Chapters 03–04

Where AI creates value

From individual tasks to end-to-end workflows: the three levels of AI value, and what AI can actually do.

03. Where the value sits

Horizontal AI optimizes tasks. Vertical AI transforms workflows.

Horizontal AI

Spread across everyone

Examples
Enterprise copilots, chatbots, writing assistants, meeting summaries, search, content generation
Primary value
Individual productivity
Profile
Easy to deploy · broad adoption · low process redesign · diffuse business impact
Vertical AI

Built into one process

Examples
Procurement, customer service, credit-risk assessment, supply-chain planning, sales qualification, research workflows
Primary value
Business process performance
Profile
Function-specific · integrated with enterprise data · embedded in workflows · potentially much higher impact · harder to scale

Horizontal AI optimizes tasks.

Vertical AI transforms workflows.

Report finding · McKinsey / QuantumBlackFewer than 10% of vertical use cases make it past the pilot stage. The value is higher, and so is the barrier.
04. The three levels of AI value

Each level changes more of the work, and creates more of the value.

Level 1

GenAI Assistance

The human stays responsible for the workflow. AI helps with drafting, searching, summarizing and recommending.

Human↓AI assistance↓Human action
Illustrative estimate · reportCall center example: 5–10% better resolution time and productivity.
Task productivity
Level 2

Agentic Automation

Agents execute selected activities inside the existing process.

  • Classify requests
  • Collect information
  • Trigger workflows and update systems
  • Propose actions, resolve simple cases
Process→Human→Agent→System→Human
Illustrative estimate · reportCall center example: 20–40% time savings, 30–50% backlog reduction.
Workflow optimization
Level 3

Process Reinvention

The process is redesigned from the ground up around human and agent collaboration.

Goal↓ Agent orchestration layer ↙ ↓ ↓ ↘
AgentAgentSystemData
↓Human oversight↓Business outcome

Agents can

  • Coordinate multiple activities
  • Work in parallel
  • Make bounded decisions
  • Monitor changing conditions
  • Resolve routine cases
  • Escalate exceptions

Humans focus on

  • Judgment
  • Oversight
  • Exceptions
  • Strategic decisions
  • Relationships
Illustrative estimate · reportCall center example: up to 80% of common incidents resolved autonomously, 60–90% faster time to resolution.
Business transformation

Estimates are McKinsey / QuantumBlack's illustrative figures for a hypothetical call center, not measured results. They show the order of magnitude between levels, not a forecast for any organization.

What can AI actually do?

Every AI use case comes down to one of three capabilities. Naming it makes the conversation concrete.

PredictEstimates what will happen.Demand, risk, churn, fraud, delays.
GenerateCreates new content.Text, summaries, code, answers, drafts.
ActTakes steps toward a goal across systems.This is what agents add: execution, not only insight.

Start with the problem, not the technology

Run every idea through three questions, in this order.

  1. Business problemWhich outcome is at stake, and what does the problem cost today?
  2. AI fitDoes it need predicting, generating or acting? Is the data there, and is the risk acceptable?
  3. ValueIs the value measurable, with a sponsor and a KPI?

If one answer is no, stop. Not every problem needs AI.

NextProcessRedesign the process. Don't speed up the old one.