Systems Thinking for AI Projects: Find the Real Bottleneck

Systems thinking helps you examine how people, information, rules, and delays interact around an AI project. Its practical value is finding where an intervention can improve the whole outcome. Making one task faster does not necessarily make the complete service faster, more accurate, or easier to use.

Before choosing an AI tool, follow the work from its initial trigger to the result that matters to the customer or colleague. Include waiting, correction, and handoffs, not just the steps visible in an application.

Follow one real case

Imagine a consulting business where proposals take a week to reach clients. The team assumes drafting is the problem and considers an AI writer. Following a recent proposal reveals a different picture: drafting takes an afternoon, but pricing approval and missing scope information account for most of the delay.

In this illustrative situation, faster writing may increase the number of drafts waiting for approval without reducing delivery time. Start by recording what actually happened, including timestamps where available. Avoid building your map solely from the official process, which may omit workarounds and informal decisions.

Map flows and responsibilities

For each step, identify the input, the person or system doing the work, the decision made, and the next recipient. Mark missing information and points where work waits. Note which team owns the overall outcome and which teams control individual parts.

Ask what returns for rework. A proposal may circulate because the initial brief does not define scope clearly. In that case, improving information at intake could reduce several later corrections. Look upstream and downstream from the task you first wanted to automate.

Distinguish activity from the bottleneck

A busy person is not automatically the main constraint. Find the step that limits the rate or quality of completed work. Examine queue length, waiting time, repeated rejection, and dependencies on scarce expertise. Use observations to test your explanation rather than assuming the most visible task is the cause.

Measure the outcome across the complete journey. For proposals, useful measures might include time from a complete brief to client delivery and the proportion returned for scope corrections. Counting generated pages would tell you little about whether the client receives a better proposal sooner.

Consider feedback and unintended effects

An improvement changes behavior elsewhere. Easier proposal generation may encourage staff to prepare more proposals, increasing review demand. Faster responses may increase incoming enquiries. A tool that removes one manual check may shift errors to a later, more expensive stage.

Write down a few plausible effects before the experiment. Decide what to monitor so that apparent local success does not hide a growing queue or declining quality. Include the people who will receive the output; they often see costs the originating team misses.

Test an intervention at the right point

For the proposal example, try a clearer intake form and a defined pricing approval route before adding broad automation. If drafting still limits performance, test AI assistance on that bounded step with appropriate review.

Compare the end-to-end outcome before and after the change, noting differences in workload and case complexity. Keep the map current as the process evolves. Systems thinking becomes useful through repeated observation and adjustment: it helps you choose where to act and recognize when a seemingly successful change has moved the problem somewhere else.

Continue learning with Innoba

Connect this exercise with Think Like an AI Strategist & Builder. For the underlying concepts, explore the book Systems Thinking for Builders and the THINK Academy. Choose a task from your own work, keep the artifacts you produce, and use the questions in this article to review your decisions as you develop the capability.

Related articles

• How to Identify and Prioritize AI Use Cases in a Business
• Redesign a Workflow Before You Automate It with AI