--- title: "Follow the friction: Five signs a process is ready for an AI workflow" date: 2026-10-27 author: "jeff schreibman" url: https://www.restontechwiz.com/articles/follow-the-friction-ai-workflow-signs/ description: "Follow-ups, status checks, missing information and repeated handoffs: the five signs that a process is ready for an AI workflow, and what to do first." categories: - "AI workflows for small business" --- # Follow the friction: Five signs a process is ready for an AI workflow ## Start with the work your team has to chase Look for the work that moves only because someone remembers to push it. A new inquiry sits in the inbox. A client has sent three of four documents. An accepted proposal waits for someone to tell operations what was sold. Those are useful places to look for an AI workflow. Spend a week noting each interruption: what was waiting, who chased it, and where the needed information lived. Start with one recurring process that has a clear owner and finish line. The examples below are illustrations to help you spot that process, rather than reports of client results. ## Sign: a reply waits more than a day An inquiry that waits a day gives you a specific delay to investigate. Is the hold-up finding the right person, reading a long message, or checking which details are missing? Separate the time spent waiting from the time needed to write a useful response. A first workflow could read the inquiry, suggest a category, and prepare a reply for the assigned person. It should flag uncertainty and keep the original message available. A receipt acknowledgment can confirm arrival; it should never suggest that a person has reviewed something they have not. Measure time to a useful first reply, alongside how often staff correct the draft. ## Sign: the same data is typed twice Repeated typing is a sign to inspect the connection between your tools. Imagine a coordinator copying names, job details, and dates from an email into a customer record, then into an onboarding sheet. Write down which record should be authoritative before connecting the others. If every input is a clean form field, a direct integration may solve the problem. An AI step becomes worth testing when the information arrives in varied language or documents and needs interpretation. Have it suggest structured fields, retain their source, and ask a person about conflicts. Check duplicates and corrections as well as time saved. Faster entry is little help if it creates a second version of the truth. ## Sign: someone chases the next step The next-step chaser is the person who knows which thread needs another message. Make their decision rules visible: what starts a follow-up, what stops it, how long to wait, and who takes over when there is no reply. For example, an onboarding workflow could compare received information with a checklist and draft a request for only the missing items. Give reminders a limit and pause them when the client responds. If the next action depends on reading the reply and choosing among permitted tools, a bounded [AI agent](/service/ai-agents/) may fit. If the route is fixed, keep the workflow simple. [Anthropic's explanation](https://www.anthropic.com/research/building-effective-agents) makes this distinction between predefined workflows and agents that choose their steps. ## Sign: status questions interrupt the work An update is useful only when it comes from a record someone trusts. Before drafting status messages, agree on what each state means, who changes it, and which events clients should hear about. A shared queue may be the first improvement. Consider a service request that moves from received to assigned to awaiting information. A workflow can prepare an update when the recorded state changes. It must not turn silence into a claim that work is underway. Show the source and its last update; route stale or conflicting records to the owner. Count status questions before and after the trial, and check whether the updates answer what clients actually wanted to know. ## Sign: the handoff drops details A handoff needs enough context for the next person to act. Ask the receiving team which details they have to request again. An accepted proposal might need an agreed scope, promised dates, exclusions, client contacts, and a named delivery owner. A workflow could gather those details, link to the source documents, and prepare an onboarding summary. Someone should confirm commitments before the next task is released. [Microsoft's approval workflow documentation](https://learn.microsoft.com/en-us/power-automate/get-started-approvals) shows how an automated flow can wait for a person's decision. Keep a route for rejection and revision, too. Measure missing-detail requests and handoffs returned for clarification, rather than counting summaries generated. ## What a first workflow looks like for each sign The [four first-project examples on our AI Workflow Automation page](/service/ai-workflows/#experience) cover these five signs. Repeated entry often belongs inside inquiry triage or the proposal handoff. Choose the smallest route that carries a real request from arrival to an agreed finish. Use the table as a starting brief. For each candidate, collect a few ordinary examples and an awkward exception. Our article on [how prototypes reveal what to build first](/articles/vibe-coding-better-scope/) explains how a rough working version can help clarify scope. Compare the proposed workflow with the simplest form, checklist, or integration that could do the job. ### Match the friction to a first project *Examples only; agree on the baseline first.* | Sign | First-project shape | Check in the trial | |------|---------------------|-------------------| | Replies wait; details are copied | Inquiry triage and follow-up | First useful reply; correction rate | | Someone chases the next step | Missing-information chaser | Complete submissions; repeat requests | | Status questions interrupt | Status updates without the asking | Check-in messages; stale updates | | Handoffs drop details | Proposal to onboarding handoff | Missing details; returned handoffs | ## What stays with your team Your team owns the commitments and exceptions. Keep prices, promises, sensitive disclosures, and unusual cases with the responsible person. Before the trial, agree on five boundaries: - Data: which records the workflow can read. - Actions: which changes it can make. - Approvals: who clears a draft or commitment. - Exceptions: who receives conflicting or missing details. - Stop control: who can pause the workflow and restart it. Set a limit on retries and reminders, record what happened, and decide who receives failures. [OpenAI's guide to building agents](https://openai.com/business/guides-and-resources/a-practical-guide-to-building-ai-agents/) recommends human intervention when failures exceed set limits or an action carries significant risk. Bring one process, a few real examples, and the name of the person who chases it. We can review where it stalls, whether AI would help, and what a useful first project would include. [Ask for a quote](/contact/) once that scope is clear.