AI automation services: scope, costs and what to expect

The short version

Buy AI automation services for a defined workflow, not a promise to add AI everywhere. Agree what starts the process, what it produces, who checks it and what happens when it fails. A useful quote covers setup, running costs, ownership, acceptance tests and support, not just a demonstration.

The service may involve connecting tools, applying fixed rules or using AI to classify text and prepare drafts. AI should earn its place. If a rule handles the task reliably, you do not need a model making that decision.

If the job is not yet clear, use our guide to choosing a first task before building an agent. This guide is about commissioning that task once you can describe it.

What should the scope include?

Describe one workflow from input to reviewed output. Name the systems involved, the data needed and the actions allowed. Include awkward cases, such as missing information, duplicates and unavailable services. State what stays manual.

For example, an illustrative enquiry workflow might suggest a category and draft an internal summary, leaving a person to choose the reply. Automatically sending that reply is a separate scope decision with different risks. Agree it explicitly rather than letting a prototype quietly become an unattended service.

Ask the supplier to write down exclusions too. Historical data cleanup, new software licences and changes to your existing tools can materially change the work. A small pilot should test the uncertain part before you commission a wider rollout.

Choose the buying route by what is uncertain

Route	Useful when	Check before buying
Configure an existing tool	The steps are understood and supported by your current software.	Who owns the configuration and handles broken connections?
Scoping consultation	You need to decide the workflow, risks or integration approach.	What written decisions or brief will you receive?
Pilot or custom build	You need working evidence or behaviour existing tools do not provide.	What testable output is included, and what would expansion cost?

What will it cost to own?

Separate the initial build from the ongoing bill. Ask about hosting, tool subscriptions, model usage, storage, monitoring and support. Check whether third-party charges are included, passed through or paid directly by you. Agree usage limits and who receives a warning before spending rises.

Our public pricing and starting ranges are in USD, not fixed menu prices. Final scope and pricing are agreed in writing. A paid AI consultation can help resolve uncertainty before a build, but it is not required to send a project enquiry.

Compare quotes against the same workflow and support assumptions. A cheaper setup that leaves your team diagnosing every failed run may cost more to operate. Ask how changes are priced, whether taxes apply and what happens if an external provider changes its API or fees. Use your own workload to assess value, not a supplier's generic savings claim.

Who owns the system and the data?

Before granting access, agree who owns accounts, configuration, source code and documentation. Clarify any third-party licences and what you can export if you change supplier. Your team should know how to revoke access and continue, replace or stop the workflow.

For AI processing, ask which provider receives the data, what is retained and which contractual settings apply. Use redacted examples during scoping. Grant only the access the workflow needs; do not send passwords or customer records through a public enquiry form.

How should you accept the work?

Agree acceptance cases before implementation. Include normal inputs, missing fields, duplicates and failures. For AI outputs, define acceptable content and errors that require rejection. Keep approval for actions affecting money, access, customer commitments or records you cannot easily restore.

Real build evidence: Transmission

Our Transmission intake system field report describes an existing build with server-side validation, a honeypot, separate recipient routing for project and Test Pilot enquiries, and visitor-safe status messages. These are ordinary validation and routing controls, not AI decisions.

What happens after handover?

Name an operational owner and agree who investigates failures, renews connections and approves changes. Distinguish defect fixes from ongoing support and new features. Read the automation maintenance checklist before signing so monitoring, recovery and human escalation are part of the agreement.

See our automation and workflow build service, then send a project brief with the task, tools and desired output. Bring redacted examples and your acceptance criteria. That gives both sides a concrete basis for a useful proposal.

Related capability

Automations

Workflow automations that save time across handoffs, repeated admin, and reporting.