The short answer
Before adding AI, map how the work actually runs today: what triggers it, the inputs it needs, every handoff and decision, the exceptions, the outputs, and who is accountable. The map shows where AI can help and where the real bottleneck is a process problem AI will not fix.
Adding AI to a workflow you have not mapped is like renovating a building without floor plans. You may improve one room and make the hallway worse. An AI pilot can disappoint when the surrounding work is unclear: a key input arrives late, a decision has no owner, or exceptions need a different path.
Mapping does not require special software or weeks of analysis. It requires walking through the work with the people who do it and writing down what really happens, not what the procedure document says should happen.
The seven elements of a workflow map
A useful map captures the same seven elements every time. Together they give you a shared picture that leaders, the people doing the work, and anyone configuring tools can all read.
| Element | What to capture |
|---|---|
| Trigger | What starts the work, such as a form submission, an email, a signed contract, or a date |
| Inputs | The information needed, where it comes from, and its usual quality |
| Handoffs | Each point where work passes between people, teams, or systems |
| Decisions | Where someone makes a judgment call, and on what basis |
| Exceptions | The cases that do not follow the normal path, and how often they occur |
| Outputs | What the work produces, and who receives it |
| Accountable owner | The one person responsible for how the whole workflow runs |
How to run a mapping session
- Invite the people who do the work, not only the people who manage it.
- Pick one recent, real example and walk it from trigger to output.
- Write each step in plain language with a verb, such as “check,” “copy,” or “approve.”
- Mark every handoff and every waiting point.
- Ask what happens when something goes wrong, then capture those exceptions separately.
- Confirm the owner out loud. If nobody can name one, record that as a finding.
Questions that reveal the real bottleneck
Once the map exists, ask questions that separate AI opportunities from process problems. Some bottlenecks are about effort, and AI may help there. Others come from unclear ownership or missing information, and no tool will fix them.
Bottleneck questions
- Where does work wait the longest, and why?
- Which steps involve copying, re-keying, or reformatting the same information?
- Which handoffs most often lose detail or context?
- Which decisions get escalated because the criteria are unclear?
- What share of the volume is exceptions rather than the normal path?
- Which steps need human judgment, and which are mostly routine assembly?
- If AI drafted or summarized this step, who would review it, and how quickly?
Hypothetical example
the sales-to-delivery handoff
This is an illustrative scenario, not a client case. Picture a company where a salesperson closes a deal and the delivery team then starts the project. On paper, sales submits a handoff form. In practice, the map shows something different.
The trigger is a signed agreement. The inputs are scattered across the CRM, email threads, and the salesperson's notes. The handoff form is often incomplete, so the delivery lead schedules a call to ask the same questions the customer already answered. A key decision about scope is made informally, and nobody owns the handoff itself: sales considers it finished at signature, and delivery considers it unstarted until the form is complete.
The map points to two changes. First, a process change: name an owner for the handoff and define which details must be confirmed before delivery accepts the work. Second, a possible AI-supported step: draft a handoff summary from the CRM record and the deal notes, which the salesperson reviews and corrects before submitting. The summary only helps because the required details and the owner were defined first.
Turning the map into a redesign
With the current state mapped, sketch the proposed workflow beside it. Show which steps AI supports, where a person reviews, which handoffs change, and who owns each step. Keep exceptions on a human path until you understand them. Then share both versions with the people doing the work and ask what would break. That conversation is where adoption starts.
A map is also your baseline. When you later measure the new workflow, you will know what changed and where to look if results are not what you expected.