Why MYLO Keeps Campaign Creation Conversational but Reporting Operational

Creating a calling campaign and operating one are different kinds of work. Campaign creation begins with business intent that is often incomplete: who should be called, which connection flow fits, when calls may run, what caller identity is appropriate, whether recording is approved, and what should happen after no answer. Conversation helps people clarify that intent. Reporting, monitoring, pause, resume, and investigation need stable controls and exact records. MYLO deliberately separates those experiences.

Give Managers an Honest Summary

A manager may ask conversationally whether a campaign “worked.” MYLO can answer only after defining the period and business meaning, then reading current operational data. It should distinguish customers attempted, customers reached, successful bridges or IVR outcomes, unresolved calls, agent-side failures, and downstream CRM results. A concise answer is valuable, but it must link back to reproducible reporting scope and avoid turning an attractive percentage into unsupported certainty.

Handle Restart and Recovery Without Guesswork

After an appliance or service restart, the operational layer should reconstruct campaign state from durable MYLO records and current telephony evidence. It should not ask a model to remember what was running. Operators need to see whether a run stopped cleanly, which calls were active, which customers remain eligible, and what reconciliation is required before dispatch resumes.

The conversational layer can explain that recovered state and guide an administrator through the approved procedure. It should never declare remaining customers safe to retry when a customer leg has an uncertain outcome. Those cases stay visible until evidence or authorised review resolves them.

Measure the Boundary Itself

Review whether conversations are producing complete proposals, how often required information is missing, where users abandon setup, and which operational errors recur after approval. Those observations can improve questions and guidance. They must not be used to weaken approval, invent defaults, or hide difficult choices.

On the operational side, measure report freshness, reconciliation gaps, duplicate processing, unresolved calls, and time to verified recovery. This makes the division practical: conversation quality is judged by clarity and completeness, while operational quality is judged by deterministic correctness, traceability, and safe control.

Conversation Fits an Incomplete Request

A manager may begin with “Call these appointment customers tomorrow.” That sentence leaves critical choices unresolved. MYLO can ask focused questions, explain customer-first and agent-first flows, identify a missing recording decision, show capacity implications, and help the manager supply a clean lead file.

The conversational layer translates business language into a proposed supported operation. It does not directly rewrite telephony or silently start the campaign. Structured services validate values, calculate capacity, prepare the proposal, and preserve any uncertainty that still needs a human decision.

A Proposal Makes Intent Reviewable

Before approval, the proposal should state campaign purpose, lead count, selected connection flow, participating agents or destinations, caller-ID source, schedule and time zone, concurrency or pacing, retry policy, recording choice, and known exclusions. The user should see how these choices affect customers and staff.

If a required decision is missing, MYLO should stop and ask. It must not invent consent, choose recording, select an arbitrary number, or expand a schedule simply to make the request executable. The approved proposal becomes an immutable reference for what the operator authorised.

Deterministic Services Create the Campaign

After approval, controlled code performs the consequential work. It validates exact fields, confirms permissions and supported resources, records the operation, and creates campaign state through the defined application boundary. A language model is useful for explanation and clarification, but it is not the authority for campaign identifiers, lead state, agent capacity, or execution status.

This separation also supports recovery. If an AI provider is temporarily unavailable after a campaign was safely created, the authorised operator should still be able to inspect operational state and use deterministic controls that do not require new semantic interpretation.

Operations Need Stable Controls

Once a campaign exists, the urgent questions become exact: is it scheduled, running, paused, completed, or failed; how many customers remain; which agents are available; what happened in the latest run; and is new work still being dispatched? Operators need visible buttons, confirmations, identifiers, timestamps, and error states rather than a chat response that can vary in wording.

Pause and resume should use supported state transitions. Pausing should stop new dispatch while allowing active calls to finish according to the product contract. Resuming should continue eligible remaining work without rebuilding the campaign or losing attempt history. The interface must not imply that a chat instruction changed state until the operation confirms it.

Reporting Must Be Reproducible

A campaign report should return the same defined relationships whether viewed in MYLO, retrieved through the API, or reconciled into a CRM. The campaign, customer, run, call, call-leg, event, and recording identifiers give that result structure. Natural-language summaries can explain the numbers, but they must be generated from preserved operational evidence.

Every report should state its scope, time zone, filters, generation time, and treatment of ongoing or unresolved calls. Counts such as reached, answered, bridged, completed, no answer, and failed need explicit definitions. A model should not calculate an attractive success rate from ambiguous categories.

Separate Customer and Agent Outcomes

Two-leg calling makes operational precision essential. A customer-first flow can reach a customer and then fail to connect an agent. An agent-first flow can fail before the customer is dialled. If both become “failed,” a later process may call an already-contacted customer again or undercount agent capacity problems.

MYLO reporting should preserve leg-aware events and derive a business outcome through documented rules. The conversational interface may explain that outcome in plain language, but it should not overwrite the underlying evidence or approve a retry on its own.

APIs Support Operational Integration

Developers can use the authenticated reporting hierarchy for campaigns, customers, calls, runs, run reports, and authorised recordings. They should use the current MYLO API catalogue for exact schemas and permissions. Retrieval APIs make operational facts available without asking a language model to reproduce database state.

Calling APIs may initiate a defined supported call, but conversational campaign creation remains useful because it resolves business choices before many calls are scheduled. A CRM request containing a customer number is not automatically a complete campaign specification.

Conversation Still Helps After Creation

MYLO can explain why a run contains unresolved customers, compare pacing with available agents, summarise a bounded event history, or suggest which evidence an administrator should inspect. It can translate technical results for a manager and guide the next authorised step.

The explanation must cite current structured evidence and distinguish fact from inference. If the operational service is unavailable or data conflicts, the interface should say so. Chat should not claim that a campaign paused, a call completed, or a configuration changed unless the deterministic operation confirms it.

Keep Roles and Approvals Visible

Campaign design, approval, operation, and reporting may belong to different roles. A configuration administrator should not automatically be the campaign operator; a reporting user should not automatically be able to start calls. MYLO should enforce permission at the operation boundary, not rely on wording in the conversation.

Consequential steps need an attributable approval record containing the proposed values, approving identity, time, execution result, and verification evidence. A rejected or failed proposal remains part of the audit history. Do not overwrite it to make the campaign history appear cleaner.

Design the Operator’s Daily View

The daily screen should foreground state, progress, active calls, remaining eligible customers, exceptions, capacity, recent runs, and clear pause or resume actions. It should make stale data and refresh time visible. Drill-down should connect a metric to customers, calls, legs, events, and recordings only where the role is authorised.

Use conversation as contextual help beside the operational view, not as the only control surface. An operator under pressure should not have to remember a phrase to stop new dispatch or ask repeatedly whether a change succeeded.

Production Boundary Checklist

1. Use conversation to clarify incomplete business intent and explain trade-offs.
2. Convert intent into a complete structured proposal with visible omissions.
3. Require appropriate approval before creating or starting consequential work.
4. Use deterministic services for validation, state transitions, and identifiers.
5. Provide explicit operational controls for monitor, pause, resume, and recovery.
6. Build reports from campaign, run, customer, call, leg, event, and recording relationships.
7. Let AI explain evidence without becoming the source of operational truth.
8. Retain immutable proposal, approval, execution, and verification records.

The boundary is simple: conversation is best where the person is still deciding what the business means; operational interfaces are best where the system must show exactly what happened. MYLO combines both so a small team can create confidently and then run, measure, and investigate campaigns without ambiguity.

From Conversation to Controlled Operation

Conversation gathers complete Level-1 information: campaign name, lead file and mobile column, one of five connection types, required agent destinations or active IVR, compliance acknowledgement, and an explicit recording choice. MYLO prepares a proposal; creation approval persists the campaign; a functional test proves the selected path; and a separate Start approval releases deterministic runtime. The Campaigns page then owns pause, resume, status and reporting rather than becoming a second creation authority.