The operating architecture your AI tools need
Operating architecture is the designed interaction between how you earn money, how you deliver work, and how your technology supports both. Most firms skip this layer. That is why their tools don't stick.
The workflow gap
McKinsey found that the 6% of organisations achieving real margin impact from AI share a common pattern. They did not simply adopt AI tools. They redesigned their workflows around AI capabilities.
Among high performers, 55% had fundamentally restructured how work flows through their organisation. Among the broader market, only 20% had done so. That is a 3x gap in the single most important predictor of AI value.
The implication is direct. If you are asking “how do we get people to use AI?”, the better question is: which workflows should we rebuild?
Six layers of operating architecture
An AI-ready operating architecture spans six layers. Each one matters. Skip any layer and the architecture is fragile.
Value model
How the firm captures value. Hours, fixed fees, subscriptions, outcome fees. How AI-driven efficiency appears economically. This layer determines whether AI savings flow to the client or the firm.
Workflow design
The defined path from client need to delivered outcome. Where decisions happen, who makes them, and what information they need. Where AI operates and where humans operate.
Data and knowledge architecture
How documents, decisions, and precedents are captured, structured, and reused. What becomes context for AI tools. Without this layer, AI tools hallucinate or produce generic output.
Technology stack and integration
The systems of record. How AI tools connect to those systems. How data flows between tools without manual intervention. This is where most firms start — and where most get stuck.
Roles, skills, and guardrails
Who is responsible for what in a world where AI drafts, summarises, and suggests. What must be reviewed by humans, and by whom. This layer protects quality and manages risk.
Incentives and governance
The metrics that encourage new behaviours. Governance for approving new tools, changing workflows, and updating guardrails. Without this layer, change is fragile.
Case study: proposal automation
Consider how these layers work together. A professional services firm generates proposals regularly. The typical process involves senior people spending four to six hours per proposal — gathering client context, selecting services, calculating pricing, and writing the document.
The proposal workflow has seven distinct sections:
1. Trigger and identification
Structured data capture. Fully automatable.
2. Discovery synthesis
AI summarises and structures conversation notes. A human checks the interpretation.
3. Product selection
Human decidesWhich services fit this client? Requires judgment about fit, scope, and sequencing.
4. Complexity scoring
Human decidesHow complex is this engagement? Sets the pricing tier and shapes the delivery plan.
5. Pricing lookup
Pricing follows defined rules once complexity is set. A calculation.
6. Content generation
AI generates the proposal draft from templates and context. Mechanical.
7. Document generation
Final formatting, review, and delivery. A human reviews and approves.
Seven sections. Only two require human judgment. A four-hour process becomes 45 minutes of focused decision-making.
But this only works because all six layers are in place. The value model defines how pricing connects to complexity. The workflow defines which decisions need human input. The data layer provides the context AI needs. Remove any layer and the automation breaks.
What “integrated” actually means
When AI is integrated into a workflow, three things are true:
It sits inside the workflow.
Not as a separate tool people switch to, but as a step in the process they already follow.
It is owned.
Someone is responsible for how AI operates in that workflow. They monitor output quality, update context, and make decisions about scope. Unowned AI drifts and degrades.
It moves one metric.
Cycle time, first-pass quality, cost per deliverable, client satisfaction. If you cannot name the metric AI is improving, the integration is not real.
These three criteria separate genuine integration from surface adoption. The firms seeing real value from AI tend to meet all three. The rest are running pilots.
Start with one workflow
You do not need to redesign your entire operating architecture to start. Choose one workflow that is repeatable, measurable, and commercially important. We will help you diagnose it against these six layers and rebuild it with intention.