Choose work with a clear beginning and end
Describe the trigger, inputs, steps and final output. For example, an internal reporting workflow might gather approved figures, draft a summary and present exceptions to a manager. This is an illustrative architecture, not evidence of a HELIX deployment.
Map the existing process first. If a reliable fixed rule solves the problem, ordinary automation may be enough. Several agents add coordination, running cost and additional failure points; use them when the task benefits from distinct roles rather than because the label sounds more advanced.
Name the owner and permitted access
Identify the sponsor, day-to-day reviewer and person who can pause the workflow. List the systems it may read or write, the records it needs and the actions it must never take without approval. Keep access limited to the defined job.
Do not send API keys, customer exports or confidential documents in the initial enquiry. Discovery should establish data permissions, retention, supplier requirements and appropriate access. Regulated decisions and sensitive uses need specific review; a general pilot brief does not establish compliance.
Define the difficult cases
Use permission-cleared examples representing normal work, missing information, conflicting records, duplicate events and unavailable tools. Define how the system asks for clarification, reports uncertainty and hands control back to a person.
Approval points should name the actual action: sending an external message, publishing a claim, accepting a cost or altering a customer record. Record what was proposed, approved and executed. A draft appearing in an interface is not the same as a successful external action.
Agree acceptance and costs before expanding
Measure whether the output is correct and useful, how much human correction it needs, task completion and cost per completed job. Set thresholds from your actual requirements and baseline; do not borrow an unsupported accuracy or time-saving percentage.
Your written scope should distinguish discovery, the pilot build, model usage, hosting, connected subscriptions and optional support. HELIX multi-agent builds start from £4,999 for a defined pilot scope. Extra workflows, scale, resilience or procurement requirements need discovery and a separate proposal.
Plan handover and a fallback
Specify who owns the code and records under the agreement, how permissions are withdrawn, how logs are reviewed and how ordinary work continues when a provider is unavailable. Include a stop control and a clear support arrangement.
A useful first enquiry names the workflow, current tools, approximate volume if known, main difficulty, accountable owner and desired outcome. Larger-company interest does not automatically mean the starting pilot is suitable for an enterprise-wide deployment.