Most COOs are drowning in AI use case decks.
You have slides that say things like:
- 40+ GenAI use cases
- 12 functions impacted
- Millions in “potential” value
And yet, a year later:
- A handful of pilots
- A few ChatGPT shortcuts
- Very little measurable impact on cycle times, cost, or quality
McKinsey’s recent work on GenAI for operations is blunt: the companies that capture outsized value are the ones that rewire how work is organized and governed, not the ones with the longest use case lists.
Their State of AI research adds: for most organizations, the shift from pilots to scaled impact is still “a work in progress,” even as adoption rises.
At the same time, Gartner finds that 70% of chief data and analytics officers are now responsible for AI strategy and operating model, which tells you AI is moving from project status into core operating model territory.
Put differently:
You do not have a use case problem.
You have an operating model problem.
This post is about that shift, specifically from a COO lens.
1. The “100 Use Case” Slide That Never Turns Into Change
You have probably seen some version of this pattern:
- Consultants or internal teams generate a big use case backlog
- Each function gets its own GenAI ideas
- A few proofs of concept launch in parallel
- Twelve months later, you have:
- Some cool demos
- A pilot here or there
- No real change in how work actually flows
Why this keeps happening:
- Use cases are scoped function by function
- The core value chains (order-to-cash, quote-to-order, hire-to-retire, ticket-to-resolution) are not redesigned
- There is no single owner of “AI in operations.”
- Governance is fragmented, so pilots never graduate to “how we do things here”
McKinsey calls out exactly this trap: future-facing COOs go beyond use case catalogs and focus on end-to-end operating model design for GenAI and Agentic AI.
2. Why Use Case Thinking Stalls In Real Operations
Use cases are a great starting point. They are a terrible ending point.
Three structural reasons they stall:
1) Fragmented ownership
- Sales owns their GenAI ideas
- Ops owns theirs
- HR and Finance have their own lists
- IT and data teams own models and infrastructure
BUt:
- Nobody is accountable for end-to-end impact on a value stream
- No one is on the hook for “order-to-cash cycle time” or “time-to-fill” across all the AI pieces
2) Pilots that live in the wrong place
- A GenAI pilot improves a step inside a function
- Upstream, the process still starts late and with bad data
- Downstream, nobody changes how decisions are made
Result:
- Local improvements
- Global outcomes unchanged
- COOs see no material difference in throughput, error rates, or unit economics
3) No unified way to govern AI in operations
Most orgs have:
- A slide with “AI principles”
- A security review process
- A list of pilots
They do not have:
- A standard lifecycle for AI in operations: from idea, to test, to rollout, to retirement
- A clear view of where AI is allowed to act, where it only recommends, and where humans stay firmly in control
- A simple dashboard that ties AI work to a short list of operational KPIs
So even good pilots become orphans.
3. What An AI-Ready Operating Model Actually Looks Like
McKinsey’s COO guidance reads less like “do more AI” and more like “change how operations work with AI built in.”
Translated into plain language, an AI-ready operating model has three big ingredients.
A. Clear roles for COO, CDAO, and IT
Instead of diffuse ownership, you see patterns like:
- COO
- Owns value streams and operational KPIs
- Decides where AI belongs in the flow of work
- CDAO
- Owns data, models, and AI operating model standards
- Ensures quality, monitoring, and reuse of components
- CIO/CTO
- Owns platforms, integration, and security
Gartner’s CDAO research reinforces this: CDAOs are being formally tasked with AI strategy and operating model, which only works if the COO is a co-owner, not a bystander.
B. A standard pattern for how AI touches processes
Across journeys you see the same pattern:
- Observe
- AI watches events, logs, tickets, transactions
- Recommend
- AI surfaces suggested actions, triage, next best step
- Act
- In low-risk situations, AI can act directly under defined rules
- Learn
- Feedback loops from humans and outcomes update the system
This gives you:
- A reusable way to plug AI into order-to-cash, hire-to-retire, case-to-resolution, etc
- A consistent way to design human in the loop, not reinvented every time
C. Real governance, not theater
An AI operating model is visible in:
- A small portfolio of AI in operations, prioritized by value and feasibility
- A clear promotion path:
- Idea
- Trial in part of the journey
- Limited rollout
- Standard way of working
- Guardrails that cover:
- Risk and compliance
- Change management and training
- Value tracking: which KPI should move and by how much
Without this, you are just adding clever tools into chaotic processes.
4. Practical Moves A COO Can Make This Quarter
This does not need a multi-year transformation plan to start.
Here are moves that fit in a single quarter.
1) Stand up a small “Ops+AI council”
Keep it tight:
- COO or head of operations
- CDAO or head of data
- CIO/CTO or head of platforms
- HR or people leader
- One or two functional leaders from big value streams (for example, CX, revenue, supply chain)
Mandate:
- Prioritize AI work in operations
- Decide where to test end-to-end journeys
- Set simple guardrails and metrics
2) Pick one end-to-end journey as a testbed
Examples:
- Order-to-cash
- Hire-to-retire
- Ticket-to-resolution
- Lead-to-renewal
For that one journey:
- Map the current flow, including systems, handoffs, and failure points
- Mark where AI already exists (or where people are using shadow tools)
- Mark where AI could:
- Take over rote tasks
- Recommend actions
- Help humans see risk earlier
3) Define one operating model template for agentic/automation work
For that journey, answer a few concrete questions:
- Where is AI allowed to act automatically? Under what thresholds and rules?
- Where must a human approve or review?
- Who is accountable if an AI-powered step fails: which team, which leader?
- Which KPIs must move for this to be considered a success?
Once you have this template working in one journey, you can reuse it elsewhere.
5. Where Business Operations Advisory Fits
This is the zone where our Business Operations Advisory work lives: not in making the AI use case deck bigger, but in making your operating model ready for AI.
In practice, that looks like:
Diagnose
- Map the current operational reality:
- Systems and workflows
- Bottlenecks and fire drills
- Shadow spreadsheets and shadow AI
- Identify where AI would currently magnify chaos vs where it could stabilize and accelerate
Decide
- Prioritize a small number of value streams and journeys
- Decide for each:
- Fix first, then automate
- Automate as is, with guardrails
- Leave it alone for now
Design
- Define roles and rhythms between COO, CDAO, IT, and line leaders
- Create lightweight runbooks for:
- How AI gets proposed, tested, and promoted into standard operations
- How humans stay in the loop for high-risk steps
- How value is tracked over time
The outcome you want is simple:
- Fewer random experiments
- Fewer pilots that never grow up
- More visible, end-to-end changes in how work actually gets done
6. A Quick Self Check For Your AI Portfolio
Pick one AI initiative in your current portfolio and ask:
- Does anyone own the full value stream this use case lives in?
- If this pilot “works,” do we know what will actually change in the way work flows?
- If volume doubled tomorrow, would we feel confident in this AI enhanced process, or would we be nervous?
If the honest answers make you uneasy, that is not a reason to slow down on AI.
It is a reason to start treating AI as a COO level operating model problem, not a list of features.
That is where Business Operations Advisory at Scale Crew HR earns its keep: turning “lots of use cases” into an operating model where AI has a real job, a real owner, and real impact on how the business runs every day.
