The question of when to hire another operations person versus when to automate a workflow is one of the more consequential decisions a growing company makes. Hire too early and you have a person spending most of their time on tasks that could run on their own. Hire too late and your current ops team is buried, dropping things, and moving slowly on the work that actually requires judgment.
This post is about five specific workflows that we see consistently eating operations time at 20-to-60-person companies, and that are good candidates for automation before a hire. These are not edge cases. They are the workflows that come up in almost every conversation we have with ops directors at companies in this size range.
A note on scope: this list applies to companies with general operations functions, not companies with heavily specialized ops requirements like manufacturing, logistics coordination, or regulatory compliance. Those categories have different automation priorities.
1. Vendor follow-ups
Vendor follow-up is the single most consistent time sink we see in ops at growing companies. It is not one big task. It is dozens of small, repetitive tasks spread across the week: checking whether an invoice was received, following up on a quote that hasn't been returned, confirming delivery timelines, chasing payment acknowledgments from vendors you've already paid.
Each individual follow-up takes five to ten minutes when you factor in finding the original thread, drafting a message, and logging the interaction. Multiply that by 20 or 30 vendor relationships and you have multiple hours per week going into work that follows a completely predictable pattern.
This workflow automates well because the inputs are structured (invoice dates, quote request dates, vendor contact info), the outputs are templated (follow-up emails follow predictable formats), and the failure mode of an automated follow-up being slightly off-tone is low-consequence compared to an invoice going unpaid because nobody followed up. The agent drafts the follow-ups, a human reviews and sends. Total review time per week: typically under 20 minutes once the templates are calibrated.
2. Expense categorization
Expense categorization is the kind of work that feels fast when you sit down to do it, but the sitting-down part is the problem. Most ops people at growing companies categorize expenses in batches, which means they accumulate until someone carves out time to go through them. The backlog creates its own overhead: reconciling categories from a month ago requires going back and remembering context that has decayed.
AI-based expense categorization has been around for a while in accounting tools, but the agent approach adds a layer that accounting software typically doesn't provide: escalation for ambiguous cases. An agent that categorizes automatically for clear cases and escalates to a human for ambiguous ones produces cleaner data than either full-manual or fully automated categorization, because the human attention is concentrated on the cases where it matters.
The agent reviews incoming expenses daily, categorizes what is unambiguous, and batches the ambiguous cases for a weekly review. This produces more current financial data and reduces the monthly reconciliation overhead that many small finance teams dread.
3. Meeting scheduling and coordination
Meeting scheduling is a well-understood automation problem and there are dedicated tools for it, but most of those tools handle the simple case (you share a link, the other person picks a slot) without handling the more complex coordination that ops teams typically own: scheduling meetings with multiple external parties who don't use the same tool, managing rescheduling when things change, handling the confirmation and reminder sequence that precedes important meetings.
A scheduling agent that sits on top of your calendar system and handles the back-and-forth coordination covers more of this than a link-sharing tool. It can draft scheduling emails in context, handle rescheduling requests with human review, and send the reminder and prep-note sequences that ops teams often manage manually for high-value meetings.
The scope boundary here is important: the agent handles the coordination and communication layer, not the judgment about which meetings should happen or with whom. Those decisions stay human. The agent saves the time that goes into the logistics after the decision is made.
4. Weekly status reports
Internal status reporting is a significant time sink that is almost invisible because it gets absorbed into the end of the week in small chunks. Someone has to pull data from multiple sources, synthesize it into a summary, format it appropriately for the audience, and distribute it to the right people. At a 40-person company, this might be a two-hour task every Friday that nobody counts as "work" because it feels like overhead.
Status report generation is a good automation candidate because the data sources are known, the report structure is consistent, and the consequence of a slightly imperfect draft is low since a human reviews before distribution. The agent pulls from the relevant systems (project tracker, CRM, finance dashboard), generates a structured first draft in the established format, and routes it to the ops lead for review and send.
The review typically takes 10 to 15 minutes rather than the original two hours, because the human is editing a draft rather than composing from scratch. Over a year, this is a substantial reclaimed capacity.
5. Inbound request routing
Growing companies accumulate informal request channels faster than they formalize processes for handling them. By the time a company has 40 people, requests come through Slack, email, project management tools, and verbal handoffs, and someone on the ops team is effectively acting as a human router: reading each request, figuring out who should handle it, forwarding it, and following up when it falls through the cracks.
Inbound request routing is automatable precisely because the classification step (what type of request is this, who should handle it, what priority) follows a learnable pattern. The agent reads incoming requests across channels, classifies them, routes them to the right queue or person, and flags items that have been sitting without a response for too long. The human router role becomes a review-and-override role instead of a from-scratch classification role.
This is often the automation that ops leaders feel most skeptical about before deploying, because request classification feels like it requires judgment. In practice, the majority of requests at a growing company fall into a small number of recurring categories that the agent handles well. The novel or ambiguous requests are the minority, and those still go to a human. The agent handles the volume; the human handles the exceptions.
The question this list does not answer
These five workflows are good automation candidates, but automating them is not a substitute for thinking clearly about what your operations function actually needs. If your company's primary bottleneck is that nobody has defined the processes these workflows are supposed to follow, automation will surface that problem quickly. An agent cannot follow a process that doesn't exist.
The teams that get the most out of these automations are the ones who used the configuration process as an opportunity to make their processes explicit: what does a vendor follow-up sequence actually look like, what are the expense categories and when is each one appropriate, what defines a high-priority inbound request. That clarity is valuable independent of the agent, and it tends to outlast any specific automation tool.
