Legal services are expensive because lawyers spend 40% of their time on work that doesn’t require a law degree. AI can fix that. But most LegalTech products fail because they optimise the wrong thing-they automate document generation when they should be redesigning entire workflows. We’ve built enough software to know the difference.
If you’re building in this space, or considering whether to build, this is where the real efficiency gains sit-and where the money is.
The Document Automation Trap
Contract generation tools are everywhere. Zappi, Lawgeex, LawGenius. They’re useful. They save time. But they’re a local optimisation in a broken system.
Here’s the reality: a mid-market law firm generating 50 contracts a month might save 5-8 hours weekly with smart automation. That’s roughly 260 hours annually, worth maybe AUD $65,000-$100,000 at standard billing rates. It sounds good. But the firm’s gross margin barely moves because they’re still billing by the hour, and they can’t raise prices without client pushback.
Document automation works best for firms that can (a) shift to fixed-fee pricing, or (b) redeploy the time saved into new client work. Most don’t do either, so the tool sits half-used.
The efficient gains worth building around aren’t in automating one task. They’re in removing entire categories of work.
Where AI Actually Saves Money: Process Elimination
Take due diligence for M&A transactions. Traditional approach:
- Junior lawyers manually read hundreds of documents and create a summary
- Senior lawyers review and flag issues
- Associates chase clarifications from the counterparty
- Process takes 4-8 weeks for a mid-market deal
An AI system that ingests all documents, extracts obligations and risks automatically, and flags contradictions can compress that to 5-10 days. But here’s the critical bit: you’re not automating the existing process slightly faster. You’re eliminating the first 60% of junior lawyer hours entirely.
For a firm doing 20 deals annually, that’s the difference between needing 2 full-time junior lawyers and needing 0.5. That’s a real cost saving-roughly AUD $140,000-$180,000 per year in salary. And it can be priced into the service without looking like a rate cut.
The firms winning with AI aren’t using it to speed up old workflows. They’re using it to eliminate entire phases.
The Architecture That Actually Works
If you’re building LegalTech, the most common mistake is overcomplicating the AI component. Founders often want to build a “smart” system that understands nuance. That’s expensive, unreliable, and slow to ship.
The real architecture looks like this:
- Extract, don’t interpret. Use structured extraction (regular LLMs are fine for this) to pull specific fields, dates, obligations, parties, and amounts from documents. Store as clean JSON.
- Flag via rules, not ML. Use straightforward boolean logic and threshold-based alerts. “If counterparty exclusivity period > 5 years, flag.” “If indemnity cap missing, flag.” These rules are maintained by lawyers, not engineers.
- Surface the delta. Show users what’s unusual or missing compared to a template or prior deal. The AI isn’t making legal decisions; it’s doing the comparative work a junior lawyer would.
- Plug in integrations last. Don’t build Slack bots or Salesforce connectors until you have a core product people pay for. Most legal software companies fail trying to be everything at once.
This is slower to sound impressive at a pitch meeting but far faster to ship, maintain, and actually deliver value. A startup can build version 1.0 of this in 8-12 weeks. A more sophisticated NLP approach takes 6+ months, and half the time you’ll find your users don’t need it.
The Pricing Model Trap
Most legal AI companies try to charge per-document or per-feature. That’s weak positioning.
Law firms don’t care about documents. They care about time saved and risk reduced. If your tool genuinely eliminates two junior lawyer hours per transaction, you should price it as a fixed fee per transaction (typically AUD $2,000-$8,000 depending on deal size and firm size), not as “AUD $5 per page reviewed.”
The firms that can’t justify that price point aren’t your market. They’re too small or too price-sensitive to buy software at all-they’re still writing contracts in Word and billing in spreadsheets. Don’t chase them.
The 200-500 person firms and in-house legal teams? They will pay per transaction if you save them 20-30 hours of labour. That’s clean unit economics and it scales without you having to increase customer support cost per user.
What Still Needs Human Judgment
Be clear-eyed about what AI can’t do. Legal advice requires judgment that sits outside data extraction. Recommending deal structure, negotiating strategy, risk tolerance-these are human calls, and they should stay human.
Your product should make those decisions faster and more informed. It shouldn’t make them for the lawyer. Every LegalTech product that’s tried to remove the lawyer from the loop has either failed or pivoted back to augmentation.
The winners are the ones that make a good lawyer 3-4x more productive, not the ones that try to replace a lawyer with AI.
Building vs. Buying
If you’re a law firm or in-house legal team reading this: before you build, check whether Doctrine, Kira, or LawGenius already solve your problem 80%. Most of the time they do. Building your own AI legal tool takes 6-12 months minimum and requires ongoing investment in training data and model maintenance. Use existing tools to get the quick wins, then think about custom build only if your problem is genuinely differentiated.
If you’re a founder building: you have a real opportunity in vertical-specific tools. General contract AI is crowded. AI for IP litigation workflows, regulatory compliance for specific industries, or automated legal research in narrow domains-those have less competition and clearer ROI stories.
If you’re exploring whether to build a legal AI product or invest in growth for an existing one, talk to Amora about your build. We’ve shipped software in this space, and we can tell you honestly whether your idea needs a full platform or a focused 28-day MVP.
The Real Play
LegalTech gets exciting when you stop trying to build smarter software and start trying to rebuild the firm’s business model around what AI enables. Fixed-fee pricing. Shorter timelines. Smaller teams handling larger deals. That’s worth building towards.
The efficiency gains aren’t in the AI. They’re in the process redesign the AI makes possible.
Got something you want built?
Amora Digital is an Australian software and AI agency. We scope it, build it, and ship it – live in 28 days. No offshore teams. No surprises.