When comparing business automation services, I would check which work the supplier takes on and which work stays with your team. One quote might cover connections; another might include exception handling, documentation and handover. If both promise “complete automation”, unpack that scope before comparing the totals.
This guide is for a business that has chosen a process and needs to turn it into a verifiable engagement. If you are still choosing where to start, use the automation prioritisation worksheet first. Here, the unit of comparison is each task and its operational owner, rather than a count of connections or screens.
Turn the chosen process into a task inventory
I would map the current journey from incoming work to the point where someone considers it complete. For every step, record its trigger, required information, destination system and what stays unresolved when something is missing. Include the less visible work: checking an import finished, correcting a field or chasing an approval.
- Input: source, required fields, identifier and the person responsible for data quality.
- Outcome: expected record or document, destination and a way to confirm it exists.
- Rule: agreed decision and the person authorised to change it.
- Exception: why a case stops, who receives it and how they return it to the workflow.
- Operations: frequency, ordinary load, known peaks and acceptable delay, with measured information or explicit open questions.
The inventory should separate building, operating and accepting the work. The implementer can demonstrate the flow; the business confirms the outcome is useful; a designated person handles unresolved cases. I would not assume all three responsibilities are included in the same service.
Compare configuration, integration and custom development
Ask the supplier to identify the smallest option that covers the inventory. If a feature in your current tool handles the journey and its exceptions, the engagement might only need configuration, testing and training. Moving data between systems may call for an integration. Custom rules, approvals or recovery requirements that existing options cannot support may justify development.
I would not buy another platform simply because it advertises more connectors. Ask for a demonstration of your workflow using the intended accounts and plans, including permissions and limits. The proposal should distinguish verified behaviour from assumptions still dependent on access, documentation or a bounded test.
Text interpretation and autonomy are separate decisions. The AI agents versus workflows comparison helps scope them. I would not add AI to a copying task with known rules. Nor would I commission implementation while the team has contradictory rules or has not agreed who can approve exceptions.
If a proposal adds AI to interpret documents or classify requests, give that evaluation its own budget and acceptance criteria. The AI consulting cost guide separates assessment, pilot and implementation. Define what a person does when an output is unacceptable and how much review remains; automating data transfer does not demonstrate correct classification.
A matrix for comparing and accepting the service
Copy these rows into your own process brief. Ask each supplier to mark items as included, optional, excluded or unresolved; identify the deliverable; and separate initial and recurring costs. Replace the roles with named owners before approval. This is a buying matrix with proposed checks, not a report of completed supplier testing.
| Task | Required deliverable | Ownership to agree | Acceptance check |
|---|---|---|---|
| Receive work | Defined input and a record of received, rejected and pending cases. | The business validates the fields; the supplier implements intake. | Valid input starts the journey; incomplete input has an identifiable reason and review destination. |
| Apply rules | Versioned rules, approvals and excluded actions. | The business approves rules and changes; the supplier implements them. | Agreed cases follow the expected branch; an unauthorised person cannot approve. |
| Deliver the result | A link between source and destination, with confirmation of the operation. | The destination system owner validates fields and states. | The final record is checked; repeating the same case does not cause another business action. |
| Handle exceptions | A pending-work list, reasons and a correction procedure. | Operations resolves data issues; technical support handles failures within agreed coverage. | An operator finds, corrects and resumes a case; connection failures remain visible. |
| Pause and recover | Instructions for stopping, manual fallback and reconciling pending work. | A designated person authorises the pause and return to operation. | A stop is rehearsed; resuming distinguishes completed work from pending cases without repeating completed actions. |
| Maintain and transfer | Agreed configuration or source code, account inventory, checks and instructions. | The business controls accounts; the maintainer receives limited access and explicit responsibilities. | Another authorised person can follow the instructions, investigate a failure and run the agreed checks. |
A dashboard does not resolve a row with no owner. An unresolved row should not be counted as included at no cost either. Before hiring, decide whether to investigate that uncertainty in a separate phase or exclude it from the first scope.
Example: two proposals for preparing internal orders
Hypothetical example: a business receives internal requests in a shared spreadsheet. Operations checks the cost centre and owner, then prepares an order in another tool. The initial scope ends with an internally approved draft; external purchases, payments and historical imports are excluded.
Proposal A copies complete rows and supplies instructions for running the flow. Staff review rejected rows manually. Proposal B adds an exception list, assignment to operations and documented recovery. Either could fit, but they buy different operating arrangements. A may be enough if the team accepts and can manage manual review. B needs to justify its cost through the work it covers, without promising savings before measuring them.
For acceptance, I would prepare a valid request, one without a cost centre, a repeated request and a case whose destination is unavailable. The business defines the expected outcome. The supplier supplies evidence for each journey. Operations demonstrates finding and correcting the incomplete request; the destination owner confirms that recovery has not created a second order.
If the destination cannot establish whether an order was saved, I would ask for that uncertainty to be resolved before authorising automatic retries. If someone changes a request after approval, the proposal needs to specify whether approval is required again. These boundaries distinguish an accepted workflow from a demonstration with one convenient example.
For HR workflows, an approval also needs to update balances, calendars and team coverage. The leave management software matrix defines checks for delegation, concurrent requests and cancellations before accepting that workflow.
Put scope, exclusions and ownership in writing
The proposal should enumerate tasks, systems, fields, data directions, users and environments. Add acceptance samples, who prepares them, who approves delivery and how a failed check is recorded. New requirements need an explicit scope and cost decision before implementation.
I would request the process map, rule inventory, agreed configuration or code, test evidence, incident procedure and a handover session. Specify which accounts and subscriptions the business controls and which materials it can pass to another maintainer. A source-code copy is insufficient if nobody has the access or instructions required to run it.
Separate historical cleanup, migration, additional departments, new channels, rule changes and support outside agreed hours. These need not always be excluded, but they must be priced when required. If a previous workflow is being replaced, request a pending-work inventory and a cutover rule so the transition does not leave cases without an owner.
For content distribution, also compare who approves each version, confirms destinations and handles partial publication. The content automation pricing and scope matrix defines those responsibilities and their operating cost.
Separate implementation, maintenance and staff effort
I would compare implementation by process analysis, configuration or development, data preparation, agreed migration, testing, launch and training. Effort changes with rules, input quality, systems, permissions, exceptions and recovery needs. I would not attach a price to “an automation” without defining those items.
For operations, ask for licences, hosting, external usage charges, monitoring, maintenance and internal review time. Each item should state who pays, its usage assumptions and what happens when the contracted allowance is exceeded. I do not give product prices or rates for my services without a specific proposal. Unconfirmed costs remain open questions.
For maintenance, distinguish fixing a failure to meet an agreed requirement, adapting to a changed tool and adding a new rule. Agree who receives alerts, support hours, priorities, response time and what needs a separate quote. Responding to an incident does not guarantee resolution within the same period.
Compare the same period and include exit costs: exporting configuration and data, transferring access and training another owner. For engagement models, milestones and scope changes, see the general software quote guide. For this purchase, also compare how many tasks remain with your team under each offer.
What I need to assess your automation project
Send the chosen process, current tools, anonymised input and output examples, common exceptions and the person who would review the result. Add known workload, possible system access and constraints on timing, budget and maintenance. If you have proposals, share the tasks included in each without credentials or personal data.
I can review a first automation and applied AI delivery with you: what to configure, integrate or build, and what should remain with a person. You can share the process context so we can define that engagement.