Mapping AI Use Cases to Business Capabilities
A list of AI use cases tells you what could be built. It does not tell you which work changes, who stays accountable, or what to do first. Mapping each use case to the business capability it acts on answers those questions, because the capability is where the work, the data, the people and the risk meet.
The method below is the one every Capstera AI-ready capability model applies to each of its capabilities. You can run it on your own capability map, or start from one of ours.
The Method, in Six Steps
1. Pick the level you will decide at
Profile capabilities at the level where you will decide and fund the work. For most organizations that is L2: specific enough to say where AI acts, broad enough to keep the number of decisions manageable. Use L3 when the L2 capabilities are too broad to rate as one piece of work.
2. Describe the work
Name the kind of work each capability involves, using a fixed list of work patterns so capabilities can be compared. The models use eight: transactional processing, rules-based decision, judgment-based decision, knowledge retrieval, content generation, coordination & orchestration, analysis & insight and relationship & empathy. A capability usually has two or three, and the mix says a lot: rules-based decisions and transactional processing suit automation, while relationship and empathy work usually stays with people.
3. Set a target AI mode
Decide how far AI should go in each capability, not just whether it applies. A five-step scale keeps the answer precise:
| Human-led | AI plays no material role; the work depends on human presence, trust or accountability. |
|---|---|
| Assist | AI retrieves, summarizes or drafts on request; a person does the work and owns every output. |
| Augment | AI proactively recommends, scores or prepares; a person decides and acts. |
| Automate | AI executes defined, bounded tasks end to end; people handle exceptions and sample-check results. |
| Agentic | AI agents plan and run multi-step work across systems toward a goal, within guardrails, escalating to people at defined checkpoints. |
4. Decide what stays human, and the risk
For each capability, write down what stays with a named person, and set the oversight that goes with the AI mode:
| Human decides | AI informs; a named person makes and records every decision. |
|---|---|
| Human-in-the-loop | AI proposes; a person approves before anything takes effect. |
| Human-on-the-loop | AI acts; people monitor in real time and can intervene or reverse. |
| Monitored autonomy | AI acts; people review samples, metrics and exceptions after the fact. |
Then assign a risk tier (Low, Moderate, High and Restricted), defined for the domain by what an error would affect. The tier decides who approves the use case and which controls it needs before it goes live.
5. Score and prioritize
Rate each capability from 1 to 5 for AI fit, value, feasibility and data readiness. AI fit has done its work in step 3, setting the mode. The ranking then weights value, feasibility, data readiness and risk, where a lower risk tier scores as easier, into a risk-adjusted priority score, and you set the weights to fit your strategy. A heatmap of the result shows leadership, on one page, where AI pays off and where it doesn't.
Data readiness is the rating that answers "how ready are we?" capability by capability: it shows where the data a use case needs is clean and accessible, and where it isn't yet.
6. Sequence into waves, and attach the use cases
Group the ranked capabilities into three waves (Wave 1, Wave 2 and Wave 3), with the lower-risk, readiest work first and the governance that later waves need set up early. Only now attach the specific use cases and agents to each capability, each with its own risk tier, controls and approver. A use case that maps to no capability usually means the map is missing something, or the use case doesn't belong in the plan.
Assessing AI Readiness
Readiness has two parts. Capability by capability, the data readiness rating in step 5 shows where the data is ready. Across the organization, a structured assessment scores the conditions every AI program depends on, such as executive sponsorship, data quality and governance, and shows which have to improve before the later waves.
Each Capstera AI-ready model includes a 40-question readiness assessment, scored in its workbook, so the result sits next to the capability profiles it qualifies.
Worked Example: Customer Service and Experience
The Customer Service and Experience model applies the method to every one of its L2 capabilities. The table shows the averages for each L1 area, the same heatmap its product page shows, with the number of L2 capabilities rated High risk.
Ranked by average value and AI fit, the areas to look at first are Field Service Management; Service Performance and Insight; and Self-Service and Digital Service. The High-risk count tells you how much oversight the first use cases in each area will need.
| L1 area | L2s | Avg. AI fit | Avg. value | High risk | Target AI modes |
|---|---|---|---|---|---|
| Field Service Management | 5 | 3.8 | 4.2 | 3 | Augment 3, Automate 1, Assist 1 |
| Service Performance and Insight | 5 | 3.6 | 4.2 | 1 | Augment 5 |
| Self-Service and Digital Service | 4 | 3.8 | 4.0 | 3 | Automate 2, Augment 2 |
| Service Channel and Contact Center Management | 5 | 3.4 | 4.2 | 2 | Augment 4, Automate 1 |
| Case and Request Management | 5 | 3.6 | 4.0 | 2 | Augment 3, Automate 2 |
| Complaint and Feedback Management | 4 | 3.5 | 4.0 | 2 | Augment 4 |
| Service Knowledge Management | 4 | 3.5 | 3.8 | 1 | Augment 3, Assist 1 |
| Customer Account Servicing | 6 | 3.5 | 3.8 | 4 | Augment 5, Automate 1 |
| Service Workforce and Quality Management | 5 | 3.4 | 3.8 | 3 | Augment 4, Assist 1 |
| Customer Experience Strategy and Design | 5 | 3.2 | 3.8 | 1 | Augment 5 |
| Customer Success and Onboarding | 5 | 3.0 | 3.4 | 1 | Augment 5 |
One capability's full profile, Case Investigation and Resolution (CX-4.2), shows what the method produces. It is part of the package's public sample, 0.79% of its content.
Case and Request Management › Case Investigation and Resolution
Case Investigation and Resolution (CX-4.2). The ability to find what is wrong, put it right for the customer and confirm that the case is resolved.
| Work patterns | Knowledge retrieval, Judgment-based decision, Relationship & empathy |
|---|---|
| Target AI mode | Augment |
| Human oversight | Human decides |
| AI fit / Value / Feasibility / Data readiness | 3 / 4 / 4 / 3 (1–5) |
| Risk tier | Moderate |
| Roadmap wave | Wave 1 |
AI opportunities
- Walk a Customer Service Representative through diagnosis step by step from the symptoms, the product holding and similar resolved cases, citing the knowledge article behind each step
- Summarize a case's history, contacts and actions so whoever picks it up next sees where it stands without rereading every note
- Suggest the resolution that fixed matching cases, and what the customer needs to hear about it, for the representative to choose and explain
What stays human
- The Customer Service Representative decides the diagnosis and the resolution and owns what is said to the customer; AI suggests and drafts
- A refund, credit or billing adjustment a resolution needs goes to the Service Team Lead within delegated limits, prepared by the representative; AI never grants or denies one
Risks
- A plausible but wrong diagnosis followed because the suggestion looked authoritative
- Summaries that drop a fact the customer gave, so the next person asks again or acts on half the story
Ratings describe a typical large enterprise and are a starting point for your own assessment.
Use Cases and Agents in Each Package
Each AI Transformation Blueprint carries the method through to specific use cases, agent specifications and controls, every one mapped to the capabilities it acts on. The Business Capabilities Map includes the capability profiles, prioritization and readiness assessment without the use cases and agents.
| Package | L2 capabilities | Use cases | Agents | Controls |
|---|---|---|---|---|
| AI Management | 61 | 72 | 45 | 49 |
| Customer Service and Experience | 53 | 63 | 44 | 44 |
| Cybersecurity | 39 | 64 | 44 | 45 |
| Data and Analytics | 48 | 64 | 43 | 45 |
| Finance and Accounting | 58 | 60 | 41 | 41 |
| Governance, Risk and Compliance | 48 | 66 | 42 | 45 |
| Human Resources | 36 | 60 | 39 | 41 |
| Manufacturing Operations | 47 | 63 | 41 | 45 |
| Marketing | 53 | 64 | 43 | 45 |
| Research and Development | 40 | 60 | 43 | 46 |
| Sales | 36 | 64 | 43 | 45 |
| Supply Chain Management | 45 | 63 | 41 | 43 |
| Sustainability | 27 | 50 | 34 | 40 |
| Technology | 82 | 87 | 48 | 48 |
| Banking and Payments | 48 | 91 | 60 | 53 |
| Energy and Utilities | 44 | 92 | 70 | 61 |
| Enterprise | 112 | 119 | 79 | 80 |
| Healthcare and Life Sciences | 61 | 113 | 70 | 61 |
| Insurance | 49 | 91 | 60 | 55 |
| Investment and Wealth Management | 60 | 110 | 65 | 54 |
| Manufacturing Industry | 64 | 91 | 57 | 54 |
| Retail and Restaurants | 48 | 91 | 59 | 61 |
All AI-ready capability models, with prices · Read the sample from an AI assistant
Questions
Should I map AI use cases to capabilities or to processes?
Map them to capabilities first. A capability stays stable when processes, systems and the organization chart change, so the plan doesn't need redoing after every reorganization. Processes then show how a use case changes the work inside the capability.
What level of the capability map should AI use cases sit at?
Profile at L2 for most organizations, and attach individual use cases at L2 or L3, wherever the work they change is specific enough to name.
How do I decide which capabilities to prioritize for AI?
Rank capabilities by a risk-adjusted score: weight value, feasibility, data readiness and risk to fit your strategy. Then sequence by risk tier: lower-risk, readier capabilities go in the first wave, while the governance the higher tiers need is set up.
Can my AI assistant read these profiles directly?
Yes. Capstera's MCP endpoint serves the public sample to any assistant without a key, and a license key opens the full package, one record per call.