The most common question we hear from enterprise leaders: "What should we budget for AI?" The most common answer they get from vendors: whatever the vendor wants to charge. Here's the honest answer, based on what we've seen across dozens of engagements in New Zealand and Australia.
What You Need to Know
- Enterprise AI costs fall into three phases: discovery ($50-100K), build ($150K-1M+ per capability), and operate ($100-250K annually). Most enterprises budget for build and forget the other two.
- Integration is 60-70% of the build cost. The AI model is the cheap part. Connecting it to your existing systems, data, and workflows is where the money goes.
- The foundation approach costs more on capability #1 but saves 40-50% by capability #4. Budget for the platform, not just the project.
- Hidden costs kill AI budgets: change management, data preparation, governance, and the "pilot to production" gap are routinely underestimated by 2-3×.
- Budget ratio benchmark: 40% technology, 30% people/change, 30% governance/operations. If your budget is 80% technology, you'll build something nobody uses.
Recommended AI Budget Allocation
Source: RIVER Group, enterprise engagement data, 2023-2024
60-70%
of enterprise AI build cost is integration, not the AI model itself
Source: RIVER Group, enterprise engagement data, 2023-2024
2.6×
average cost overrun for enterprises that skip formal AI discovery
Source: McKinsey & Company, The State of AI in 2024
Phase 1: Discovery - The Investment That Saves Everything
Discovery is where you determine what to build, whether it's feasible, and what it will cost. Skipping it is the single most expensive mistake in enterprise AI.
What Discovery Includes
| Activity | Investment | Duration | Output |
|---|---|---|---|
| AI readiness assessment | $10-30K | 1-3 weeks | Maturity scorecard, gap analysis, prioritised action plan |
| Use case identification | $10-25K | 1-3 weeks | Ranked use case portfolio with ROI estimates |
| Technical feasibility | $10-30K | 1-3 weeks | Architecture recommendation, risk assessment, build estimate |
| Data assessment | $10-20K | 1-2 weeks | Data quality report, pipeline requirements, sovereignty review |
Total discovery investment: $50-100K
This feels expensive until you compare it to the alternative: building the wrong thing. The 2.6× cost overrun for enterprises that skip discovery means a $500K build becomes a $1.3M lesson.
The Discovery Credit
Many AI delivery partners (including RIVER) credit discovery investment against the build phase if you proceed within 90 days. Discovery isn't a sunk cost. It's the first phase of delivery.
Discovery Red Flags
If your discovery phase produces any of these, pause and recalibrate:
- No clear business case with measurable outcomes. "AI would be nice to have" is not a use case
- Data that doesn't exist or can't be accessed. AI without data is a demo, not a capability
- No executive sponsor with budget authority. AI initiatives without sponsorship stall at the first organisational hurdle
- Estimated ROI below 3x the total investment. The risk-adjusted return needs to justify the organisational effort
Phase 2: Build - Where the Budget Goes
The build phase is where most enterprises focus their budgeting, and where the biggest misunderstandings live.
The Cost Breakdown
For a typical enterprise AI capability (e.g., intelligent document processing, claims automation, contract analysis):
| Component | % of Build Budget | What It Covers |
|---|---|---|
| Integration and data engineering | 35-40% | Connecting to source systems, building data pipelines, ETL, APIs |
| AI/ML development | 20-25% | Model selection, fine-tuning, prompt engineering, orchestration |
| Application development | 15-20% | User interfaces, workflow tools, reporting dashboards |
| Testing and quality assurance | 10-15% | Accuracy validation, edge case handling, security testing |
| Deployment and infrastructure | 5-10% | Cloud setup, CI/CD, monitoring, scaling |
Enterprise AI Build Cost Breakdown
Source: RIVER Group, enterprise engagement data, 2023-2024
The critical insight: Integration and data engineering is the largest line item, not AI development. This is where vendor estimates consistently mislead. An AI vendor quoting $100K for "the AI solution" is quoting 25% of the real cost. The other 75% is connecting that AI to your actual business.
Build Cost Benchmarks by Capability Type
| Capability | Typical Range (NZ/AU) | Timeline |
|---|---|---|
| Document classification and routing | $80-150K | 6-10 weeks |
| Intelligent data extraction | $120-250K | 8-14 weeks |
| Knowledge base with AI search | $150-300K | 10-16 weeks |
| Workflow automation (multi-step) | $200-400K | 12-20 weeks |
| Customer-facing AI assistant | $250-500K | 14-24 weeks |
| AI foundation platform (first capability + shared infrastructure) | $300K-800K | 16-30 weeks |
$300K-800K
typical NZ/AU investment for an AI foundation platform with first capability
Source: RIVER Group, enterprise engagement data, 2023-2024
Foundation vs Project Economics
Here's the maths that should drive your budgeting approach:
Project approach (4 separate capabilities):
- Capability 1: $250K (14 weeks)
- Capability 2: $250K (14 weeks)
- Capability 3: $250K (14 weeks)
- Capability 4: $250K (14 weeks)
- Total: $1M, 56 weeks
Foundation approach (4 capabilities on shared infrastructure):
- Capability 1 + foundation: $350K (18 weeks)
- Capability 2: $150K (8 weeks)
- Capability 3: $120K (6 weeks)
- Capability 4: $80K (4 weeks)
- Total: $700K, 36 weeks
4 Capabilities: Project vs Foundation Approach
Source: RIVER Group, enterprise engagement data, 2023-2024
The foundation approach saves 30% on investment and 36% on time, and the savings compound with each additional capability.
Phase 3: Operate - The Budget Everyone Forgets
AI systems are not "build and forget." They require ongoing investment to maintain performance, adapt to changing data, and comply with evolving governance requirements.
Annual Operating Costs
| Category | Monthly Range | What It Covers |
|---|---|---|
| Model monitoring and maintenance | $3-8K | Performance tracking, drift detection, model updates |
| Infrastructure and hosting | $1-5K | Cloud compute, storage, networking, scaling |
| Governance and compliance | $2-5K | Audit logging, policy updates, regulatory alignment |
| User support and training | $1-3K | Help desk, new user onboarding, workflow updates |
| Enhancement and iteration | $3-8K | New features, accuracy improvements, scope expansion |
Annual operating budget: $120-350K depending on the number of capabilities and their complexity.
The Hidden Costs
These line items consistently catch enterprises off guard:
Change management ($30-80K per major capability). AI changes how people work. If you don't invest in training, workflow redesign, and stakeholder communication, adoption will be low and ROI will suffer. This is not optional.
Data preparation (variable, often $50-150K). Your data is rarely AI-ready. Cleaning, structuring, labelling, and building pipelines to maintain data quality is a prerequisite the budget must account for.
The pilot-to-production gap ($50-200K). A pilot that works on clean data in a controlled environment is not the same as a production system handling edge cases, integrating with legacy systems, and meeting enterprise security requirements. Budget 2-3× the pilot cost for the transition.
Governance framework development ($30-60K upfront). Building the governance framework (policies, risk classification, monitoring, incident response) is a one-time investment that pays dividends across every capability.
The Budget Template
Here's a practical template for a 2025 enterprise AI budget:
| Category | Year 1 | Year 2 | Year 3 |
|---|---|---|---|
| Discovery | $50-100K | $20-40K | $15-25K |
| Build (capabilities) | $300-800K | $200-400K | $150-300K |
| Integration | Included in build | Included in build | Included in build |
| Change management | $50-100K | $30-60K | $20-40K |
| Governance (setup) | $30-60K | - | - |
| Operations (annual) | $120-250K | $150-300K | $180-350K |
| Contingency (15%) | $80-195K | $60-120K | $55-110K |
| Total | $630K-1.5M | $460K-920K | $420K-825K |
Year 1 is the highest investment. By year 3, you're primarily operating and extending, not building from scratch. This is the compound advantage in financial terms.
The Contingency Rule
Always budget 15% contingency. Enterprise AI engagements encounter surprises: data quality issues, integration complexity, organisational change resistance. The organisations that handle surprises well are the ones who budgeted for them.
Five Budgeting Mistakes to Avoid
-
Budgeting for technology only. If your entire AI budget is vendor fees and cloud costs, you've missed 60% of the real investment. People and process matter more than technology.
-
Comparing AI costs to SaaS costs. Enterprise AI is not a subscription product. It's a capability investment with different economics. Comparing it to your CRM licence will make everything look expensive.
-
Expecting immediate ROI. The first capability takes longest and costs most. ROI compounds over time as the foundation matures. Budget with a 2-3 year horizon, not a quarterly one.
-
Under-budgeting for operations. The build gets executive attention and funding. Operations quietly gets underfunded, and AI systems degrade because nobody budgeted for maintenance.
-
Budgeting per project instead of per platform. Every separate AI project budget creates incentives to build in isolation. A platform budget creates incentives to build on shared infrastructure.
The honest answer to "what should we budget for AI?" is: more than you think for year one, less than you think for year three, and differently than you think for everything in between.