PROJECT BUYER GUIDE

Automation, retrieval or fine-tuning?

Choose the approach after defining the problem. Different kinds of friction call for different kinds of work.

When the problem is a sequence of actions

If people repeatedly copy fields, move requests between tools or chase approvals, begin by mapping the workflow. Automation can connect those steps. Use conventional rules for deterministic decisions and AI only where interpretation or drafting adds value.

When the problem is missing software

A spreadsheet workaround, fragmented user journey or absent internal tool may need focused software development. Define the users and the essential journey. Adding a model is optional; the application should make sense as a product even before AI is considered.

When the problem is finding information

If answers live in changing documents, retrieval can bring relevant source material into the workflow. Evaluate search quality, source support and permissions. Fine-tuning is not a substitute for keeping a current document collection available.

When adaptation may be justified

A repeated, well-defined task with representative labeled examples may justify a model experiment. Check data rights and quality, compare against a prompting baseline and agree an evaluation plan. Do not begin with a guaranteed improvement target.

A useful decision sequence

Describe the task and unacceptable errors. Check whether rules or software can solve it. If interpretation is needed, test prompting with representative examples. Add retrieval when approved knowledge is missing. Consider adaptation only after identifying a specific baseline limitation.

What to bring to discovery

Bring the current process, a few safe examples, existing tools and your definition of useful output. Ask for an assessment of alternatives and the assumptions behind the proposed scope. The simplest maintainable approach may be the right one.

START A CONVERSATION

Turn a business problem
into a clear project.

Bring a task, a bottleneck, or a product idea. The first step is a clear scope.

Discuss your project