Every small team has some hidden work that is repetitive and unseen. For me, that was triage, invoice matching and the Monday report. I spend about 15 hours a week on these tasks and none of them require any real brain power, but they do need your attention, which is why they’re very costly. Eight months ago we decided we weren’t going to take this hidden work for granted. Here’s how we set up our agents, what went wrong and what we’d recommend to any team of 10 or less.
Start with the Friction
The most important thing we did was spend a week understanding our processes. We spent two weeks logging any tasks we felt were repetitive. This gave us three clear agents.
It’s important to note that smaller teams can easily get caught up in the novelty of agents. Whilst it’s impressive to have an agent that can generate clever marketing copy, it’s much more valuable to have an agent that removes a task no-one wants to do. This will ensure your agent sticks around after the first month.
Operation One: The Inquiry Triage Agent
We have an inbox where we receive around 120 messages a week. These can be anything from billing questions, feature requests and the odd sales lead. This needs to be triaged, tagged and routed to the correct person, taking around 6 hours.
Our agent reads each message in the inbox and sorts them into the correct category out of 7. It also then generates a draft response using our documentation and routes it to the correct person. However, it will never send the messages automatically. All drafts are put into a review queue.
We made it reliant on human input. During the first month, we had a high error rate of around 1 in 8 messages being in the wrong category. Usually this was due to the message not clearly belonging to one category, but as the messages were being checked by humans we didn’t lose any customer confidence, and only lost a few seconds. By the third month we had fine tuned our categories and provided examples of edge cases. Our accuracy increased to over 95% and now most drafts are approved with only a few changes.
Operation Two: Invoice Reconciliation
This was really what we needed this project for. Matching invoices from vendors to purchase orders requires you to spot patterns. However, there are lots of tiny issues that can occur. Sometimes numbers are swapped round, rounded numbers or line items are mentioned in 3 different ways.
Traditional automation struggles with this as it can only match exact matches. Our agent uses Language Models to understand what near matches might be and can explain why they think they match. We extract the information from the invoice pdf, compare it to our records and sort them into 3 categories; certain matches, likely matches that just need a quick check and actual mismatches.
Related:
- How AI Is Rewriting the Rules of Business Opportunity
- The Future of Independent Businesses
- Transforming Industry Expertise into Intelligent, Purpose-Built Solutions
Our finance lead now spends maybe 15 minutes reviewing the flagged items instead of 200. We’ve reduced our time spent on this from 5 hours a week to less than 1 hour. We’ve also found 2 billing mistakes that wouldn’t have been found.
Operation Three: The Monday Report
Our weekly metrics required us to pull numbers from 4 different places. We then had to paste them into a template and write up some commentary. It was a bit of a pain and the commentary wasn’t always great as we were tired.
Our agent pulls the data and calculates the changes from week to week. It then creates a story based summary of the data. We still have a human read through it to add some context that we couldn’t get from the data and send it on. Interesting to note, our reports are better. As we don’t have to create them we’re spending our energy on interpreting instead.
The Lessons That Transfer
A few lessons were learnt that can be used in other areas.
Specific is better than general. Each agent has a specific purpose. Our initial attempt at creating an ‘operations assistant’ was very unreliable and difficult to troubleshoot. Specific agents will have predictable issues that can be resolved.
Write down your logic. We saw the most improvements in accuracy when we documented the rules we had in place. Using automation means you’re able to see what knowledge you have within your organisation.
Be honest with your measurements. We measured how much time we saved, how many errors we had and how much review we had. We didn’t have success with everything. We tried to create a 4th agent to organise meetings but gave up as the coordination was too much.
What we actually gained
Overall we have saved around 12 hours a week meaning we have the equivalent of a part time employee. However, we’ve also gained culture. No-one is scared of Mondays anymore. We’re focusing more on the thinking side of things instead of the repetitive.
AI agents aren’t magic. They aren’t a replacement for people who think. They are a way to give your small team the leverage that large teams have had. You can absolutely do this but will you know what friction to take away first?
Digital entrepreneur and content expert I help businesses with AI, SEO and the latest tech trends. I started Silicon Valley Weekly to make complex tech concepts easy to understand and use for business growth. I know a lot about systems and help startups, entrepreneurs and brands navigate the fast-changing world of tech and online marketing.
I build strategies that use data, search optimization, content marketing and AI tools to get visibility, engagement and revenue. I love finding ways for businesses to grow increasing their presence and turning new ideas into successful businesses. My goal is to connect the technology, with practical business use so brands can succeed online.