AI has a role here. Authority is not it.

Dialed uses artificial intelligence to help wholesale coffee roasters handle operational work. This policy explains what AI does in the Service, what it does not control, the limits you should expect, and the safeguards built around it.

This policy supplements our terms of service and privacy policy. If it conflicts with the Terms, the Terms control.

AI may help Dialed:

• classify incoming business messages and documents;
• extract fields such as products, quantities, prices, dates, and destinations;
• match information across orders, invoices, payments, and connected systems;
• draft order confirmations, collection messages, summaries, and explanations;
• surface exceptions, missing information, and work that needs review; and
• propose a structured action for Dialed’s trusted systems to evaluate.

Dialed may use different models or providers for different bounded tasks. We evaluate them for the job they perform and may change them as the Service develops.

The part of Dialed that reads and reasons cannot move money, approve an action, or write directly to your business systems. It can prepare a typed proposal. A separate trusted plane validates that proposal, applies your permissions and policy, checks whether human approval is required, executes through the authorized connector, and records what happened.

Some routine actions may execute under rules your organization has configured. Actions that require approval can be approved only by an authenticated human with the necessary permission. An AI agent cannot approve its own proposal or turn an untrusted message into authority.

AI output may be inaccurate, incomplete, inconsistent, or out of date. It can misread a document, confuse a unit, map the wrong product, miss context, perform faulty arithmetic, or produce confident-looking text that is not supported by the source. A polished answer is not proof that it is correct.

Ambiguous emails, scans, handwriting, unusual templates, local abbreviations, changing catalogs, and missing records make mistakes more likely. Confidence signals help prioritize review, but they are estimates and not guarantees.

You and your authorized users are responsible for:

• keeping connected records, mappings, permissions, and business rules current;
• reviewing exceptions, approval requests, unusual transactions, and material decisions;
• checking output against source records before relying on it where an error could matter;
• using appropriate accounting, security, quality-control, and continuity practices; and
• reporting corrections so the workflow and its controls can be evaluated.

Dialed is an operational service, not legal, tax, employment, investment, medical, or other professional advice. Do not use it for emergency, personal, or safety-critical decisions.

A café, supplier, document, or linked webpage may contain text that looks like an instruction to Dialed. The ingestion system treats that content as untrusted business data. It has no money, send, approval, contract, database, or credential tools. Any proposed action still has to pass structured validation, authorization, policy, and audit controls outside the AI system.

We do not use Customer Data to train AI models, and we do not authorize model providers to use it for their own model training. We send providers only the information reasonably needed to perform the requested task and apply contractual and technical controls to that processing.

Evaluation and test datasets contain only synthetic or fully anonymized records. We may also use aggregated service information that cannot reasonably identify a customer or person to test extraction, safety, and reliability. Our privacy policy explains how information is collected, disclosed, retained, and deleted.

We use structured schemas, source links, field-level checks, confidence thresholds, policy evaluation, permission checks, idempotency controls, audit records, and deterministic tests around AI-assisted workflows. We also test for prompt injection, unauthorized actions, tenant isolation, and false confidence on fields that affect money.

No set of controls removes every error. We monitor system behavior, investigate reported failures, and adjust tests and safeguards as failure patterns become clear.

Dialed is designed to show the source, proposed action, relevant confidence or exception, required approval, and resulting audit record where the workflow supports it. Communications identify the roaster; where supported and configured, they may also indicate that Dialed assisted with delivery.

We may update this policy as our AI systems, safeguards, or legal obligations change. We will post the revised policy with a new effective date and provide additional notice before a material change when required. A change to this policy does not give AI new authority over your account; authority remains governed by product controls, permissions, and your agreement with Dialed.

If Dialed produces an incorrect result or behaves unexpectedly, contact support@withdialed.com. Include the workflow or record identifier, what you expected, and what happened. Do not send passwords, access tokens, or full payment-card details.