Identify the work to delegate
We work with your teams to pinpoint the activities that eat up time and the tasks where delegation can deliver a tangible benefit.
We assess how often they occur, how complex they are, what information they require, and what an error would cost. You come away with clear priorities, an initial scope, and criteria to measure results.
| Work to delegate | How it can be handled | Human role |
|---|---|---|
| Customer requests | Identify the topic, look up the relevant information, and handle routine requests. | Take over disputes and special cases. |
| Periodic reporting | Gather the data, run the defined checks, and draft the commentary. | Validate the analysis and decide on actions. |
| Invoice follow-up | Flag overdue invoices and prepare or send authorized reminders. | Make the call on sensitive accounts and payment terms. |
| Document processing | Extract the information, check that nothing is missing, and record it in the right system. | Review inconsistencies and approve sensitive transactions. |
| Monitoring and tracking | Watch selected sources, summarize what’s new, and flag relevant events. | Assess what they mean for the business. |
Design the interfaces and the human touchpoints
We define how users hand off work, track its progress, and step in, whether from a messaging app, a business application, a portal, or a dedicated interface.
The solution can start on demand, on a schedule, or when an event occurs. Employees have what they need to review results, approve an action, correct a request, or take back control.
Build the architecture, workflows, and agents
We design an architecture suited to the work being delegated and the level of autonomy you want.
| Approach | How it works | Possible use |
|---|---|---|
| Automated workflow | Runs predefined steps and rules. | Send a notification when a deadline is reached. |
| AI-enabled workflow | Uses AI for certain steps within a defined process. | Read a request, classify it, and draft a reply for approval. |
| AI agent | Chooses some of its own steps and tools to reach a goal, within authorized limits. | Review a case file, find missing information, and take permitted actions. |
These approaches can be combined with human approvals.
We also plan the environment in which agents run and are controlled, known as the harness. It manages their access to information and tools, keeps track of the state of the work, enforces controls, and handles approvals, errors, and retries.
Connect your data and tools
We define the connections your applications and information sources need: CRM and sales systems, accounting, documentation, messaging, or databases.
Each solution gets access only to the information and functions its job requires, with the appropriate permissions. Technology choices take into account your existing environment, your security requirements, and running costs.
| Category | Example solutions | Possible role |
|---|---|---|
| Enterprise-connected agents | Dust | Carry out tasks using internal knowledge and take action in connected tools. |
| Agent builders | Microsoft Copilot Studio | Configure agents that can handle requests and take authorized actions. |
| Automation and orchestration | n8n — Germany | Connect applications, chain steps together, and build in human approvals. |
| Agents built into business software | Salesforce Agentforce | Handle sales or customer service activities within the configured scope. |
| Agents and software robots | UiPath | Combine analytical capabilities with actions executed in applications. |
| Custom agents | Mistral Agents API — France | Build agents that use tools and coordinate multiple actions. |
Sergio Castagna Consulting is technology- and vendor-agnostic. The European and international solutions mentioned here are examples meant to illustrate what’s possible. Our recommendations depend on your needs, your constraints, and your existing environment.
Set up governance and secure every action
We define who is accountable for the delegated work, what the solution can do on its own, and which situations require approval.
Security is built in from the start: data protection, limited access, controls on actions, and protection against malicious instructions hidden in the content being processed.
Restrictions are enforced at the system level. Mechanisms are in place to suspend activity, revoke access, and handle incidents.
Roll out the solutions and support your teams
We run tests on real-world scenarios, including exceptions and potential failures, then support a phased go-live.
Your people learn to oversee delegated work, handle approval requests, and step in when a situation falls outside the defined scope. Responsibilities for operations, maintenance, and support are defined before rollout.
Oversee execution and measure results
We set up tracking for quality, turnaround times, costs, and incidents. Actions and approvals are logged so they can be audited.
Benefits are assessed net of the time spent on oversight and corrections. Changes to models, rules, or connections are tested before they go live.
Start with a clearly defined piece of work
A first engagement can focus on a single recurring task or a limited workflow. Together, we define the expected outcome, the scope, the deliverables, and the success criteria before considering a broader rollout.