Any team can wrap an LLM in a chatbot in a weekend. The hard part is everything above the model — and that's what you're actually buying.
Each layer is independently irreproducible by a Claude prompt. Together they're the product. Read top-to-bottom — the layers get harder to replicate as you descend.
DARA reads from AML engines, KYC platforms, integrity feeds, treasury systems, POS layers, BMS frameworks, industrial historians. Read-only, no rip-and-replace. The LLM doesn't normalise an OPC UA stream into a regulator-aligned event log — DARA does, before the LLM ever sees the data.
A Claude prompt can't connect to industrial historians, BMS frameworks, AML/KYC engines, or integrity-data feeds. The plumbing is the product.
Every signal, report, and audit-log entry is routed to the correct regulator stack. AUSTRAC + NCPF + ACMA for AU; UKGC + AMLD6 + FCA for UK; FinCEN + state DGEs for US; MGA + AMLD6 for EU. An operator carrying 4 licences gets ONE compliance brain, not 4 regional teams running disconnected tooling.
An LLM doesn't know that AUSTRAC's 24-hour SMR clock is different from FinCEN's 30-day SAR clock. DARA does, because the rulebooks are config, not prompt.
Every read, every output, every reasoning step is audit-logged at the database layer. Defensible under AUSTRAC, NCPF, UKGC, FinCEN, MGA review without a fire drill. Not a transcript log — a signed, schema-versioned, regulator-aligned evidence chain.
A Claude response is opaque to a regulator. DARA's audit trail is structured evidence that an examiner can subpoena, replay, and cross-reference.
Per-operator database. Per-operator AES-256 encryption key. Zero cross-tenant queries. Every read audit-logged. Engineered at the architecture layer, not promised in a contract.
If two operators ran the same Claude prompt, neither can prove their customer graph wasn't visible to the other. With DARA, the tenant boundary is provable by inspecting the database, not by reading the SLA.
DARA produces ONE morning brief covering AML/CTF · Responsible Gambling · Integrity · Advertising — across every licence the operator holds. The Head of Compliance gets a 5-minute read before standup instead of stitching 5 separate dashboards built by 5 different vendors.
Anyone can prompt an LLM with one signal. Nobody is synthesising 4 regulator pillars × N jurisdictions × M point-tool outputs into one operationally useful brief in 5 minutes. That's a year of product work, not a prompt.
Not 11 lookup tables — 11 living rulebooks that route signals, structure reports, format evidence chains, and version themselves as regulators update. Adding a new licence to an existing operator is configuration, not a six-month engineering project.
By 2027 the underlying LLM will be commoditised — Claude, Gemini, OpenAI, Mistral, Anthropic will all be interchangeable. The rulebooks won't. They take years to assemble and they are the durable asset.
Every brief, PDF, regulator report, and analyst email DARA produces renders in the operator's brand — not in DARA's. Switching off DARA means switching off months of operator-branded audit artifacts that may already be cited in a regulator file.
Switching costs aren't built by integration depth alone — they're built by the existence of branded, regulator-cited artifacts the operator now relies on. Every day DARA runs, the lock-in deepens.
Every operator considering DARA arrives with one of these four objections. The honest answer to each.
“We already use Claude internally for compliance summaries.”
Good — so you've validated the demand. DARA isn't a Claude prompt. It's the seven layers above the prompt. Your engineering team built layer-01 (a chatbot). The other six take 12-18 months and require domain depth no engineering team carries by default. We've already shipped them.
“Our engineers can build this internally.”
They can. At ~A$200k/year per senior engineer × 2-3 engineers × 18 months = A$600k-900k of internal build cost, plus ongoing maintenance, plus opportunity cost (those engineers aren't building your product). DARA's Paid POC is A$45-60k for 6-8 weeks. The math doesn't favour build at your scale.
“Anyone can wrap Claude in a chatbot.”
Correct — and DARA isn't a chatbot. It's an orchestration layer with jurisdiction routing, regulator-grade audit trails, tenant isolation, and 11 jurisdictional rulebooks. The chatbot at the bottom is the part you can replicate in a weekend. The other 95% is the product.
“What happens when the LLM gets better / cheaper / replaced?”
DARA upgrades the LLM. Your operational continuity doesn't change. The rulebooks, audit trail, tenant isolation, and brand-aware artifact engine all remain. We swap the model behind the curtain. Operators who hardcoded Claude into their internal tooling have to re-architect.
They each sell the product layer their customers actually use — the workflow, the orchestration, the structured outcome — and they each disclose their infrastructure subprocessors in their privacy and security pages because that's what mature B2B SaaS does. DARA's the same shape. The LLM is one of our ingredients. What you're buying is the meal.
If the seven layers landed for you, the next conversation is whether DARA fits your operator stack. Operator-to-operator. Honest read on whether we're a 6-week paid POC, a 12-week Lighthouse, or not the right fit right now.