The work that comes after choosing an AI model

Conversations about AI often begin with a demonstration: a document is summarised, a question gets an answer, or an agent completes several steps. That is a useful way to explore possibilities. A business owner also needs to understand what changes when the demonstration becomes part of everyday work.

Who can use it? Where does the information come from? What happens when it makes a mistake? Where is the decision recorded? These questions define much of the development that follows.

What the Faculty acquisition tells us

Accenture announced the completion of its Faculty acquisition on 16 March 2026. The transaction brings AI design and implementation capabilities alongside Frontier, Faculty’s decision intelligence product. Financial terms were not disclosed. Source: Accenture.

The earlier acquisition of Halfspace in March 2025 also involved specialists in AI solutions for business processes. Source: Accenture.

Our reading of these cases is specific: products and implementation work can complement one another. A reusable platform provides a foundation; understanding each organisation determines what must be connected, adapted and tested. These transactions cannot establish the return on an unrelated project or predict the valuation of a small company.

The model responds. The application organises the work

Consider a services company receiving requests with supporting documents. This is an illustrative example, not a client case study.

A demonstration might extract the applicant’s name and summarise the request. A useful application also needs to attach the document to the right record, identify missing fields, display the original and let someone correct the suggestion before saving it to the business system.

There are three distinct stages:

• Intake. Documents arrive through a defined channel, with access restricted to the appropriate people.
• Review. The team sees the AI suggestion and can consult the supporting evidence before accepting it.
• Record. The validated result enters the working system and retains a history of changes.

AI has a specific role within that process. The project should be able to explain the other steps in equally concrete terms.

What to ask before commissioning a project

Start with a recurring task and an identifiable group of users. Agree on the examples used for testing, the errors that require review and how the result will be compared with the current process.

The proposal should separate development from model, hosting and maintenance costs. It should also identify who administers access and how information can be recovered if the provider changes. These answers make it easier to compare proposals that initially look similar.

At Kenea, we scope an initial development around that complete process: a defined need, the necessary connections and a delivery the team can review. Where conventional automation is better suited to part of the work, it belongs in the solution too.