DIRECT ANSWER
A proposal is not the price of an AI tool. It covers a tested, owned and transferable workflow.
A normal first phase covers workflow diagnosis, scope, data mapping, implementation, exception handling, testing, documentation and handover. Price changes with system count, data quality, permissions, human approval, failure risk and service requirements.
Third-party software, API usage, WhatsApp messages, hosting and ongoing maintenance should normally be listed separately. Compare inclusions and exclusions before comparing totals.
A complete first phase normally includes six types of work.
Workflow diagnosis and success criteria
Define triggers, inputs, outputs, monthly volume, normal path, exceptions, owner and measurable before-and-after outcomes.
Scope and system mapping
State inclusions, exclusions, the system of record and which fields can be read or written.
Build and integration
Connect forms, CRM, WhatsApp, sheets, databases, accounting or other existing platforms using approved rules.
Exceptions and human approval
Define what happens for missing data, low confidence, duplicates, mismatches, special terms and system failures.
Testing and acceptance
Use safe representative samples to test normal, failure and edge cases, then accept against expected results.
Documentation, handover and launch
Document accounts, permissions, rules, monitoring, manual fallback, common failures and change procedures so knowledge is transferable.
Seven cost types should not be hidden inside one vague total.
Third-party software
CRM, automation platforms, document tools, databases, email and other SaaS should list account ownership and payer.
API, model and messaging usage
Usage-based document, token, message, storage and execution fees should be estimated from real volume.
Hosting, monitoring and logs
Production needs runtime, error records, alerts and retention, not one successful run on a development machine.
Data cleanup and legacy systems
Duplicates, missing fields, legacy formats, absent APIs and unclear permissions add preparation and testing.
Training and internal time
Rules, samples, testing and decisions from process owners, users, IT, finance or management are part of project cost.
Ongoing maintenance
Platform changes, credentials, failures, model evaluation and monitoring need a defined response scope, not a vague support promise.
Material changes
New platforms, changed business rules, expanded data or redesigned workflows are normally change requests rather than fixes to the agreed scope.
Similar projects can differ greatly because of risk and exceptions.
The clearest cost drivers are system count, data consistency, stable read and write access, human approval, and whether failures require retries, alerts, reconciliation and recovery.
Internal drafts allow correction. Formal quotations, payments, inventory, customer messages and accounting records require stronger testing, permissions, logs and approval.
Understand scope before asking what a project should cost.
One form to CRM
Fixed fields, official APIs, deduplication, routing and notification are normally simpler than cross-system document workflows. Platforms and exceptions still need confirmation.
Invoice extraction and inventory matching
Supplier formats, inconsistent names, low-confidence data, duplicates, inventory records and formal posting add mapping, testing and review.
Website, WhatsApp, CRM and sales handoff
Beyond integration, define consent, message timing, qualification, routing, stop conditions, human takeover and conversion tracking.
Build cost covers initial delivery. Maintenance covers later change and failure.
The build normally covers diagnosis, mapping, implementation, testing and handover within the agreed scope. Maintenance may cover health checks, alerts, credentials, dependencies, minor fixes and periodic model evaluation. New business rules, platforms or workflows should be separate change requests.
Align ten items before comparing proposals.
Goal, scope and exclusions
Does each proposal solve the same workflow and state what is excluded?
Systems, data and ownership
Who owns accounts, source, credentials, data, documentation and exit arrangements?
Exceptions, approval and failure handling
Beyond the normal demo, how are duplicates, missing data, timeouts, low confidence and platform failure handled?
Testing, acceptance and launch
What samples, expected results, pass criteria, pilot, rollback and named owner are used?
Total first-year cost
Separate setup, platforms, APIs, messaging, hosting, maintenance, training, changes and internal time.
Four situations do not justify a custom project yet.
Volume is too low
For a few monthly runs taking minutes, an existing tool, template or simple configuration is usually more sensible.
The process changes daily
If the team has no agreed process, stabilise the SOP, responsibility and exceptions before automating confusion.
Existing software solves most of it
If mature SaaS solves most of the need, use it first and integrate only the business-specific parts.
There is no process owner
Without someone to provide rules, test, approve changes and handle exceptions, a provider cannot own daily operations for the business.
Use published cases to understand value without treating hours as guaranteed cash savings.
An AutoBrand manufacturing case processes roughly 3,000 invoices monthly and reduced average handling from around ten minutes to one. That releases around 450 handling hours per month, but financial value still depends on how capacity is used, error cost and adoption.
Invoices per month
Average minutes per invoice
Monthly handling capacity released
Not guaranteed cash savings
Frequently asked questions
Does AI automation have a fixed package price?
Usually not. A custom project needs a defined workflow, systems, data, exceptions, testing, permissions and risk scope before a formal proposal can be prepared.
Should a proposal include third-party monthly fees?
They should be disclosed but may not be included in the build fee. List platforms, APIs, messaging, hosting and usage separately with account ownership.
How long does a typical project take?
A clearly scoped first phase typically takes around two to four weeks. Legacy systems, cleanup, multiple approvals, many exceptions and complex permissions extend it.
What is the difference between maintenance and a change request?
Maintenance covers monitoring, failures, credentials, dependencies and minor fixes in the agreed workflow. New platforms, rules, data scope or features are normally change requests.
Should every business commission a custom project?
No. Low volume, unstable processes, no owner or a strong existing product normally favour a simpler option first.
EXPERT AUTOMATION CONSULTATION
Want to confirm where your business should start?
Diagnose one real workflow first, clarifying value, risk, data, systems and the minimum viable scope before deciding whether to build.