How to Build an Operating Model for the Agentic AI Future

In this webinar

An agentic AI operating model is the structure an organization uses to govern, build, secure, and measure a mix of rules-based automation and AI agents. Most automation CoEs were designed for deterministic RPA, where every input of a given type produces the same output. Agents behave differently, so risk, delivery, staffing, and value all need new rules.

In this session, Ashling's Head of Advisory and Enterprise Architect break the model into 6 modular pillars. 

What you'll learn:

• Where traditional automation operating models strain once agents enter the pipeline
• How to tier AI risk by data sensitivity, autonomy, and business impact so low-risk agents ship faster
• Why skills badging scales a team better than fixed roles
• How to put a dollar value on cycle time, avoided fines, and customer retention
• What to do this quarter in each of the 6 pillars

Why Traditional Automation Operating Models Strain Under Agentic AI

Traditional operating models were built for deterministic processes. They assume rules-based bots, focus on cost reduction and documentation, and bundle strategy through build into one model.

The agentic AI operating model is designed for a blend of deterministic automation and AI agents. It balances risk, cost, and upside. Its components are modular, so you can augment the CoE you already run without starting over.

Strategic Direction & Governance

Strategic direction and governance sets the program's goals, decision rights, and funding logic. Its 4 components are vision and strategy, program governance, a benefit realization framework, and change management.

The most common friction point: leadership wants to move fast on AI while risk and security teams still assess every use case the traditional way. Agents pile up in a governance backlog, and the program slows. Benefit realization is where most of the work goes. Labor cost alone misses what agents deliver. If speeding up customer onboarding by 3 days raises retention by a few percent, multiply that lift by customer lifetime value. A cycle-time metric that used to sit in days now has a dollar figure leadership can prioritize.

Proof from the field: A super-regional bank was struggling to adapt its automation CoE to agentic AI. Ashling updated its benefit realization framework to treat token spend as a variable cost per run, replacing the fixed-cost model built for RPA. Every initiative now traces back to enterprise priorities, and the program delivered more than $2.4 million in value in 2026.

Key artifacts: AI vision, AI roadmap, governance and reporting structure, benefit realization framework, AI and automation funding model, and communication and stakeholder engagement plan.

Architecture & Security

Architecture and security decides what an agent can safely do and how it gets approved. Its components are the solution design document (SDD), a business user guide template, architecture review, and a tiered AI risk assessment.

Documentation carries more weight for agents. A screen recording can document an RPA bot. An agent needs its data sources, inputs, and decisions written down so its behavior stays traceable and explainable.

Ashling adapts failure mode and effects analysis (FMEA), a Six Sigma method from manufacturing, to score each agent on severity, likelihood, and detectability. Multiplying the 3 scores produces a risk priority number (RPN) that ranks risk and points to the right mitigation. An agent that gathers public research data for human analysts scores low. An agent that handles account takeover with client financials scores very high.

Proof from the field: A utilities provider running separate automation and AI CoEs needed one security standard for both. A governance committee defined criteria for 4 risk tiers, and every agent now gets an FMEA score at ideation. Low-risk agents can clear approval in about 2 weeks, compared with 8 weeks for complex, high-autonomy ones.

"There's a lot of overlap between those two CoEs. In general, we're trying to do the same thing, but having to incorporate two different standards because there's two different governing bodies involved becomes complex." Kevin Huggard

Key artifacts: SDD, platform build and provisioning documentation, security policies, data archiving and logging policies, business user guide template, and tiered AI risk model.

Talent & Velocity

Talent and velocity sets how much a team can deliver and who maintains it afterward. Its components are organization and people (roles and responsibilities) and a competency tracker with a badging process.

Roles are blending. With coding assistants, analysts and project managers build more, and developers work directly with business stakeholders. Ashling tracks skills over job titles. Each critical skill (UiPath, Microsoft, Python, data engineering, architecture) has tiers, and people earn badges as they prove capability. A silver badge in data engineering tells a planner exactly which projects someone can take on.

"I've been asked before as a consultant, doesn't that work you out of a job? And the answer is frankly, yes. But that's kind of the goal." Filip Swat

Proof from the field: A group benefits provider wanted to scale automation and AI quickly without depending on outside help long term. A competency tracker matched existing skills to program needs. It showed where Ashling support was needed now and which work the client team could take over later. The provider now plans its pipeline around real team capacity, with dynamic pods that scale up and down.

Key artifacts: organization and people chart, roles and responsibilities matrix, training and enablement plans, and badging framework.

Pipeline & Priority

Pipeline and priority decides what gets built and with which tool. Filip calls it the beating heart of any operating model. Its components are an intake approach (top-down engagement plus bottom-up ideation), opportunity assessment (value and effort), and a tech decisioning framework that matches each opportunity to RPA, low-code, GenAI, or agentic AI.

T-shirt sizing and directional benefit estimates keep assessment moving and prevent analysis paralysis. A shared tech decision framework also keeps ideas from bouncing between teams debating whether something is an agent or a low-code app.

Proof from the field: A private equity owned hotel chain needed to modernize its back office and grow revenue across finance, guest management, and operations. Ashling ran discovery workshops with agent-aided interviews, scored every idea on value and effort, and routed work to the client's existing ERP and CRM wherever they fit. The result was an $8M pipeline that prioritizes high-value, low-complexity solutions.

Key artifacts: opportunity intake methodologies, opportunity assessment framework and business case, prioritization criteria, and tech decisioning framework.

Delivering Outcomes

Delivering outcomes makes delivery consistent, reusable, and safe. Its components are a delivery methodology, coding and logging standards, technical best practice, repeatable project plans, and an AI testing plan.

Sprints suit agentic work because teams often can't predict how a model will behave until the data is connected. Testing is the hardest shift.

"Traditional code will give you an error. If it fails, it will very visibly break. An AI agent very rarely breaks in an obvious way. They'll just give you a wrong answer or a partial answer or a low-confidence answer." Filip Swat

An AI testing plan defines what counts as an error: a low-confidence flag, a second agent checking the first, or more intensive human review.

Proof from the field: A global commercial real estate firm had competing requests across regions and service lines and no sprint planning method. Ashling introduced DevOps sprint planning with planning poker, led discovery workshops, and built a reusable component library. The firm now has a $12.8M pipeline of high-value, low-complexity solutions and shorter delivery cycles.

Key artifacts: delivery methodology and engagement model, delivery lifecycle document templates, solution design principles and development standards, repeatable project plans, coding and logging standards, and AI testing plan.

Value Tracking & Insights

Value tracking and insights keeps the program honest. It answers whether the work performs as expected, whether anyone sees it, and whether the business cares. Its components are a value framework, innovation scanning, and quarterly impact reports with success stories.

Credible measurement starts with the right metric. Moving from a 1% sample to full review can look like hundreds of millions of dollars under a minutes-saved calculation, which overstates the impact. Stories carry results further. Filip has seen a single success story turn a struggling program around.

Proof from the field: An investment advisory firm needed to measure automations that catch processing errors before they become fines or reputational damage. Ashling's automation moved quality control from sampling under 1% of transactions to monitoring 100%. Value is measured in avoided fines and NPS detractors, and dashboards show every error found and corrected. Delivered value now exceeds $54 million.

Key artifacts: value tracking dashboard and reporting, innovation radar and trends assessment, and quarterly impact reports and success stories.