Build it yourself vs. DoThat
What it actually takes to put governed, client-facing AI in production — not a demo, the real thing.
| Capability | DIY on a generic LLM wrapper | DoThat |
|---|---|---|
| Basic chat integration | 1–2 weeks · $5k–15k | Included, live in days |
| Security & data perimeter (encryption, access control, secure data handling, pen testing) | 3–6 months · $80k–250k | Included |
| Regulatory compliance (GDPR, EU AI Act, data residency, DPIAs, audit trails, ongoing legal) | 4–8 months · $100k–350k+, plus ongoing counsel | Compliance-ready infrastructure |
| Prompt security & guardrails (injection defense, brand rules, output controls) | 1–2 months · $20k–60k | Built in |
| RAG & vector infrastructure (embeddings, retrieval, data pipelines) | 2–3 months · $40k–120k | Included |
| Engagement analytics & insight | 1–2 months · $25k–80k | Included |
| Ongoing maintenance, compliance upkeep & model updates | 2–3 engineers + compliance, permanently · $250k–450k/yr | Covered |
| Total to reach production | 12–24 months · $600k–1.5M+ plus a permanent team | Days, one plan |
The question is not what it costs to succeed. It is what it costs to be wrong.
Most AI ideas do not work. Build in-house and the floor cost of finding out is a team, an infrastructure bill and two quarters. Adopt, and the floor cost is a week and some usage — so you get more attempts for the same money, and it takes attempts to find the one that works.
Killing an idea here costs nothing structural. Switch it off. You are not left holding a codebase and a hiring plan attached to something that failed.
These aren't edge cases — they're the minimum for putting AI in front of clients with confidence. Every row is something we've already built.
Skip the build. Choose a plan, or book a demo to see it in action.