ROI, TCO & Business Case Analysis
AI investments need financial justification. Learn how to build ROI analyses for AI-powered business solutions, calculate total cost of ownership, and present business cases that get executive approval — including when to build, buy, or extend from a cost perspective.
Making the money case for AI
Your CEO doesn’t care about model parameters or prompt engineering. They care about one question: “Will this make us money or save us money?”
ROI (Return on Investment) measures whether the benefits of an AI solution outweigh its costs. TCO (Total Cost of Ownership) captures ALL costs — not just the software licence, but development, training, compute, maintenance, and the time your team spends managing it.
As a solutions architect, you need to build a business case that proves the investment is worth it — with numbers, not just excitement about AI.
ROI criteria for AI-powered business solutions
ROI isn’t just about cost savings. The exam expects you to select the right criteria for different types of AI investments.
| Feature | Criteria Type | How to Measure | Example |
|---|---|---|---|
| Direct cost savings | Quantitative — reduced headcount, fewer errors, lower processing costs | Reduction in manual data entry hours, decreased call centre volume | Agent handles 60% of support tickets, saving 4 FTE |
| Revenue uplift | Quantitative — increased sales, faster deal cycles, higher conversion | Improved lead scoring accuracy, faster quote generation | AI-powered lead scoring increases win rate by 12% |
| Productivity gains | Quantitative — time saved per employee per task | Hours saved on report generation, meeting preparation, email drafting | Copilot saves 30 minutes per employee per day on average |
| Risk avoidance | Semi-quantitative — reduced probability of costly events | Compliance penalties avoided, security incidents prevented, audit failures eliminated | AI compliance monitoring prevents a potential regulatory fine |
| Strategic value | Qualitative — competitive advantage, innovation capability, employee satisfaction | First-mover advantage, talent retention, customer experience improvement | AI-powered customer service rated 20% higher in satisfaction surveys |
Exam tip: selecting ROI criteria for the scenario
The exam often asks which ROI criteria are most appropriate for a given scenario. Match the criteria to the business context:
- Cost-cutting initiative — Direct cost savings (FTE reduction, error reduction)
- Revenue growth initiative — Revenue uplift (sales velocity, conversion rates)
- Digital transformation programme — Productivity gains + strategic value
- Compliance/risk project — Risk avoidance (penalty prevention, audit readiness)
- Customer experience improvement — Strategic value + revenue uplift (satisfaction, retention)
The wrong answer is usually “just measure cost savings.” The right answer includes multiple criteria types — because AI investments often deliver value across several dimensions.
Total cost of ownership: the full picture
TCO for an AI-powered business solution includes costs that many people forget:
| Cost Category | What’s Included | Often Forgotten? |
|---|---|---|
| Platform licensing | Copilot Studio, M365 Copilot, Foundry compute, D365 AI add-ons | No — this is the obvious cost |
| Development | Solution architecture, agent building, testing, integration work | Sometimes — internal team cost is often underestimated |
| Data preparation | Cleaning, structuring, indexing, connector development | Yes — this is typically 40-60% of the initial effort |
| Model compute | Inference costs (per token), training/fine-tuning compute | Sometimes — scales with usage, hard to predict |
| Change management | Training users, updating processes, managing adoption | Yes — the technology works but people don’t use it |
| Ongoing maintenance | Model monitoring, knowledge source updates, prompt tuning, security patches | Yes — AI solutions need continuous care, not “set and forget” |
| Opportunity cost | What else could the team have built with the same time and money? | Yes — rarely quantified but important for prioritisation |
Scenario: Adrienne builds the business case for Vanguard's AI programme
Adrienne needs to justify a $2.4M AI investment to Vanguard’s board:
Costs (TCO over 3 years):
- Platform licensing: $800K (Copilot Studio + Foundry + M365 Copilot licenses)
- Development: $600K (internal team + Cloudbridge consulting fees)
- Data preparation: $400K (data cleansing, connector development, index building)
- Model compute: $300K (inference costs, estimated based on usage projections)
- Change management: $200K (training, adoption programmes)
- Maintenance: $100K/year = $300K over 3 years
Benefits (projected over 3 years):
- Direct cost savings: $1.8M (40% reduction in call centre volume — 12 FTE)
- Productivity gains: $1.2M (30 min/day saved for 500 employees using Copilot)
- Risk avoidance: $500K (estimated reduction in compliance penalty risk)
- Revenue uplift: $400K (faster loan processing → more loans closed)
ROI calculation: ($3.9M benefits - $2.4M costs) / $2.4M = 62.5% ROI over 3 years
Payback period: ~18 months (benefits exceed costs in the second year)
Marcus (CISO) adds a note: “The risk avoidance figure is conservative. Our industry’s average compliance fine is $5M. Even a 10% reduction in probability justifies the security investment.”
Build, buy, or extend: the cost perspective
Module 4 covered the architecture fit decision. This module adds the economic lens:
| Feature | Upfront Cost | Ongoing Cost | Time to Value | Total 3-Year Cost |
|---|---|---|---|---|
| Prebuilt agent | Low (licensing only) | Low (included in platform) | Days to weeks | Lowest — typically included in existing D365/M365 licensing |
| Extend M365 Copilot | Medium (plugin development) | Low (Copilot licensing + minor maintenance) | Weeks | Low to medium |
| Build in Copilot Studio | Medium (design + development) | Medium (licensing + content updates + monitoring) | Weeks to months | Medium |
| Build in Foundry | High (models + compute + engineering team) | High (inference costs scale with usage) | Months | Highest — but justified for high-value, unique use cases |
Deep dive: the hidden costs of building custom
The exam often tests whether you understand the hidden costs of custom AI solutions:
Talent costs: Finding and retaining AI engineers is expensive. A Foundry-based solution needs data scientists, ML engineers, and platform engineers.
Maintenance debt: Custom models need retraining as data changes. Knowledge indexes need rebuilding. Prompts need tuning as user behaviour evolves.
Integration complexity: Custom solutions need connectors to D365, M365, and external systems. Each connector is a point of maintenance.
Governance overhead: Custom models need responsible AI review, security audits, and compliance checks — none of which come free.
The exam’s message: Build custom only when prebuilt and extended solutions genuinely can’t meet the requirement. The most cost-effective architecture uses prebuilt for 70%, extends for 20%, and builds custom for only the 10% that truly needs it.
Creating an ROI analysis: step by step
For the exam, know this structured approach:
- Define the business process — what specific process will AI improve?
- Baseline current state — how long does it take today? What does it cost? What are the error rates?
- Project the AI-enabled future state — how much faster, cheaper, more accurate?
- Calculate benefits — map improvements to financial metrics (time saved × cost per hour, errors avoided × cost per error)
- Calculate TCO — all costs over the analysis period (typically 3 years)
- Calculate ROI — (Total Benefits - TCO) / TCO × 100%
- Determine payback period — when do cumulative benefits exceed cumulative costs?
- Sensitivity analysis — what if adoption is 50% lower? What if costs are 30% higher?
Exam tip: sensitivity analysis is the mature architect's answer
If an exam question asks about presenting an ROI analysis to leadership, the best answer includes sensitivity analysis — showing what happens if assumptions change. This demonstrates:
- Intellectual honesty (you’re not overselling)
- Risk awareness (you’ve thought about what could go wrong)
- Professional maturity (executives trust ranges more than point estimates)
A single ROI number is less credible than: “Our base case shows 62% ROI. In a conservative scenario (50% adoption, 30% cost overrun), ROI drops to 18% — still positive.”
Flashcards
Knowledge check
Jordan is building a business case for CareFirst Health's patient scheduling agent. The primary goal is to reduce call centre volume and improve patient satisfaction. Which ROI criteria should Jordan prioritise in the analysis?
A retail company is evaluating a prebuilt D365 agent for customer service (included in existing licensing) versus a custom Foundry-based agent with advanced product recommendation capabilities. The custom agent would cost $400K to build and $150K/year to maintain, but is projected to increase average order value by 8%. Annual revenue is $50M. What should the architect recommend to leadership?
When presenting TCO for an AI solution, which cost is MOST commonly underestimated by project teams?
🎬 Video coming soon
Domain 1 complete! You’ve covered all the planning essentials — from requirements analysis and strategy to multi-agent design, model selection, and ROI justification.
Next up: Copilot in D365 Customer Experience & Service — entering Domain 2 where we design the actual AI solutions, starting with Copilot customisation in Dynamics 365.