Overview

You've successfully deployed three major AI systems in your operations function. Procurement AI is preventing supplier quality issues before they happen. Compliance AI is detecting regulatory violations with 96% accuracy. Demand Forecasting AI has improved inventory turns by 18%. The systems are working. The business is seeing real value. Your peers are watching, and they're asking questions. Finance leader: "We have 40 people handling invoice exceptions. 70% of their time is spent investigating discrepancies. Can AI help?" HR leader: "We get 200 resumes per opening. Screening is expensive. Can AI help?" Legal leader: "We review 50 supplier contracts annually. Standard clauses take hours to review. Can AI help?" Suddenly, you're the AI expert in the company. Everyone wants AI. But you have a choice: (1) become the universal AI delivery team, build systems for every function, hire more and more people, become the bottleneck that everyone depends on, or (2) position operations as the internal AI consultancy that helps other functions build their own systems, scales without scaling headcount, keeps standards and governance intact, and builds AI capability across the entire organization. This chapter teaches you how to choose option 2 and execute it brilliantly.

The difference is crucial. In option 1, operations builds everything. Finance's invoice system, HR's resume screening, Legal's contract review, operations' responsibility. When a system fails, operations is blamed. When it succeeds, operations is credited, but also becomes responsible for maintaining it forever. Operations becomes essential and stretched. In option 2, operations consults and governs. Finance builds their invoice system (with operations guidance on risk and data quality). HR builds their resume screening (with operations oversight). Legal builds their contract review (with operations ensuring fairness and compliance). When systems succeed, functions are credited and take responsibility. Operations is the trusted advisor and quality gatekeeper, not the delivery engine. This model scales infinitely. The more functions you support, the more AI capability exists across the organization, without operations headcount growing proportionally.

The AI Consulting Model: From Delivery to Advisory

Imagine two organizations, both with successful operations AI. Organization A built delivery model: Operations has 15 people. Finance asked for invoice automation. Operations took ownership, hired data scientists, built the system, now maintains it. HR asked for resume screening. Operations built that too. Supply chain asked for demand forecasting. Operations built all three systems and now runs all three. Operations has become the AI delivery engine. When Compliance needs violation detection, Operations has to hire more people because there's no capacity. Every function's AI initiative depends on operations. Operations is the bottleneck.

Organization B built consulting model: Operations has 8 people. Finance asked for invoice automation. Operations spent 2 weeks understanding Finance's problem, helped Finance build a business case, advised on data quality, reviewed Finance's approach, helped train Finance's team. Finance hired a vendor (Docusign for contract automation, Thoughtworks data scientists, whatever). Finance owns the system. HR asked for resume screening. Operations screened the idea (good fit for AI, data is available, HR is committed), helped HR develop business case, recommended vendors, advised on bias testing. HR chose a vendor. HR owns the system. When Compliance needs violation detection, Operations provides consulting support and governance oversight. Operations capacity has barely grown. The organization now has three functioning AI systems (invoice, resume, violations), each owned by the relevant function, each governed by operations' standards, and operations' team of 8 is not stretched.

This consulting model is how you scale. Organizations that scale AI most successfully move from delivery to advisory.

In the consulting model, Operations provides: Strategic guidance (is this a good AI opportunity?), governance frameworks (how do we ensure this is done safely?), risk assessment and templates (what could go wrong?), data quality guidance (is our data good enough?), vendor evaluation support (should we build or buy?), model design review (is the approach sound?), compliance oversight (is this still within standards?), training on standards, shared learning (lessons from one function help other functions), best practices repository (playbooks and templates for reuse).

The business function provides: Domain expertise (they understand their problem), project ownership and sponsorship (they're accountable for success), budget (they fund the project), user engagement (they'll actually use the system), adoption responsibility (they drive adoption). When Finance's invoice automation succeeds, Finance's leader says "We built this with operations' governance and guidance." When it fails, Finance owns the problem, not operations.

Why this matters operationally: In delivery model, operations' team gets pulled in 10 directions. In consulting model, operations provides scale without growing headcount proportionally. A small team can support a large portfolio of AI initiatives across many functions because they're not building everything. They're advising, governing, and sharing learning.

Intake and Screening: Not Every Problem is an AI Problem

The first critical responsibility as consultant is screening. Not every business problem is an AI problem. Not every AI problem is the right fit for your organization right now. Your job is helping functions understand what's feasible and what isn't, and preventing wasted effort on problems that won't benefit from AI.

Phase 1: Initial Consultation (30-60 minutes)

Business function leader pitches their problem. You listen for: What specific problem are they trying to solve? Who experiences the problem? What's the current cost (money, time, quality)? What would success look like? Why do they think AI is the solution? You ask questions: "Walk me through a specific invoice exception. What goes wrong? How long does it take to resolve?" or "Take a specific resume. How long does screening take? What mistakes happen?" Questions are more valuable than listening to their prepared pitch because they force clarity.

You provide rapid feasibility check: Is this actually an AI problem or a process problem? Do we have the data needed to train a model? Is it technically feasible? What's the ballpark timeline and cost? Is the business case compelling enough to justify the investment? Do we have stakeholder support? Any regulatory constraints?

Common problems that sound like AI problems but aren't: "We make too many mistakes", maybe the process is poorly designed, not an AI candidate. "We want to automate this completely", maybe you need process redesign first, not AI. "We want to predict what customers will do, but we have no historical data", not feasible. "This is our competitive advantage; we can't automate it", probably shouldn't. Understanding these distinctions prevents wasting time.

Outcome of Phase 1: Go/no-go decision. If go, schedule detailed assessment. If no-go, explain why clearly and suggest alternatives. You might say: "The invoice exception problem is 70% reconciliation (data quality issue) and 30% genuine judgment. Fix the data quality first, then revisit AI."

Phase 2: Detailed Assessment (1-2 weeks of operations time)

Operations digs into the problem with the function. You're studying: (1) Current process in granular detail (not how they describe it, but how it actually happens). (2) Data availability and quality: What data exists? Where is it? How fresh is it? Can you access it? How accurate/complete/consistent is it? What cleaning is required? (3) Stakeholder landscape: Who will use this system? Who will resist? Who approves decisions? Who's impacted by errors? (4) Regulatory and compliance requirements: Are there laws constraining how you can use this data or build this AI? (5) Success metrics: How will you measure whether the AI system works? What's "good enough"? (6) Risk profile: What could go wrong? What's the cost of errors? How would errors be caught?

Outcome of Phase 2: Assessment report including: Feasibility rating (high/medium/low), risk score, data quality assessment, estimated timeline and cost, recommended governance category, risks and mitigation strategies, and final go/no-go recommendation. This report is fact-based, not opinion. It answers: "Can we actually do this? Should we?"

Phase 3: Business Case Development (if assessment is positive)

Operations works with Finance to build the business case. Financial analysis: What's the cost-benefit? Quantify: "Invoice exceptions take 120 minutes per person per week (60 people × 120 = 12,000 minutes). At $50/hour that's $100K/month in labor. If AI eliminates 80% of exceptions, that's $80K/month saved." Then subtract cost of AI system (vendor license, internal time, maintenance). Is ROI compelling enough to justify investment? Also include: ROI projection, timeline, resource requirements, risks and mitigation, governance plan, success metrics, exit strategy (what if it doesn't work).

Phase 4: Approval and Funding Decision

Function's leadership and operations leadership review the business case together. Decision: Go (the function secures budget and starts project) or no-go (the function doesn't pursue this right now). If it's a high-risk project (autonomous decisions affecting people, regulatory-sensitive), it goes to steering committee for final approval.

Important: Operations provides consulting and governance at no charge (it's part of operations' responsibility). But the function funds the actual AI system development (either vendor licensing or hiring data scientists). This prevents functions from treating operations as unlimited free labor and helps them think carefully about whether the project is worth the investment. A function that's willing to pay for AI is more committed than a function expecting operations to build it for free.

Service Menu: Defining Operations' AI Consulting Services

Be explicit about what operations offers. This prevents scope creep, sets expectations, and helps functions understand what they're responsible for. Your service menu should have two tiers: standard services (included for all projects, part of governance responsibility) and premium services (available for cost, function pays for them).

Standard Services (included, no charge to the function): Initial feasibility screening (consultation with function leadership), detailed assessment support, governance framework guidance and approval process navigation, risk assessment methodology and risk register development, data requirements analysis and data quality assessment, vendor selection support and RFP review (if they're buying), model design review (is the proposed approach technically sound?), bias testing methodology training, compliance monitoring and quarterly audits, testing methodology guidance, monitoring and alerting design, incident response playbook development, change management and adoption strategy consulting, training on operations AI standards.

Premium Services (available at cost, function's budget): Building the AI model (operations' data scientists do the work), data engineering (preparing data for model), infrastructure and hosting (running the system), ongoing model management (retraining, monitoring, optimization), change management execution (training, communications).

This menu clarifies that operations is the quality gatekeeper and advisor (standard services), but functions are responsible for actually building and running their systems (premium services or they hire vendors). This distributes work appropriately.

Knowledge Transfer and Training: Building AI Capability Across the Organization

Part of operations' value is teaching other functions how to do AI responsibly. Create structured training that functions must complete before their AI systems launch.

Required curriculum (all functions' AI projects must complete): AI governance and approval processes (1 hour), risk assessment and bias testing (2 hours), data requirements and data quality (1 hour), model evaluation and testing (2 hours), monitoring and alerting (1 hour), incident response and escalation (1 hour). Total: 8 hours. It's mandatory before approval. This ensures consistency and prevents functions from making predictable mistakes.

Advanced curriculum (for functions doing multiple projects): Building your function's AI capability, vendor management and contracting, ongoing model management and drift, advanced testing strategies. These are available after the basic training for functions that are scaling AI internally.

Making training mandatory before approval sends a message: AI projects don't just happen. They go through a process. Standards matter. Consistency matters.

Shared Learning Repository: Your Competitive Advantage

As you help multiple functions deploy AI, you accumulate knowledge. Centralize it in a searchable repository. This becomes one of operations' most valuable assets.

Repository should contain: Use case playbooks (how to build resume screening, how to build invoice automation, how to build violation detection, specific steps, templates, common pitfalls), data quality best practices, risk assessment templates and examples, testing checklists, monitoring templates and dashboards, incident response playbooks, vendor scorecard and comparison, lessons learned from every completed project. When your second resume screening project launches, the team doesn't start from scratch. They use the playbook from the first project and launch 50% faster. When your third invoice automation project launches, it's even faster and better because you've learned from the first two.

This repository compound in value over time. After 10 projects, operations has documented thousands of hours of learning. New functions can access all that learning without repeating the work. This is a competitive advantage that grows with every project.

Managing Demand: Saying No and Setting Boundaries

As operations becomes the go-to AI consultancy, demand will exceed capacity. You need clear boundaries and a prioritization process.

Establish transparent prioritization criteria: Strategic alignment (does this advance corporate strategy? high=yes, medium=partially, low=no), financial impact (what's the ROI or value created? high=$1M+, medium=$100K-1M, low=$0-100K), risk reduction (does this reduce enterprise risk? high=major, medium=moderate, low=minor), resource availability (do we have capacity?), technical feasibility (can we actually do this?). Score each incoming initiative against these criteria. The initiatives with highest scores get prioritized.

Establish a quarterly intake pipeline with clear timelines: "We have capacity for 2 major AI projects this quarter. If you want to be considered, submit your intake form by April 15. We'll review and prioritize by April 30. Selected projects start May 1. Decisions are final." This prevents every function from assuming they'll get what they want and keeps your team from being overwhelmed by unlimited requests.

Set clear boundaries about support level: "We provide governance and risk assessment. You hire data scientists or use a vendor." "We advise on your data strategy, but you own data quality." "We help design your monitoring and incident response, but you operate them." This distributes the workload and prevents operations from becoming a bottleneck doing everything for everyone.

The clearest way to communicate boundaries: "We are not building AI systems for you. We are helping you build them. You own the outcome." This distinction is crucial. Functions that own their AI systems care about success. Functions that have operations build for them treat it as operations' responsibility when things fail.

Scaling the Model: From One Function to the Whole Organization

Don't try to support all functions simultaneously. Start with one function (Finance's invoice automation is a common first choice, high value, good data, clear success metrics). Work through the entire process with them: intake, assessment, business case, implementation, governance monitoring. Document what worked and what was hard. Document lessons learned. Refine your process based on learning. Then move to the next function with improved processes and templates.

After three or four successful projects, you have: documented intake process, assessment templates, business case templates, risk assessment templates, testing checklists, monitoring templates, incident response playbooks, and lessons learned from multiple domains. You're now a mature consulting operation. New functions can be supported much faster because the infrastructure exists. A function that's your first (took 6 months from intake to launch) is your longest. Your fourth function might launch in 3 months because you're not reinventing the process.

This is the scaling advantage of consulting model. Each project makes the next project faster and better.

What to Do Monday Morning

• Define your service menu: standard services (consulting/governance, no charge) and premium services (building/operating systems, function pays). Be explicit about what operations does and what the function does.
  - Design your intake and screening process. Create a simple template that functions fill out when they have an AI idea. Questions: What problem? Why AI? What data? What's the business case?
  - Create assessment and business case templates. Standardize your approach so functions know what to expect and you work efficiently across projects.
  - Develop required training curriculum. What must every function's AI team know before launching? Create 8-10 hours of required training.
  - Start a shared knowledge repository. Document lessons learned, playbooks, templates, vendor comparisons. Make it searchable and accessible.
  - Pick your first consulting client. Don't try to support all functions simultaneously. Choose one (usually Finance, supply chain, or compliance), work through the entire lifecycle with them, document lessons learned.
  - Establish clear prioritization and boundary-setting processes. Define how you'll prioritize competing requests. Be clear about what operations does vs. what the function does.
  - Set up quarterly review and intake pipeline. Regular cadence for reviewing demand, making allocation decisions, and managing expectations.
  - Train your operations team to consult, not just deliver. Consulting is different from delivery. Invest in coaching and skill development for your team.

Key Takeaways

• Transition from "operations builds all AI" to "operations consults on all AI" to scale impact without growing headcount proportionally.
  - Screen ruthlessly. Not every problem is an AI problem. Use assessment process to separate good opportunities from wasted effort.
  - Define a service menu: standard services (consulting/governance, no charge), premium services (building/operating, function pays).
  - Make functions own their AI systems. Operations provides expertise and governance oversight. Function funds and operates.
  - Design intake process that forces clarity about problems, data, business case, and commitment before work begins.
  - Train all functions on operations AI standards before their systems launch. Consistency and standards matter.
  - Build shared knowledge repository that grows with every project. After 10 projects, new functions benefit from accumulated learning.
  - Set clear boundaries about capacity and support level. You cannot do everything for everyone. Be explicit about what you're not doing.
  - Start with one consulting engagement, master the model, document lessons, then scale. Your first client teaches you how to serve all clients.
  - Position operations as trusted advisor and quality gatekeeper, not delivery bottleneck. This is the mindset shift that enables scale.