Key takeaways
- Price is set by five drivers: systems touched, record quality, exception paths, how much human approval is required, and who runs the system after launch.
- Software licences are usually the smaller line. Most of the spend goes into mapping, integration, testing and ownership after launch.
- A useful quote separates mapping, integration, context and rules, controls and approvals, testing, tools and model usage, and operation after launch.
- A fixed quote issued before anyone has traced the records is a price on uncertainty, and it grows through change orders once the build finds the real exceptions.
- Credentials expire, APIs change and new exception types appear, so the running cost belongs in the budget from the start.
What are the five cost drivers?
Two companies can ask for the same thing, such as an onboarding system, and receive quotes that differ several times over. The difference is rarely the AI model. It is the operating work around it.
- Systems touched: each tool the workflow reads from or writes to needs authentication, field mapping, error handling and a readback test.
- Record quality: if the data the workflow depends on is incomplete or inconsistent, part of the budget goes to cleaning or deciding which source wins.
- Exception paths: the happy path is cheap. Every "unless" in the process (missing documents, changed scope, a client on a special contract) adds design and testing.
- Human approval: deciding which actions need a person, building the review step and logging decisions takes real design time.
- Operation after launch: someone has to watch errors, renew credentials, adapt to tool changes and handle new exception types.
What do you pay for, line by line?
A useful quote separates the lines below. If a proposal shows one number with no breakdown, ask for one.
| Line | What it covers | What moves the number |
|---|---|---|
| Mapping and scoping | Tracing the real workflow, records, owners, delays and exceptions, and writing the acceptance test | Number of teams involved and how documented the process is |
| Integration | Connecting and testing each system the workflow reads or writes | Number of systems, API quality, permission model |
| Context and rules | The approved documents, examples, terminology and rules the AI steps use | How much company knowledge is written down and current |
| Controls and approvals | Review steps, permissions, limits, audit log, failure handling | How many actions reach clients or staff |
| Testing and acceptance | Running real cases against the written acceptance test | Number of cases and exception paths |
| Tools and model usage | Plan upgrades, automation platform fees, hosting, model API calls | Volume, current plan tiers, hosting choice |
| Operation after launch | Monitoring, fixes, credential renewals, changes, reporting | Whether your team or the partner operates it |
How much of the cost is tool licences?
Check what your current subscriptions already allow before anyone scopes new software. Plan tiers are a common hidden line.
- HubSpot: creating workflows requires a Professional or Enterprise subscription on the relevant hub, according to HubSpot's knowledge base. A company on a Starter plan may need an upgrade before any CRM automation is possible.
- n8n: without a licence key, self-hosted n8n runs as the free Community edition, according to n8n's documentation. The software is free. Hosting, updates and maintenance are not.
- Zapier: pricing is set by the number of tasks run each month, so cost grows with volume.
- Model APIs: language model usage is billed by volume. For a single operational workflow it is usually small next to integration and operation, but estimate it from real volumes.
Why is a fixed quote before mapping a warning sign?
A quote issued before anyone has traced the records is a price on uncertainty. It is either padded to cover what nobody has looked at yet, or it will grow through change orders once the build finds the real exceptions. A paid mapping step that ends in a written scope and acceptance test is usually cheaper than either outcome.
Be careful with published benchmarks too. Gartner reported in July 2024 that generative AI deployment approaches for business model innovation cost between 5 and 20 million US dollars. Those figures describe large enterprise programmes. They are not a guide to installing one workflow in a mid-size services firm, and neither is a no-code template price.
How do you keep the first project bounded?
Cost overruns in AI projects usually come from scope that was never written down. Five constraints keep a first install contained.
- One workflow with one owner.
- A named list of systems the workflow may touch, and no others.
- Review mode first: the system proposes, a person approves, until the acceptance test passes.
- An acceptance test agreed before the build starts.
- A handover plan that names who operates the system after launch.
What should you budget for year two?
Credentials expire, APIs change, models are updated and new exception types appear. A system nobody owns breaks without anyone noticing. Put the running cost in the budget from the start, whoever operates it, so the system still works when you consider the next workflow.
Frequently asked questions
Is it cheaper to build in-house?
It depends on whether AI is becoming a permanent capability for your company. For one bounded operational workflow, hiring a team is usually slower and more expensive than a scoped install. The guide on partner, in-house or software covers the trade-offs.
Does Outproduce publish prices?
Not yet. Outproduce scopes through a paid Operating Map before quoting a fixed-scope install, because the number depends on the drivers described above.
What is the cheapest way to start?
Use what you already pay for. Many CRM, project and help desk tools include automation features on higher plans. If the workflow still needs connections across several tools, AI steps or human approvals, that is when a scoped install makes sense.
Who owns the system after launch?
Someone has to watch errors, renew credentials, adapt to tool changes and handle new exception types. The handover plan names who operates the system, and managed operation and documented handover are both valid outcomes.
What happens when the system gets something wrong?
When a consequential action is uncertain the system creates a visible exception instead of guessing. A queue that waits is cheaper than an automated mistake nobody sees until a customer does.
What if we want to hand the system to our own team?
Ownership, credentials, code, data, documentation, hosting and support responsibility are written into the engagement, so the system keeps running in accounts you own if the relationship ends.
What should we budget for year two?
Credentials expire, APIs change, models are updated and new exception types appear. Put the running cost in the budget from the start, whoever operates it.
Sources
This article describes operating principles and public sources. It is not a client result. Build Records carry the basis for every Outproduce system claim.