What is the difference between AI in the workflow and AI beside it?
AI beside the workflow lives in another tab: a portal, a report, a chatbot someone has to go visit. AI in the workflow appears inside the tool the person already has open, at the moment they need it. Adoption is decided almost entirely by this. A mediocre suggestion inside the dispatch board gets used. An excellent one behind a separate login does not.
The extra-tab tax
Every additional destination you ask a busy person to visit costs you most of your users. Not because they are lazy, but because their day is already a queue of interruptions and the new tool has to beat the next interruption for attention. It rarely does.
This is why so many AI projects show strong pilot results and weak usage six months later. Nothing about the output degraded. The novelty that carried people to the extra tab wore off.
The pattern is consistent enough to plan around. When usage falls off after the first month, the first thing to examine is not output quality but delivery surface.
Where 'in the workflow' actually means
- On the record. The call summary as a note on the job. The lead score as a field on the contact.
- In the queue they already clear. A task assigned in the CRM beats an email about a task.
- In the channel they already read. The message thread, the group chat, the morning brief.
- In the tool of the role that acts. A dispatcher, a CSR and an owner need different surfaces; one shared portal serves none of them well.
The diagnostic
Ask the person who is supposed to act: what were you doing right before you saw this? If the honest answer is that they went looking for it, the AI is beside the workflow. If the answer is that it appeared while they were doing the work, it is inside it.
The same output can be either. A coaching note is beside the workflow in a monthly PDF and inside it when it lands attached to the rep's own call, which is why coaching data is delivered per rep rather than as a report.
Timing counts as much as location. A flagged unbooked call delivered at 8am while the customer is still shopping is in the workflow. The same flag in a Friday summary is a history lesson.
When beside the workflow is the right answer
Exploration is different. When someone is investigating a question nobody anticipated, they need a place to slice, filter and compare, and that belongs in a dashboard or an analyst tool. Trying to embed exploratory analysis into an operational surface produces clutter that everyone learns to ignore.
The rule of thumb: recurring decisions go in the workflow, open-ended questions go in the analysis layer, and the two read from the same underlying data so they never disagree.
Topics: adoption · workflow · ux · integration
Have a version of this question about your own business?
The useful answer usually depends on which systems you run and how they're connected. That's a conversation, not a blog post.