At UBI Business School, our work on AI started with a practical question: how could we help colleagues spend less time finding information and preparing routine work, and more time acting on it?
The result was DeeJay, an AI assistant available through Microsoft Teams and connected to operational information in Odoo. We often describe DeeJay as an AI colleague. That description is useful because it brings the conversation back to work: what the assistant is responsible for, what information it needs, and when a person should take over.
Building DeeJay has reinforced a conviction behind IJK’s AI implementation work. The value comes from fitting the technology into the organisation’s daily routines, with a clear job to do.
Start with questions people already ask
Consider a familiar admissions question: has a particular piece of correspondence been sent, and has the applicant replied? Answering it can involve opening a record, finding the relevant conversation and checking the sequence of events.
We built a correspondence tool that brings together the relevant Odoo conversation history and distinguishes outgoing messages, incoming replies and internal notes. The assistant can use that timeline to help a colleague understand what happened and what may need attention.
This is a modest task in isolation. Its usefulness comes from how often questions like it arise. Making the answer easier to obtain reduces the preparation needed before a person can make the next decision.
Make the output part of the workflow
A useful answer is only one form of output. For recurring operational work, colleagues also need something they can review, filter and act on.
In the finance workflow, our work included preparing instalment follow-up worklists and staging covering or reminder emails for staff review. The worklist separates different types of action, so the person responsible can examine the underlying case rather than starting by assembling the list.
Drafts belong where the colleague already works. A prepared email in an authorised staff member’s Outlook Drafts is easier to review and use than text that has to be copied out of a separate chat. That small design choice matters to adoption.
Give business rules the same attention as the model
The work also taught us how much context sits behind an apparently simple request. A missing invoice reference, for example, does not by itself establish that an amount is owed. Different arrangements and the underlying records have to be considered.
Similarly, a management report needs agreed definitions of its population, categories and dates. We developed repeatable reporting logic for intake and receivables work so the calculations could be produced consistently, with the assistant helping people ask questions and interpret the result.
Domain knowledge became part of the implementation. The model needed tools and rules that reflected the way UBI actually operated.
Build a colleague people can work with
The practical productivity benefit is less effort spent gathering context, assembling a first draft or rebuilding a familiar report. People can start closer to the point where their judgment is needed.
We evaluate that benefit through specific workflows: how long preparation takes, how much correction is required, whether the output gets used and whether the next action happens sooner.
Our lesson from UBI is to give an AI colleague a useful role, connect it to the right information and improve it through real feedback. That is also how IJK’s AI implementation consulting arm approaches work with other companies.