deployments in adjacent companies get cheaper and faster. The marginal cost of each new operator falls. **Structural moat via the imagination and exposure gaps**: Beyond the 7 Powers framework, Evos benefits from two structural barriers that protect its position. The [[Observed Exposure — Anthropic's AI Penetration Metric|observed exposure gap]] means the theoretical capability for AI to automate operations already exists — but real-world penetration lags massively (office/admin: 90% theoretical, 14% observed). Every competitor that ships a generic tool and expects the user to configure it will run headfirst into [[The Gap of Imagination]]: users cannot picture what autonomous operations look like, so they never adopt. Evos is one of very few companies architected to close both gaps simultaneously — expertise capture closes the capability gap, experiential deployment closes the imagination gap, and graduated autonomy closes the trust gap. Companies that only solve one of these three will leak value at the other two. These powers reinforce each other. Proprietary expertise (cornered resource) feeds faster deployment (process power), which proves value faster (switching costs), which expands within accounts and across PE portfolios (scale), while incumbents watch from business models they can't abandon (counter positioning). Competitors face a cold-start problem: they need codified expertise to build competitive operators, but they need competitive operators to attract the customers who provide the expertise. ## Traction (March 2026) - 5 pilot deployments (logistics, manufacturing, retail). Secured within 2 weeks of cold outreach. - 30+ qualified pipeline from 6 weeks of cold outbound. 12+ enterprises didn't qualify for the waitlist. - 15 hrs → 2 hrs/week per employee: first deployment results. - 2/5 pilots expanding to multiple operator systems pre-launch. - Customers asking to invest and join the team. - 3 months from concept to proven product and demand. Solo founder. - Target: 50+ paying customers and $3M+ ARR in 12 months. ## Business Model Two pricing options, both delivering software margins at service-level outcomes: **SaaS**: $5K/month per operator system. 4 seats included. Month-to-month or annual. ~$180K ACV for 3 operators. **Outcome-based**: $2,500 setup + 20% of value created monthly. Tied to mutually agreed KPIs. If value isn't created, don't pay. Zero downside for the customer, higher revenue ceiling per deal. The outcome-based model is the land strategy: aligned incentives, zero risk for buyer, and it likely converts to SaaS once trust is established. This is the business model version of graduated autonomy. Both models map directly to the emerging pricing architectures in [[AI Eats Services Not Software]]. Coatue identifies three pricing models displacing traditional SaaS seats: the enterprise model (sell outcomes to the enterprise buyer), the outcome model (charge per result delivered), and the fractional employee model (price as a fraction of the human it replaces). Evos's SaaS tier is the enterprise model. Evos's outcome-based tier is literally the outcome model — billing for value created, not access granted. And the ~$20K/year savings per employee positions each operator as a fractional employee at a fraction of the cost. The structural insight: Evos is not selling software. It is selling work. The $5.5T services market is the real TAM, not the $0.2T software market. Every dollar currently going to consultancies, outsourced ops, and manual processes is addressable. The companies that win this market are the ones that price per output AND close the imagination gap — Evos does both. ## Key Links - [[7 Powers]] - [[Counter Positioning]] - [[Wright's Law]] - [[Azraq Data Flywheel]] - [[Consultancy-to-Platform Transition]] - [[Defensibility Principles MOC]] - [[First Principles and Mental Models MoC]] - [[Industrial AI MOC]] - [[Bespoke Engineering in Industrial AI]] - [[Industrial AI Unit Economics]] - [[AI era Defensibility]] - [[AI Eats Services Not Software]] - [[Observed Exposure — Anthropic's AI Penetration Metric]] - [[The Gap of Imagination]] --- Tags: #deeptech #kp #systems #firstprinciple #AIstrategy #verticalAI