The important question is not which AI product currently has the highest profile. Institutions need to establish whether they require a general productivity aid, an integrated specialist function or a digital service of their own. Data, process ownership, interfaces and ongoing operations determine the answer.
Decision Table
| Criterion | Standard AI Is Enough | A Custom System Makes Sense |
|---|---|---|
| Task | General drafting, summarising and structuring ideas | A binding professional workflow with status, rules and exceptions |
| Data | Approved, low-sensitivity content | Personal or professionally sensitive information |
| Integration | Manual transfer is acceptable | Specialist applications, registers or document stores must connect |
| Accountability | A person fully reviews the output | Processing steps and approvals must be recorded |
| Permissions | Basic user management is sufficient | Granular roles and institutional responsibilities |
| Operations | Vendor support is sufficient | Monitoring, maintenance and professional development form part of the service |
| Adaptation | The supplied feature set fits | Workflow and interface need to reflect the institution |
| Exit | The product’s export options are sufficient | Data formats, interfaces and handover must be secured contractually |
Three Typical Scenarios
Scenario A: A General Productivity Aid
Staff want to summarise approved text, organise a draft or improve wording. The institution can define acceptable use, permitted data and mandatory review. Assessment: a contractually and legally reviewed standard tool may be sufficient.
Scenario B: An Assistant Using Institutional Knowledge
Staff or members of the public need answers based on approved policies and information. Sources, currency, access rights and escalation for uncertain cases must be controlled. Assessment: a custom system built on existing models is usually more appropriate than an unrestricted chat tool.
Scenario C: An End-to-End Professional Workflow
Applications, documentation, approvals and reporting need to work together. Beyond any AI support, the service requires forms, roles, interfaces, logging and operations. Assessment: this calls for a custom specialist solution or an adaptable case-management product; the language model is only one possible component.
Lifecycle Cost Rather than Licence Price
Two cost categories are often overlooked:
- Introduction and operations. Data protection review, configuration, training, support, quality assurance and further development apply to both standard and custom systems.
- Integration and exit. Missing exports or proprietary interfaces can make later changes difficult. Custom development is only more portable when documentation and handover rights are agreed properly.
“Standard AI is a tool. An institutional system is an accountable service with a professional workflow, data model, permissions and operations.”
What We Recommend
- Use standard tools for bounded, low-risk tasks where the contract and data processing arrangements are appropriate.
- For professional workflows, define roles, data, exceptions, interfaces and evidence first.
- Build a custom system only where an existing product cannot meet the institutional requirements reliably.
Which Type of System Fits Your Project?
We can assess the task, data, integration and operating model in an initial, no-obligation conversation.