First describe what someone needs.
An employee opens several screens to answer one question. Data is exported, combined and checked again. It is tempting to immediately imagine a new dashboard or application.
First make the information need concrete. Which decision or action must someone be able to take? Which data is required, where does it originate and when must it be available? This reveals where connections are missing.
See what the existing setup can support.
The right route depends on the process, the systems' capabilities and the maintenance the organisation can provide. These three directions offer a starting point:
| Direction | What to investigate |
|---|---|
| Reconfigure existing software | Can the application already support the desired process with a suitable setup? |
| Connect systems | Can available interfaces exchange data and actions at the right time? |
| Build focused custom software | Is a specific way of working, interface or function missing that the process needs? |
A combination may fit. Existing applications might remain the source of data while a limited custom solution supports the everyday task. Establish who manages which information and what should happen when errors or changes occur.
Explore where AI can make information more useful.
AI can take on a supporting task within a digital solution. Examples include summarising selected information, initially classifying a message or suggesting an answer to a question.
For an assistant using business knowledge, sources, access rights and answer checks belong in the design. Explore how current information becomes available and how users can check the basis of a proposal.
Make the handover between systems explicit.
For each connection, describe the information exchanged, the trigger and the expected outcome. Agree which system is authoritative for which data and how to handle missing or conflicting information.
For an AI step, add what the proposal may contain and which review follows. A message classification might help an employee choose the right next step. Decide separately which subsequent actions may happen automatically.
Consider use and maintenance in your choice.
A solution should remain understandable after initial delivery. Who notices a failed handover? Who can change the process? What happens if one of the connected systems changes?
For AI, also consider data use, review of outcomes and monitoring changes in the application. Maintenance needs and recurring costs are part of deciding which setup fits.
Your first step: draw one flow of information.
Choose one situation where information is now collected or handed over manually. Draw the systems involved and note the information needed at each handover. Add the task or decision that follows.
This provides a practical basis for comparing configuration, integrations and custom software. hfc helps examine the connections and can build the chosen solution, including a focused AI application where it supports the work.
The practical approach in this article was developed by hfc. Background on AI workflows and evaluating generative AI: