Preparing the information
The support manager opens the knowledge folder and finds three versions of the same policy. A current exception appears only in an email. Instructions assume employees understand shorthand a new system will not necessarily interpret correctly.
Preparing that material takes time from people who already have jobs. They may need to clarify policy, remove outdated content and agree who approves changes. If their time is absent from the estimate, the cost has been transferred rather than removed.
Ask what information must be ready before development begins, who prepares it and who decides that it is good enough. A dependency without an owner can become a delay nobody budgeted for.
Testing difficult cases
A test question with an obvious answer is useful for showing the idea. Everyday support includes incomplete requests, disputed information and situations where a wrong answer has consequences.
The team needs representative examples and a way to judge responses. Someone must define when the assistant can answer, when it should ask a question and when it should involve an employee. Customer-facing changes need review before release.
Budget for that evaluation and the work it uncovers. The same applies to connections with customer records or ticketing systems. Access, permissions, failure handling and changes to the underlying software all affect what it takes to keep the process dependable.
Life after launch
On launch day, the assistant does not become its own manager. Customers may phrase questions differently from the test set. A policy can change. A useful answer can still leave someone unable to complete the next step.
Name the person responsible for those signals. Give the team time to review performance, correct information and investigate issues. Include ongoing usage costs and the effort required to handle escalations.
Training belongs in the budget too. Employees need to know when to trust the assistant, what to check and how to report a problem. If checking its work takes longer than doing the task directly, the supposed saving needs to be reconsidered.
The first year
Return to the approval meeting with a fuller picture: preparation, implementation, integration, testing, training and ongoing operation. Separate one-time work from recurring costs and state what could cause either to change.
Compare a few realistic usage scenarios. How many requests might the assistant receive? What proportion may need review? What happens if it expands to another team or needs more information? Use explicit assumptions rather than a single optimistic forecast.
The largest cost will vary. A narrow assistant using clean material may be inexpensive to operate. A process involving fragmented data and consequential actions can demand more preparation and oversight. A credible budget explains the difference.
The supplier’s scope
Before signing, ask the supplier to walk through a policy change, a failed connection and an incorrect response. Who does what? Which work is included? What requires a new scope? What can your team manage without specialist help?
Those questions help you compare proposals on the same basis. They also expose whether the business is ready to own the result.
At Superfactor, we can help review the scope behind an AI estimate and identify the work needed to put the capability into use. Bring the quote and the workflow it is meant to improve. The important number is the cost of making that improvement work in your business.
