Last year we did something that felt slightly uncomfortable at the time. We used our own product on ourselves.

I know that sounds obvious. But there is a version of building AI tools where you spend all your time deploying them for clients and never actually live with the friction yourself. We had been doing that. We knew what our agents could do in theory. We had seen the results at client companies. But we had not fully committed to replacing our own internal workflows.

So we did. Three of them. Here is what happened.

The three workflows we replaced

I want to be specific here because vague case studies are useless. The three workflows we targeted were lead qualification, weekly reporting, and onboarding.

Workflow 1: Lead qualification

Before the agent, this looked like this. A new lead would come in through our website or through an introduction. Someone on our team would look at the lead, do a bit of research, figure out whether they looked like a real fit, draft a reply, and put them into our CRM at the appropriate stage.

Simple on the surface. But across a week, with 20 to 30 incoming leads, it was eating three to four hours of someone’s time. Not because any individual task was hard. Because each one required context-switching, a bit of research, a judgment call, and then execution.

The agent we deployed handles all of this. It reads the incoming inquiry, pulls publicly available information about the company or person, compares their profile against our criteria for a good client fit, assigns a lead score, drafts a personalised first-touch email, and updates the CRM with all the context it gathered. The whole thing runs in under two minutes.

What would have been a four-hour weekly task for a person now takes us maybe 20 minutes of review. We read what the agent found, approve or adjust the draft email, and move on.

Workflow 2: Weekly reporting

Every Monday morning we used to spend a chunk of time pulling numbers. How many leads came in? Where are we on revenue? What did the team work on? How is the platform performing?

The information existed. It was just scattered across six different tools. So someone had to log into each one, find the relevant numbers, copy them into a shared doc, and write a short summary. Start to finish, about two hours.

The agent now does this overnight on Sundays. By the time I sit down Monday morning, I have a single document in my inbox with everything in it. Lead volume and sources. Revenue pipeline. Platform usage stats. Key notes from the team’s activity logs. It pulls from every system and assembles it in a format I actually find useful.

Two hours went to zero.

Workflow 3: Client onboarding

When a new client signs on with Omni, there is a stack of things that happen in the first few days. Welcome email with next steps. Setup questionnaire sent and tracked. Kick-off call scheduled. Access provisioned for the relevant systems. Initial briefing doc created based on their intake form.

Before the agent, one of our team members handled all of this manually. It was reliable, but it was entirely dependent on that person having bandwidth. When things got busy, onboarding slipped. Welcome emails went out a day late. Questionnaires did not get chased.

The agent now fires the moment a deal is marked as closed in our CRM. It sends the welcome email, triggers the questionnaire, follows up if the questionnaire is not completed within 48 hours, schedules the kick-off based on calendar availability, and creates a starter briefing document from whatever information we already have on the client.

The agent does not replace the human relationship. Our team still does the kick-off call. They still build the actual strategy. But the admin scaffolding that holds the process together now runs automatically.

The honest friction

I want to tell you about what went wrong too, because nobody tells you this part.

The first week of the lead qualification agent was rough. The scoring logic was too aggressive. It was filtering out leads that deserved a proper conversation because they did not match our typical profile on the surface. We had to tune the criteria three times before it felt right. If we had just set it and walked away, we would have missed good opportunities.

The reporting agent hallucinated once. It pulled the wrong date range for a revenue figure and I nearly sent an incorrect number to an investor. Caught it in review. That was a reminder that review is not optional. The agent drafts. A human approves.

Onboarding automation surfaced a process problem we did not know we had. When we tried to write the agent’s logic, we realised our onboarding steps were not documented anywhere. They lived in the head of the person who always did it. The agent could not be built until we wrote the actual process down. That turned out to be a useful side effect.

What changed in our team

The clearest change has been in how our team spends their time, and in how they feel about their work. This is the real AI story — not replacement, but what your team can do when the mechanical work gets handed off.

Before the agents, there were days where a talented person spent most of their day on tasks that required zero of their talent. Email triage, data entry, chasing questionnaires, formatting reports. The kind of work that is necessary but not interesting.

After the agents, that same person is working on strategy, client conversations, and creative problem solving. The agent handles the mechanical parts. The human handles everything the agent cannot do.

It is a better use of everyone’s time. And honestly, it is better for the quality of the work.

Let me put some numbers to it, because I think the specifics matter more than the narrative.

Lead qualification was consuming three to four hours per week across the team. It now takes about 20 minutes of human review time. That is a reduction of roughly 85 to 90 percent.

Weekly reporting went from two hours to zero, because the output is ready before anyone shows up. The only human time involved is the ten minutes I spend actually reading the report.

Onboarding was not a time problem before, it was a consistency problem. Steps got missed. Timing was unpredictable. Now it runs identically every time, in the same sequence, with the same quality. We have had clients comment on how responsive and organised our onboarding feels. That reputation used to depend on whoever happened to be managing onboarding at the time. Now it is structural.

The total time saving across the three workflows is somewhere between five and eight hours per week. That is a meaningful chunk of a small team’s capacity. But the number I actually care about more is the quality improvement. The consistency. The fact that nothing falls through the cracks because someone was busy.

What I would do differently

A few things, knowing what I know now.

I would start with documentation before I started with technology. We wasted time during the onboarding agent build because the process was not written down. If I had spent a day documenting the process before we touched any agent tooling, the build would have been faster and cleaner.

I would also set a clearer review cadence from day one. We were casual about checking the agents’ output in the first few weeks. Fortunately nothing went badly wrong. But the reporting agent issue, where it pulled the wrong date range, could have been more serious. Now we have a standing weekly review where someone on the team checks a sample of each agent’s output. It takes 20 minutes and it catches things before they compound.

And I would define success criteria before deploying. “Does it work?” is not a useful question. “Does it qualify leads with 85 percent accuracy compared to our best human qualifier?” is a useful question. Having a specific benchmark makes it easier to tune the agent and to know when it is performing well enough to trust.

Why we do not recommend going it alone

I debated whether to include this section because I know it sounds self-serving. But I am going to say it because it is true.

We tried to build our first agent internally, without a proper ops structure around it. It took about three weekends. It worked well for six weeks. Then an API changed, the agent started behaving strangely, and nobody had time to debug it properly. It sat broken for almost two months.

That experience is what shaped how we built Omni Ops. Every agent we deploy for clients has monitoring, maintenance, and an ops layer behind it. When something breaks or needs tuning, it gets fixed. Not next month. Now.

Building your first agent is achievable. Keeping it running while you also run a business is the hard part.

What to do if you want to start

If you are reading this and thinking about your own workflows, here is where I would start.

Pick the workflow your team hates most. Not the biggest or most complex one. The one that grinds people down. The one that is purely mechanical, requires no real judgment, and just needs to get done. If you want a clearer sense of what agents are actually doing inside businesses right now, this breakdown covers a day in the life of an AI agent.

Document it. Write out every step as if you are explaining it to a new employee on day one. This is not just prep for the agent. It is also useful for you. Most people discover, as we did with onboarding, that their process is less documented than they thought.

Then come and talk to us. We run a discovery session where we look at the workflow, map the agent logic, and tell you honestly whether this is worth building. If it is not, we will say so.

The three workflows we replaced are not special. Every business has versions of them. The question is just whether you are ready to hand them off.

Book a discovery call for Omni Ops — we will show you exactly what this looks like for your business.