Four useful areas to investigate are research, customer follow-through, internal coordination, and reporting. Below, I explain what a possible workflow could involve, where a person remains responsible, and what I would evaluate before calling it an improvement.
These are illustrative possibilities, not completed client projects. My starting point is the work the business needs to do, not a list of tools.
Where does someone spend time gathering information? Which handoffs need repeated attention? What makes it difficult to respond to a customer or decide what matters next?
Those questions can reveal opportunities. They do not determine in advance which system should be built.
Here are four areas I would consider during discovery. They are starting points for investigation, not four products every business needs or a required order of implementation.
1. Research and finding information
A team may repeatedly search documents, websites, or past assignments to prepare the next piece of work.
An AI-assisted approach could help organize information, identify potentially relevant material, or prepare a first-pass summary. The useful system also needs source references, appropriate access, and a way to review the output.
Consider a hypothetical team preparing an account brief. Faster text generation is only one part of the job. The facts need to be current, relevant, and traceable. The system should distinguish between information it found and an interpretation it produced.
I would evaluate the time to a reviewed result, the amount of correction required, and whether the work can be reused. Generating more pages is not necessarily progress.
2. Customer-facing work and follow-through
Another opportunity may be the work around answering an inquiry, preparing a proposal, onboarding a client, or keeping someone informed.
AI could help draft an update or organize a request. Ordinary automation could handle routing and reminders. A portal might make required information and next steps easier to understand.
These do not need to be separate competing projects. The right approach depends on where the experience is breaking down.
For example, a hypothetical service business’s onboarding problem may look like a need for better email writing. Discovery may show that clients are receiving inconsistent instructions instead. The first improvement could be a clearer process, with AI supporting only part of it.
Useful measures might include time to an approved response, missing-information requests, or whether clients complete the required steps. Customer communication should have review appropriate to its importance and consequences.
3. Internal coordination and repeated administration
Look for work that involves copying information, chasing status, assigning tasks, or checking whether something happened.
A connected system might reduce that effort without using AI at every step. Defined rules can handle predictable routing. AI may be useful when incoming information varies and needs interpretation.
The important question is what happens when the interpretation is wrong or the expected information never arrives.
I would want clear ownership, visible exceptions, and an understandable recovery path. Quietly moving the wrong information faster is not an improvement.
Anthropic’s engineering guidance distinguishes predefined workflows from more open-ended agents. That supports choosing the form of the system around the task rather than treating autonomy as the goal.1
Evaluate the completed process: not only fewer clicks, but fewer missing handoffs, less rework, and a clearer picture of what still needs attention.
4. Reporting and decision support
Sometimes the work is already recorded, but understanding it requires opening several tools or rebuilding a report.
A useful reporting system starts with a decision. What does the owner or team need to notice? What action would follow? How current must the information be?
AI might help explain a change or prepare a narrative summary. The underlying numbers still need defined sources and calculations. A confident explanation should not obscure incomplete data or an unsupported conclusion.
Consider a hypothetical weekly delivery review. A useful view might show which assignments need intervention and why. A less useful one might display twenty attractive charts with no clear action attached.
Evaluate whether the information helps someone make a better-informed decision, not whether the dashboard looks sophisticated.
Choose based on the business, not this list
The best first opportunity may be large or small. It should have a worthwhile purpose, feasible implementation, appropriate safeguards, and people willing to use the result.
Microsoft’s AI planning guidance makes a similar distinction between business impact, feasibility, and user desirability.2 An idea can be technically impressive and still be the wrong priority.
A business with a strong pipeline but slow delivery may need a different starting point from one that lacks qualified opportunities. A team with unreliable data may need to address that before adding more analysis.
At Fidelis, growth strategy helps establish what matters. Discovery identifies where the work needs to change. AI and custom systems are ways to make the right improvements happen.
For a way to compare these possibilities, read how to identify AI opportunities in your business. You do not need to select one of these categories before contacting me. Describe the business and what you hope to improve. We can investigate the opportunity together.
Related reading
How to Identify AI Opportunities in Your Business
AI Implementation for Small Businesses: From Plan to Working System
Anthropic, “Building effective agents.” https://www.anthropic.com/engineering/building-effective-agents↩
Microsoft Learn, “Business plan for AI agents.” https://learn.microsoft.com/en-us/azure/cloud-adoption-framework/ai-agents/business-strategy-plan↩