How we built an AI tool smart enough to know our business inside and out, solving one of our most tedious tasks in the process.
How we built an AI tool smart enough to know our business inside and out, solving one of our most tedious tasks in the process.
Lauren Leamon
Consultant and Designer
Lauren designs practical, human-centred solutions at the intersection of people, technology and business. With experience across healthcare, retail, FMCG and tech products, she translates insights into actionable strategies that drive alignment and impactful outcomes.
Known for her energy, curiosity and strategic mindset, Lauren brings people together through storytelling and collaboration.
Working in a small company means everyone does a bit of everything. Strategy, sales, delivery, operations, admin, marketing… You get it. The reality is that some days, most of the tasks eating up your time aren’t the ones that need you. They’re the ones that just need doing.
We wanted to see how AI could actually help with that. Not in a “replace the team” way but in a “stop spending the afternoon on something a machine could do in a couple of minutes” way.
To be completely honest, I still don’t really know how I feel about AI. The ethics are murky, the pace is unsettling, and I’m protective of the skills that make our work human. But, in a small team, any time you can reclaim for the work that actually needs you is worth exploring, and any time we can save procurement teams from having to evaluate non compliant RFQs is appreciated, I’m sure! So we started exploring.
What this covers:
The first question is choosing what to hand over
There’s a lot of work that shouldn’t be automated or handed over to AI. So before choosing what to start testing new AI tools on, I set a filter, the task I chose to automate had to tick most of these boxes:
If it ticks those boxes it’s fair game.
The problem I chose: qualifying tenders
Here’s the one that was at the top of my frustrations list. As a small company, we’re always scanning for opportunities, tenders, panels, EOIs. The qualifying stage is rough, not because it’s hard, but because it’s time-consuming, repetitive, and often ends in “nope, not this one.” You could manually read through all the documentation, but tender documents can run to dozens of pages and the critical details like non-negotiables, weighting criteria, hidden requirements are scattered throughout.
My workaround wasn’t terrible. I’d copy and paste Avian’s capability overview into Copilot or Claude, drop in the tender documents, and ask it to assess our suitability. But I was constantly going back and forth. “Have you looked at this section? What about our experience on this project?” I trusted the output, but I had to double-check everything. Then, if we were a good fit, I’d ask it to find our most relevant case studies, which meant crafting the right prompt for that application, waiting for the platform to search through our case study database, and flagging areas that needed rewriting.
And that’s only the qualifying which is a tiny fraction of the actual proposal process. After that comes project planning, the pitch strategy, filling out documentation, writing the response (the enjoyable part).
None of this was the end of the world, which is exactly why it persisted.
This was the problem that I highlighted this as an efficiency that wouldn’t take away from the team’s skills or take away from work we love doing.
Building the tool with Claude
Building this qualifier tool with Claude genuinely impressed me. I prompted it to build exactly what I needed: a platform that could ingest tender documentation and qualify it against our capabilities.
It built something. I had feedback. Quite a few rounds actually, but each time I was adding complexity, not just repeating myself which I often found myself doing on other platforms.
Each feedback round was adding capability. “Can it find relevant case studies too? Can it rewrite them to fit? Can I have a chat function to ask specific questions about the documents? Can we keep a history so we can return to the same tender later?”
Then we got into the experience layer. “Can we brand it? Can we reorder the information? Can we fix these UX issues?” All making it feel good to use, not just functional.
Then I started getting logistical (and a little ambitious). “Can I publish this somewhere my team can access it? Can we make it private so it’s just ours?”
Next thing I know, Claude (and Josh, our tech advisor) were teaching me how to publish this platform through GitHub so the whole team could access the tool and its history.
Why Claude was a good fit for this job
Three things made Claude the right tool for this:
Visual output. It wasn’t just giving me text back, it built an interface I could see and interact with, making feedback fast and specific.
Quick iterations. Changes didn’t take days. I could describe what I wanted, see it, refine it, and move on. The feedback loop was tight.
A shared platform. The end result lives somewhere my whole team can use it. It went from my workaround to an actual business tool.
Watch the demo below.
*for privacy purposes we have cut the video off at the ‘find case studies’ step.
The result
The platform works. I drop tender documentation in and it qualifies it for us, a score out of 100, the reasoning behind it, who we could partner with to boost our capabilities, any non-negotiables we don’t meet, and which case studies we should reference in the application.
While my feelings around AI remain skeptical, this is a use case I’m happy to stand behind. We’re not asking it to think for us or replace the craft of us creating and writing a compelling response. We’re asking it to scan two sets of data, our large bank of case studies and the requirements for a tender, connect the dots and surface the facts. This platform has shifted from generic AI efficiency to deeply personalised efficiency. Of course, we check the outputs because AI is a tool not a decision-maker. I have gained back some time which I can now spend on parts of the proposals I enjoy, the creative strategy, the human-centred thinking, the pitch. As an added bonus, I somewhat know my way around GitHub.
The takeaway for small businesses
If you’re working in a small company wondering where to start with AI, start here: find the task that frustrates you, doesn’t need your creativity, and follows clear rules. Build a solution around that and make sure you always keep humans in the loop.
We’re here to help—and always happy to chat.