The cost of a custom AI implementation depends on the work it must complete, the systems and records it can use, the review steps, and the support it needs after launch. A useful quote separates discovery, implementation, software usage and ongoing support. Horizon scopes these for your business before quoting. The example below explains the calculation using invented inputs; it is not a market average or a Horizon price offer.
What should an AI implementation quote include?
Ask for a defined outcome, the systems included, the actions the workflow may take, the review point and a clear acceptance check. Two proposals described as an AI assistant can cover very different work.
A document summary for staff review needs a different scope from a system that also retrieves client records, creates tasks and prepares changes in a finance platform. The quote should name those differences.
- Discovery: process mapping, access checks, sample review and a written scope.
- Implementation: connections, workflow logic, interface and review controls.
- Validation: representative examples, failure handling, permission checks and staff acceptance.
- Handover: training, documentation, ownership and the support contact.
- Ongoing costs: software plans, AI usage, hosting, monitoring and maintenance.
For the delivery stages, see AI implementation services. For the connection work, see AI integration services.
A worked example: how a budget can be calculated
Suppose a team wants a draft client briefing using approved CRM records and email. The following invented effort and rate show how to make a proposal transparent. They are illustrative assumptions, not observed project costs, a quote or an Australian pricing benchmark.
- Scope and access review: 8 hours.
- Connections and workflow: 24 hours.
- Testing and staff review: 12 hours.
- Training and handover: 6 hours.
The assumed effort is 8 + 24 + 12 + 6 = 50 hours. At an illustrative rate of A$180 per hour, the calculation is 50 × A$180 = A$9,000, excluding GST. An agreed contingency of 10 hours at the same assumed rate would add A$1,800, giving an illustrative total of A$10,800 excluding GST. These inputs are chosen only to demonstrate the arithmetic.
Software subscriptions, usage and ongoing support are excluded from this example and would be listed separately. Replace every input with the supplier's actual estimate. Ask which assumptions would trigger a scope change, and whether contingency is a fixed charge, an allowance or only billed if used.
This guide previously presented cost multipliers as Horizon cost data. Those figures did not have a documented project sample and have been removed. The worked example above is a planning illustration with visible assumptions.
What makes one workflow cost more than another?
Useful scoping questions concern access, data quality, actions and operating conditions. There is no reliable fixed multiplier for adding a second system or a particular industry.
- Available access: can the software expose the specific records and actions needed, and does the client have permission to use them?
- Data quality: are documents readable, records consistent and identifiers reliable?
- Review: who approves an output, how are exceptions handled and what must be recorded?
- Volume: how many cases, pages and requests will run, including peak periods?
- Operating needs: what happens when a vendor changes, a connection fails or the process changes?
A good proposal explains these assumptions in business terms. It also identifies the native product features or existing integrations that can do part of the work before adding custom software.
What is free, and when does a paid assessment begin?
Horizon offers a free 30-minute introductory call to discuss the workflow and whether we can help. A paid assessment is a separate, agreed engagement with written deliverables. An implementation is separately scoped and quoted.
Use the introductory call to explain the work, the software involved and the result you want. It may identify a useful next step, but it is not a completed technical assessment or a guaranteed fixed quote.
Where deeper investigation is needed, agree the assessment's fee and deliverables before it starts. These can include a process map, access findings, candidate workflows and an implementation scope. Any credit toward a later build must be stated in the proposal; it is not automatic.
See the workflow assessment for that stage, or book an introductory call.
How should you estimate ongoing AI costs?
Use the expected workload and current supplier prices. Separate AI consumption from subscriptions, hosting and human support so the recurring bill can be checked against actual usage.
For a usage-priced model, estimate input and output volume per case, multiply by expected monthly cases and the provider's current unit prices, then include retries and unusually large cases. Test with representative examples before treating the estimate as a budget.
Ask who owns each subscription, what limits or alerts apply, how logs are retained and who responds to a failed workflow. A support arrangement should specify response expectations and included changes. There is no universal monthly amount or fixed percentage of build cost that fits every workflow.
How do you compare two proposals fairly?
Compare the same outcome and operating responsibilities. The lower headline price may cover fewer records, less testing or less support; a larger price should still explain the extra work.
Ask each supplier to demonstrate the intended result using an approved example, name the review and acceptance steps, and show which costs recur. Check the handover, ownership and exit terms. Compare a native feature, configuration work and a custom build where each is a plausible solution.
Measure the current process before estimating value: cases per week, time per case, rework and delay. After a pilot, compare the same measures and include the time people spend reviewing outputs. Treat estimated time savings as a hypothesis until the workflow has been used and measured.