Why use an unrestricted model for this
Mainstream AI assistants are tuned to be agreeable. Ask one to tear apart your pricing and it praises your effort. Ask it to think like your toughest competitor and it gives you a balanced overview. Ask it to write a realistic scam call so your staff can practise spotting one, and it declines.
We run this work on our own unrestricted models, the same abliterated builds we publish on Hugging Face, so the analysis is blunt and complete. They run on our hardware in Melbourne, so your plans, numbers and client lists are never sent to a cloud AI provider.
Unrestricted means the model doesn't soften the analysis. Our rules for the work still apply: public sources only, your own business as the target of any attack, and nothing we wouldn't put our name to. See the ground rules.
What we do
Look outward at the market, inward at your own weak points, or both
Competitor teardowns
A full picture of who you're up against, built from public sources: their websites, pricing, ads, reviews, job listings, social channels, search rankings and public company records.
- Pricing and offer comparison, side by side
- Review mining: what their customers complain about
- Ad and messaging analysis, with the gaps they leave open
- Hiring and expansion signals
- Where you can win, ranked by effort
Red-team your own plans
We point the model at your business plan, pitch deck, pricing, website or launch and have it attack them as a sceptical investor, a ruthless competitor, an unhappy customer or a regulator.
- Pre-mortem: "it's a year from now and this failed, why?"
- Hostile investor and buyer objections
- How a competitor would respond to your launch
- Website and sales copy critique without the flattery
- A prioritised fix list, not just criticism
Negotiation war-gaming
Rehearse the hard conversation before it happens. The model plays the other side, whether that's a supplier, landlord, buyer, a big client or a union rep, and it plays them as tough as they're likely to be.
- Their likely position, leverage and walk-away point
- Live role-play rounds you can repeat
- Your best alternatives if the deal falls over
- A one-page brief to take into the room
Scam & social-engineering drills
Invoice fraud, fake supplier emails and "it's the boss, I need a transfer" calls are hitting small businesses hard. We build realistic, harmless test versions aimed at your own team so they learn to spot the real thing.
- Your own staff only, with written sign-off
- Phishing emails, texts and pretext call scripts
- A map of where your payment process is weakest
- A short training session on the results
What you get back
Plain-English reports you can act on, with every claim traced to its source
| Deliverable | What it is |
|---|---|
| The report | Findings ranked by how much they matter to your business, with the source for every fact about a competitor. Things the model inferred are labelled as inference, so you know what is evidence and what is a judgement call. |
| Action list | What to change, in priority order, with a rough effort and cost for each item. |
| Walkthrough call | We go through the findings with you and answer questions. You're welcome to push back on the model's conclusions. |
| Your data back | Anything you gave us is returned or deleted at the end of the job, whichever you prefer. Nothing is kept for training. |
How it works
Every job is scoped on a free consultation and quoted before we start
- Free call. Tell us who you're up against, what you're planning, or what's worrying you.
- Scope and quote. We pick the parts that matter for your situation and send a fixed quote. No surprises.
- The work. Research, red-team runs and war-game sessions, with a person checking every finding.
- Handover. Report, action list and a walkthrough call. Ongoing competitor monitoring is available if you want it.
Want to run this yourself? We can set up the same private, unrestricted model on your own hardware. See AI setup and quantisation & abliteration.
Ground rules
An unrestricted model makes the analysis sharper. It doesn't change what's legal, or what we're prepared to do.
We will
- Use public information about competitors: websites, ads, reviews, listings, public records
- Attack your own plans, processes and staff, with your written OK
- Be blunt with you, even when it isn't what you hoped to hear
- Keep your material private and off cloud AI services
We won't
- Hack, log in to or scrape anything behind a password or paywall
- Contact or deceive a competitor's staff, or pose as anyone
- Write fake reviews, smear content or anything defamatory
- Profile private individuals or collect personal information
- Run a scam drill on anyone outside your own business
Questions we get
Is competitive intelligence legal?
Yes, when it's built from public information, and that's all we use. Big companies have whole teams doing this. We give smaller businesses the same view without the corporate price tag.
Why not just use ChatGPT?
For a lot of things you should. For this work, mainstream assistants hedge, soften criticism and refuse the adversarial parts. They also run on someone else's servers. Our models are blunt, and they're private.
Can AI get things wrong?
Yes. That's why every fact in the report comes with a source, inference is labelled as inference, and a person checks the findings before they reach you. The model does the heavy lifting and we check what it produces.
Will my competitors know?
No. Reading a public website or a review page doesn't notify anyone. We don't contact competitors or their staff.
Find out what your competitors already know
Tell us who you're up against, or what you're about to launch, and we'll scope it on a free consultation.