Use Cases
Three Ways to Roll Out Enterprise AI
Same scenario, three radically different approaches. A 500-employee company wants to deploy AI across Engineering, Customer Support, and Finance using AWS + Azure + GCP. Here's what each path actually costs. In time, money, and risk.
The Scenario
A 500-employee company wants to deploy AI across 3 departments. Engineering, Customer Support, and Finance, using AWS + Azure + GCP.
Three Approaches, One Goal
The Consulting Route
Timeline: 6-12 months
Year 1 cost: $500K-$2M
Ongoing: $100K-$200K/yr maintaining custom integrations
Token usage: ~2M tokens/month once live (single-vendor, unoptimized)
Models: GPT-4o for everything (Azure-only, vendor lock-in from consultant's recommendation)
Monthly AI spend: ~$30/month at 2M tokens… but they don't get here for 6-12 months
Total Year 1: $500K-$2M+ with AI not live until month 8-12
- e.g., Deloitte, Accenture
- Consulting engagement + internal resources
- Consultant picks one model, one cloud. Strategy is stale by delivery
- Models change quarterly; their recommendation is a point-in-time snapshot
- Compliance recommendations delivered as PDFs, not automation
The Patchwork Route
Timeline: 4-6 months
Year 1 cost: $200K-$400K
Ongoing: $8K-$15K/yr tooling + 2-3 platform engineers at $150K+ each
Token usage: ~10M tokens/month (multi-cloud, no cost-optimized routing)
Models: GPT-4o (60%), Claude 3.5 Sonnet (25%), Gemini Flash (15%). No intelligent per-task routing
Monthly AI spend: $200-$400/month at 10M tokens on premium models
Total Year 1: $250K-$450K with 2-3 FTE tied up in platform maintenance
- DIY with point solutions: Portkey ($499/mo routing) + Helicone ($150/mo observability)
- Custom compliance scripts + LangChain/CrewAI for agents
- Separate RAG pipeline per cloud provider
- Every new department = new integration project
- Team spends more time maintaining the stack than building features
Bonito
Timeline: Same day → 1 week
Year 1 cost: ~$17K
Ongoing: $499/mo Pro + 3 agents at $349/mo each
Token usage: ~10M tokens/month (same workload, smart routing)
Models: Nova Lite (60%), GPT-4o Mini (20%), Gemini 2.5 Flash (15%), GPT-4o (5%), auto-routed per task
Monthly AI spend: $60-$80/month at 10M tokens with smart routing
Total Year 1: ~$18K all-in
- Same-day connection to all 3 clouds, 1 week to all departments live
- Smart routing sends each task to the optimal model automatically
- 60% of requests go to Nova Lite for classification and drafts
- Only 5% of requests need GPT-4o, complex analysis only
- Built-in governance, compliance, and cost attribution from day one
Side-by-Side Comparison
Consulting | Patchwork | Bonito | |
|---|---|---|---|
| Time to first AI in production | 6-12 months | 4-6 months | Same day |
| Year 1 all-in cost | $500K-$2M | $250K-$450K | ~$18K |
| Monthly AI inference spend | ~$30 (single model) | $200-$400 (unoptimized) | $60-$80 (smart routing) |
| Monthly token volume | ~2M (single use case) | ~10M (multi-team) | ~10M (multi-team) |
| Models in use | 1 (consultant's pick) | 3-4 (manual selection) | 4-6 (auto-routed per task) |
| Unified governance | ❌ PDF recommendations | ❌ Manual scripts | ✅ Built-in real-time |
| Agent governance | ❌ | ❌ DIY | ✅ Default-deny, budget caps |
| Compliance automation | ❌ One-time audit | ⚠️ Custom scripts | ✅ SOC-2/HIPAA/GDPR checks |
| Cost attribution | ❌ | ⚠️ Partial (per-tool) | ✅ Per-key, per-team, per-request |
| New department rollout | New engagement ($$$) | New integration (weeks) | Add an agent (minutes) |
| Vendor count | 1 expensive one | 5+ | 1 |
| Engineers required | 0 (outsourced) then 2-3 | 2-3 FTE | 0 (self-serve) |
Risk Deep-Dive
Every approach has trade-offs. Here's an honest look at the risks.
The Consulting Route
- Model obsolescence. GPT-4o today might not be optimal in 6 months, but the consultant's strategy is locked in
- Vendor lock-in. The consultant picked one cloud, one model vendor. Switching costs are enormous.
- No operational layer. You get a strategy deck and architecture diagrams, not running infrastructure
- Knowledge walkout. When the consultant engagement ends, institutional knowledge leaves with them
- Compliance is a snapshot, one-time audit recommendations in a PDF, not continuous automated checks
The Patchwork Route
- Integration fragility. One vendor pushes an update and breaks the chain. You're now debugging 5 vendor APIs.
- Security gaps between tools. Portkey handles routing, Helicone handles logging, but neither handles the gaps between them
- No unified audit trail, compliance has to stitch together logs from 5+ tools to answer 'who accessed what, when'
- Agent sprawl without governance. LangChain/CrewAI agents run without default-deny or budget caps
- Engineering team becomes the 'AI platform team'. They spend more time maintaining integrations than building product features
Bonito
- Newer product. Less battle-tested than established consulting firms or Portkey's 3+ year track record
- Smaller provider catalog: 3 cloud providers (AWS, Azure, GCP) vs Portkey's 60+. If you need a niche provider, it may not be supported yet.
- Single vendor dependency. Mitigated by OpenAI-compatible API format (portable) and standard IaC (Terraform) for infrastructure definitions
- No SOC-2 Type II yet. In progress, but not complete. May be a blocker for some enterprise procurement processes.
The Bottom Line
$500K-$2M
Consulting Route
6-12 months to first AI in production
$250K-$450K
Patchwork Route
4-6 months, 2-3 FTE dedicated
~$18K
Bonito
Same day to connected, 1 week to all departments
Bonito CLI
Deploy AI agents across any provider from one YAML file
Sound like your team?
If you're running AI workloads across multiple cloud providers and want unified control without the infrastructure overhead, Bonito was built for you.