Disclaimer: This article was AI generated from my YouTube video transcript.
Why tenders are so painful
If you run a business here in New Zealand, chances are you’ve bid on tenders before, whether through GETS or a private process. These tenders can be worth tens to hundreds of thousands of dollars over a multi-year contract, so it really matters that you put your best foot forward every time.
The hard part is the sheer volume of documents you have to reference and understand. And when you bid on multiple tenders, you end up copying and pasting the same information over and over. That’s exactly where AI comes in. In this article I’ll show you how to set Claude up to look at your previous winning tenders, use them as reference, and automatically fill out new ones as they come through - even drafting the more complicated paragraphs, tweaked to the new brief.
What you need to get started
First, create a Claude account over on the Claude website. You’ll want the Pro plan, which works out to about NZD 40 per month. That gives you three things that matter here:
- Extended usage - so you don’t hit limits mid-draft.
- Data privacy over the information you feed into the tool, which is really important when you’re working with sensitive tender material.
- Claude Co-work - the big piece we’ll be using for this whole automation.
Make sure you also download the Claude desktop app, because that’s how we access Co-work rather than running it from the web browser.
Co-work vs Chat
What makes Co-work different to Chat? Chat is like ChatGPT - it can connect to your different tools and give you a text response back. Co-work actually works on your computer, inside a real file system. That means when you have a large amount of documentation, Co-work can go into all the different files, interpret them, edit them and create new ones. It’s working on your machine and your files, which is exactly what we need for tenders.
Load the tender docs into Co-work
Grab your tender documents - whether that’s a private tender or a relevant one you’ve found on GETS - and download all the related files into a single folder on your computer. For this walkthrough I used a real example: the New Plymouth District Council put out a chatbot solution tender, so I pulled all those files into one folder.
Then open the Claude app and head over to Co-work. Click into the “work in a project” option, choose a different folder, and select where your tender documents are sitting. It will ask whether it’s allowed to change files in that directory - choose always allow so it can edit and create files in there.
Connect your apps
At this point Claude has the tender docs, but it doesn’t know anything about you or your business yet. That’s where connectors come in. Go to Customize and connect your apps - there’s a whole range depending on where your business information lives:
- Microsoft / SharePoint / OneDrive - if your health and safety, company info and previous tenders sit there, search Microsoft, log in, and it can pull from there.
- Google Drive - search Drive and connect it if that’s where your files live.
- Notion - same again, connect it up.
Once connected, every time Claude answers a question it will look across these data sources for answers you’ve used before and reuse them.
If you don’t have good examples sitting in another tool, there’s a simpler option: open your project folder and create a new folder called company information. Drop anything relevant in there - company branding, previous example tenders, project reports, any background that would help someone draft a response without knowing you. Ask yourself: if a stranger had to answer this tender for me, what would they need? Put all of that in the folder.
Build a tender-drafting Skill
Next we create a Skill - a repeated set of instructions that Claude follows every time to answer a tender for you. Go back into Customize, choose create new skill, then create with Claude. A window pops up where you tell it everything you’re trying to achieve.
Here’s roughly what I told it: I’ll provide all the documents in a Co-work folder with the full request for tender from GETS. I’ll also give it access to my Google Drive, which has my company information and previous tenders in a tenders folder. My company is Harkness AI, and it can look at my website and any other connectors for projects and information. I want it to read all the documents, look at everything it has on me, and answer each question inside the actual RFT template - then colour-code each answer by confidence:
- Green - very confident in the answer (like dropping in my website or full legal name).
- Yellow - it’s provided an answer, but it needs a review.
- Red - it couldn’t find any information to answer that field.
You can customise this for whatever your business needs and improve it over time. Get a first version in place, run it, then tweak. Claude may ask a few clarifying questions in a nice pop-up window - answer those and it generates a skill.md file. Read through it, make sure everything is correct, and if anything needs changing (NZ spelling, key details) just write back and ask it to rewrite. When you’re happy, hit save skill in the top right and it syncs across every Claude session - phone, Chrome browser, different laptops. From then on, every tender draft uses that Skill.
One last check: back in Co-work, hit the plus sign, go to connectors, and confirm they’re switched on so they actually pull data.
Run it and read the output
Now you simply say “draft my tender response”. Claude has everything in the folder relating to the tender, a Skill with step-by-step instructions, and live connections into your software. This takes a bit of time to run through.
On the right-hand side you get a progress report - it plans out every task, pulls all the information and contacts, maps out the response pieces, drafts them, reviews them and packages them up. You can read through every skill and tool it used along the way.
My demo came back after about 20 minutes with 217 green fields, 22 yellow and 24 red, plus pricing. It exported the result to Word and PDF in the exact template that was completely blank at the start. Trading name, full legal name, address, website - all populated. It built out the executive summary, the solution architecture and the proposal based on other work I’ve done, and even pulled in references for a joint proposal I was running. Importantly, it flagged the thin areas for review - for example, it knew I hadn’t covered much on Maori implications and cultural considerations, so it highlighted that as something I needed to look at myself.
Bonus: auto-scan GETS for new tenders
As a quick bonus, you can set Claude up to scan GETS for you and surface tenders worth applying to - instead of trawling emails or the GETS site yourself. On the left, go to Schedule, then new task. Give it a name, a description and a set of instructions, then set a frequency - weekly, daily, whatever suits.
For a demo I ran, the task scanned GETS, pulled back the relevant tender information step by step, and for each relevant tender outlined the details plus a high, medium or low priority rating for whether to apply. It’s a simple addition that keeps your pipeline full without the manual checking.
Refine it over time
Like any AI tool, this gets you roughly 80-90% of the way there. There’s still refinement to do around your business information, formatting and the finer detail - but that’s the easy layer on top. Once the setup is running, the trick is to run it a few times, note the pieces you want changed, and write those back into the Skill. It updates its memory and instructions, and your next result gets better and better. It’s a bit of a time game to teach it to respond properly, but the more past winning tenders you feed it, the stronger the outputs become.
I’m Blake, an engineer who’s helped over 25 businesses across New Zealand and Australia with AI implementation. A lot of my recent work has been setting up Claude to connect into different systems and building Skills to handle the repetitive tasks. If you’re not sure how to get started and want a hand, reach out to me via the link in the description below. Cheers.
