What does AI
implementation cost?
An AI subscription is only one part of an implementation budget. The work needed to connect systems, prepare information, test outputs and support the team can matter more than the model bill. This guide explains cost drivers; it is not a price list or quotation.
What changes the initial scope?
- Workflow boundaries: one well-defined task is different from replacing an end-to-end business process.
- Data preparation: inconsistent records, missing fields and access restrictions need attention before reliable testing.
- Integrations: documented APIs may simplify a connection; unsupported systems and manual exports can add work.
- Quality requirements: the consequences of an incorrect answer determine the review and testing effort.
- Rollout: documentation, staff training, migration and a fallback process are part of delivery.
Separate one-off and ongoing costs
A proposal should distinguish discovery, design, implementation and initial testing from recurring expenses. Ask which licences you need, who pays for usage, what maintenance includes and what happens when an upstream system changes.
- Model or API usage, including retries and longer documents.
- Software subscriptions, hosting and storage.
- Human review and handling exceptions.
- Monitoring, updates, support and periodic quality checks.
- Exit costs: exporting information, replacing a supplier or retiring the workflow.
Who maintains the software after launch?
Agree this before the build starts. Your internal team, Astir under an agreed support scope, or another provider can take responsibility. The proposal should name who handles updates, access changes, failed jobs and questions from staff.
Budget for checking that integrations still work when a supplier changes its system. AI workflows also need reviews of output quality and usage costs. Agree support hours, response arrangements and which changes need a separate quote.
Ask for operating instructions, an account and dependency register, and a tested recovery process appropriate to the work. See what a workflow handover should cover.
Model a range, then test the assumptions
Estimate a lower-volume and higher-volume operating month. Include the time needed to review the output. The purpose is to expose what drives the decision rather than present a precise forecast built on uncertain inputs.
For cash-return analysis, compare verified financial benefits with both setup and ongoing costs over the same period. Keep capacity released separate from cash savings unless the business has a clear way to realise that saving.
Ask these questions before approving a pilot
- What will be delivered, and what is outside scope?
- How will success and unacceptable errors be measured?
- Which data, accounts and access permissions are required?
- Who owns the output and has responsibility for ongoing operation?
- What would trigger additional work or a revised quote?
- How can we pause, roll back or stop using the solution?
A real software cost example
Brulee's grocery-shop example reports $5,000+ annual transaction and subscription fee savings against a $1,000 initial cost. It illustrates the value of examining operating costs. It is a retail software project, not an AI implementation benchmark or a typical expected return.
Read the Brulee case and calculation
Comparing online-store costs?
Use our Shopify to WooCommerce migration calculator to compare setup, payment fees and upkeep across the same period.
Scope before committing
Astir starts with the problem, systems and intended result so the work can be discussed concretely. We do not publish a universal AI package price because the appropriate scope depends on the workflow and its constraints.
Discuss scope and value ↗