Today
Today, AOS is the operating model behind Deloitte Digital's agency services: clients buy workflows from Deloitte; practitioners run them using AOS.
Defining AOS
AgencyOS
AOS is the agency in a box.
AOS is the agency in a box. Most AI marketing tools sit on top of an agency and accelerate parts of it. AOS is the agency itself: Deloitte Digital's full marketing agency, with AI agents running every workflow and intelligence compounding with every engagement, so the work gets faster, cheaper, and smarter the longer it runs.
Today, AOS is the operating model behind Deloitte Digital's agency services: clients buy workflows from Deloitte; practitioners run them using AOS. Tomorrow, clients license AOS directly and run the workflows themselves. Same product, two commercial wrappers; every engagement today hardens AOS for tomorrow.
The outcomes
Three operational outcomes the CMO measures directly. AOS commits not just to delivering them, but to improving them the longer the system runs:
Time-to-market
Lower time-to-market per output
Cost
Lower cost per output
Throughput
Higher throughput
All three improve over time.
The problem
Marketing has a lot of repeatable work that today gets done expensively and slowly by FTEs whose experience doesn't transfer between engagements. AOS turns that repeatable work into agentic workflows that run cheaper, faster, and smarter the more they run.
The vision
Tomorrow
Tomorrow, clients license AOS directly and run the workflows themselves. Same product, two commercial wrappers; every engagement today hardens AOS for tomorrow.
Who AOS is for
Buyer
The CMO. The commercial decision-maker who engages Deloitte.
Operator
The Worker. Today, a Deloitte practitioner operating AOS on the client's behalf.
02Architecture
How AOS is structured. What clients buy.
AOS is built on seven marketing capabilities. Workflows are the SKU: what clients buy. Each workflow breaks into tasks, the atomic unit of work. Outputs are the deliverable end-state the client receives. Two engine layers power everything underneath.
Architecture mode
Capabilities
Work column
Run an example workflow
Seven capabilities at the top organize all marketing work. Each is a marketing domain. A distinct kind of marketing work. Together they close the universe of marketing work. Audience cuts (B2B), industries (healthcare, retail), and channels or tactics (SEO, paid social, podcast) cross-cut the capabilities as workflow parameters; they do not expand the L0 layer. Each capability carries its own domain-trained agents, accumulated domain intelligence, and domain-specific governance.
Workflows are repeatable processes that produce defined client outcomes. Housed in one capability (the address where buyers find them); tap multiple capabilities during execution. Tasks are the atomic units of work that compose workflows. Outputs are the deliverable end-state of workflow execution: the artifact or operating condition the client receives, counts, and measures.
The engine powers everything. Two engine layers sit underneath the work layers and power every workflow the same way regardless of which capability houses it. The AOS Operating Layer is the machinery. Closest analogy: an operating system. macOS doesn't write your documents, but every app on your Mac uses macOS for the universal jobs. AOS's Operating Layer is the OS; each capability's workflows are the apps that run on it. The AOS Intelligence Layer is the body of knowledge AOS draws on at every gate. Four types: functional, client, Deloitte, learned. The Operating Layer is constant; the Intelligence Layer grows with every workflow that runs. The Intelligence Layer is what compounds.
What clients buy
Workflows. An engagement can be a single workflow or a collection of workflows packaged as a retainer or a transformation engagement.
Clients buy workflows. A workflow can be a project (a one-and-done deliverable like Brand Architecture Build) or a cadence (ongoing work like CRM nurture or performance reporting). An engagement bundles one or more workflows. Pricing is per workflow.
03Buyer's journey
From objective to learning loop. Seven beats.
The buyer's path through AOS. Hover or click any beat to read it in full.
1
The buyer states an objective.
"I need a brand refresh." "I need to acquire customers in a new segment." "I need to stand up a CRM program."
04Operator's journey
The Worker is the human at the boundary.
The Worker is the human at the boundary of every workflow. The Worker never sits inside the workflow. The engine (Operating Layer + Intelligence Layer) and the agents do the work. The Worker bounds them in across two phases of the engagement and a set of five roles.
Act 1
Phase 1. Configuration.
The Worker provides context. This is the longest single human activity in the engagement, and it happens once per client before any workflow runs.
The agents arrive pre-programmed; functional intelligence is built into them. Deloitte intelligence (industry research, benchmarks, first-party and licensed data) routes into the Intelligence Layer automatically. Neither is configured per client. The only thing built per client is client intelligence: brand, voice, audiences, products, history, competitive landscape, business rules. The Worker assembles it working alongside intake agents that interview, scrape public material, and structure brand documents into the format the Intelligence Layer can act on. The Worker reviews and approves what the intake agents produce; the Worker stays the source of truth.
The Worker also configures the workflows the client is buying: which workflows are on, what cadences they run, and who authorizes each one. When Phase 1 closes, AOS is ready to run.
Worker
The human
- Assembles client intelligence (brand, voice, audiences, products, history, competitive landscape, business rules)
- Works alongside intake agents that interview, scrape public material, and structure brand documents
- Reviews and approves what intake agents produce; stays the source of truth
- Configures workflows: which are on, what cadences, who authorizes each one
System
AOS
- Agents arrive pre-programmed with functional intelligence
- Deloitte intelligence routes into the Intelligence Layer automatically
- Client intelligence populates as the Worker provides context
- Workflows stage for the run phase
AOS is ready to run.
Act 2
Phase 2. Run.
Workflow type
Worker
Steps in at defined moments. Never sits inside the workflow.
Workflow executionEngine running
Agents execute. Operating Layer governs. Intelligence Layer informs.
Output, then
Output, then loop or terminate.
When a workflow completes, it produces an output. What happens next is a property of the workflow, not a per-run Worker decision.
Cadence
Cadence workflows loop.
CRM nurture, performance reporting, optimization recommendations. These keep running on their cadence. Learned intelligence feeds back into the Intelligence Layer between runs; the next run is smarter.
Learned intelligence routes back to the Intelligence Layer
Project
Project workflows terminate.
Brand Architecture Build, GMMM Build. These finish when the deliverable is produced. Learned intelligence still feeds back into the Intelligence Layer for adjacent workflows.
Learned intelligence routes back to the Intelligence Layer
The work getting faster, cheaper, smarter the longer AOS runs is emergent from Phase 2 running over time. The Worker does not trigger it; it is on by default.
Boundary
What the Worker does not do under 1.3.
Govern the agents (Operating Layer does). Evaluate agent output for quality (Operating Layer does). Enforce brand code (Operating Layer does). Sit alongside agents while they work. Interrupt workflows mid-task. When a workflow seems to require a Worker inside the task sequence, that signals a system gap; the gap gets logged and closed.
The Worker is a Deloitte practitioner. AOS runs as Deloitte Digital's marketing agency; the Worker is the practitioner operating it on the client's behalf.