| name | 14-ai-integration |
| description | Use when producing or reviewing the 14 ai integration component of a business plan; applies its specialist evidence, decisions, and acceptance tests instead of neighbouring pipeline skills. |
| metadata | {"portable":true,"compatible_with":["claude-code","codex"]} |
AI Integration & Efficiency Skill
Overview
Generate Section 14 of the business plan: the AI integration and efficiency section. Use this skill to show where AI materially improves economics, speed, quality, or scalability, and where human control must remain.
Use meta-digital-transformation first when the broader question is how the business should digitise, redesign its model, or prioritise systems investment. This skill should then specify the AI layer inside that wider digital strategy.
Use When
- Use when drafting or revising the AI integration section of a modern business plan.
- Use when the business wants to show credible automation, augmentation, and process redesign opportunities.
- Use when investors or operators need to see the commercial logic, governance, and limits of AI adoption.
Do Not Use When
- Do not use to bolt trend language onto a business that has no realistic AI use case.
- Do not recommend automation before understanding the underlying process and economics.
- Do not imply that AI removes the need for ownership, controls, or customer trust.
Generate a practical AI utilisation plan that demonstrates the business is AI-smart a competitive requirement in 2026.
Why This Section Exists
Investors in 2026 expect businesses to articulate their AI strategy. A business plan without an AI section signals either ignorance or inefficiency. This is not about being trendy it is about demonstrating operational intelligence.
What to Generate
Required Elements
- AI readiness assessment Current state of AI adoption
- Workflow automation inventory Processes that AI can handle or augment
- AI tool stack Specific tools, platforms, and APIs the business will use
- Cost-benefit analysis AI investment vs. labour savings and efficiency gains
- Implementation roadmap Phased AI adoption plan
- Team AI competency Training needs and hiring for AI skills
- Data strategy What data the business collects and how AI uses it
- AI governance and ethics Responsible AI use policies
- Competitive AI advantage How AI creates differentiation
- Risk and limitations What AI cannot do and human oversight requirements
- Process-redesign-first logic What is removed, simplified, or accelerated before automation
- CX impact How AI improves the end-to-end customer and employee experience
Workflow Automation Inventory Template
| Process | Current Method | AI Solution | Tool/Platform | Time Saved | Cost Impact | Priority |
|---|
| [Process] | Manual/Semi | [AI approach] | [Tool name] | X hrs/week | -$X/month | High/Med/Low |
AI Tool Categories to Evaluate
- Customer service Chatbots, ticket routing, sentiment analysis
- Marketing Content generation, ad optimisation, personalisation
- Sales Lead scoring, CRM automation, proposal generation
- Operations Inventory forecasting, scheduling, quality inspection
- Finance Invoice processing, expense categorisation, fraud detection
- HR Resume screening, onboarding automation, performance analytics
- Product Feature prioritisation, user behaviour analysis, testing
- Development Code generation, testing, documentation (Claude Code, etc.)
Cost-Benefit Analysis Format
AI Investment:
- Tool subscriptions: $X/month
- Implementation cost: $X (one-time)
- Training cost: $X
- Total Year 1: $X
Savings/Gains:
- Labour hours saved: X hrs/month @ $Y/hr = $Z/month
- Error reduction: X% fewer errors = $Y saved
- Speed improvement: X% faster processing
- Revenue impact: [Additional revenue enabled by AI]
ROI: [Savings - Investment] / Investment x 100 = X%
Payback period: X months
AI Governance Framework
- Data privacy What data AI accesses and how it is protected
- Bias monitoring How outputs are checked for fairness
- Human oversight Which decisions require human approval
- Transparency How AI decisions are explained to stakeholders
- Compliance Relevant AI regulations (EU AI Act, local laws)
Process Automation Readiness
Before automating any process, assess readiness using the BPM principle: redesign first, automate second "don't pave the cow path" (Dumas et al., 2013).
Automation decision criteria automate when a process scores high on:
- Volume (high repetitions), Standardisation (low variability), Rule-based (codifiable decisions), Data availability (digital inputs), Error impact (costly mistakes), Speed requirement (fast turnaround needed)
Leave manual when: complex judgement required, low volume, high variability, or human touch adds customer value.
Automation maturity levels:
| Level | Description | Example |
|---|
| Manual | Paper-based, no IT | Handwritten records |
| IT-supported | Standalone apps | Spreadsheets, email |
| IT-enabled | Integrated systems, manual coordination | ERP, CRM |
| BPMS-managed | Automated flow, some manual tasks | Workflow routing |
| Fully automated | Minimal human involvement | Straight-through processing |
Build vs Buy vs Outsource
For each technology component, evaluate (Dennis et al., 2021):
| Strategy | Best when | Risk |
|---|
| Custom build | Unique need, in-house skills, time flexible | High (long timeline) |
| Packaged software | Common need, short timeframe | Medium (vendor dependency) |
| Outsource | Non-core function, skills gap | Medium (loss of control) |
The Algorithm Principle: Automate Last
Use McNeill's five-step order before recommending any AI workflow:
- Question every requirement
- Delete every possible step
- Simplify and optimise
- Accelerate cycle time
- Automate last
If the plan jumps straight to automation, it is weak. The section must show what process friction was removed before AI was introduced.
AI Redesign Checklist
For each proposed AI use case, state:
- which requirement or approval was challenged
- which steps were removed
- what was standardised
- what delay or handoff was shortened
- why AI is now appropriate
Customer-Experience-Led AI
AI should improve the full customer experience, not just reduce internal labour.
Map each AI use case to one or more of these moments:
- discovery and enquiry handling
- sales conversion or proposal turnaround
- onboarding and first value
- service quality and response time
- retention, repeat purchase, or referral
If AI lowers cost but degrades trust, clarity, or service quality, reject or redesign the use case.
AI Literacy and Adoption
Every AI roadmap must include:
- executive sponsor
- process owner
- user training and AI literacy
- pilot period with success criteria
- human-review boundaries
- feedback loop to refine, expand, or stop the use case
Do not treat tool subscriptions as transformation.
Generation Process
- Ask for: industry, current tech stack, team size, pain points, budget constraints
- Audit all business processes for AI automation potential
- Prioritise by impact and implementation difficulty
- Select specific AI tools and platforms
- Calculate cost-benefit for top 5 automation opportunities
- Build phased implementation roadmap aligned with section 13
- Define governance policies and human oversight requirements
- Apply the five-step algorithm to each priority use case before recommending automation
- Show how AI improves the end-to-end customer experience, not just internal throughput
- Define literacy, pilot, adoption, and measurement requirements for each major use case
Quality Criteria
- Automation opportunities are specific to this business, not generic AI hype
- Tool recommendations are current and available (not vapourware)
- Cost-benefit analysis uses realistic numbers
- Implementation is phased not "AI everything on day one"
- Governance section addresses data privacy and bias
- Human oversight is maintained for critical decisions
- Process redesign happens before automation
- AI use cases are tied to customer or employee pain points, not generic hype
- Adoption plan includes literacy, ownership, and pilot metrics
- ROI is measured using operational or commercial outcomes, not novelty language
AI-on-SaaS Cross-Section Integration (when AI is customer-facing in a SaaS plan)
When AI is a customer-facing feature in a SaaS / ICT business plan (not just internal efficiency), this Section 14 is the cross-section integrator. The deep AI-on-SaaS skills sit in their natural homes:
- Section 03:
skills/pipeline/03-products-services/saas-ai-product-strategy-and-roadmap/ — AI product strategy (build/buy/host/orchestrate; model router; eval-driven dev; AI roadmap by ARR with cost gating)
- Section 04:
skills/pipeline/04-market-analysis/saas-ai-market-and-tam/ — AI-aware TAM with attribution discipline
- Section 06:
skills/pipeline/06-competitive-analysis/saas-ai-moat-and-defensibility/ — 7-question moat test + false-moat catalogue + Wardley placement + foundation-model platform-risk
- Section 07:
skills/pipeline/07-marketing-sales-strategy/saas-ai-pricing-strategy/ — AI pricing architecture (tier × model × allowance × overage × FX)
- Section 09:
skills/pipeline/09-management-team/saas-ai-talent-strategy/ — AI talent (roles by ARR, African talent map, retention)
- Section 10:
skills/pipeline/10-financial-projections/saas-ai-unit-economics-and-cogs/ — AI COGS waterfall + AI GM trajectory + AI-cost-%-of-ARR
- Section 10:
skills/pipeline/10-financial-projections/saas-ai-cost-of-tenant-calculator/ — per-tenant cost calculator with sensitivity matrix
- Section 11:
skills/pipeline/11-funding-request/saas-ai-funding-stage-playbook/ — stage ladder + investor archetype + grant + commercial blend
- Section 11b:
skills/pipeline/11b-grant-proposal/saas-ai-for-good-grant-proposal/ — AI-for-good grant proposal
- Section 12:
skills/pipeline/12-risk-analysis/saas-ai-risk-and-stress-test/ — 14-category AI risk register + 6 quantified stress scenarios
- Section 16:
skills/pipeline/16-sustainability-strategy/saas-ai-sustainability-and-ethics/ — AI ethics + sustainability + governance committee + AI-incident protocol
- Meta:
skills/meta-finance/meta-ai-bankability-and-investor-readiness/ — AI bankability scorecard
- Meta:
skills/meta-finance/meta-ai-valuation-adjustments/ — AI premium / discount logic
- Meta:
skills/meta-strategy/meta-living-plan-governance/SKILL.md — AI cadence (eval weekly, hallucination monthly, cost-per-tenant monthly, retraining trigger, model deprecation watch)
- Meta: — AI section of quarterly board pack
This section (14) remains the operational AI integration layer (where AI is used inside the business for efficiency / augmentation) and the cross-section anchor that ensures the AI-on-SaaS plan reconciles across all sections.
References
- Process automation readiness: See
references/process-automation-readiness.md for automation levels spectrum, automation decision framework (6 criteria), automation readiness checklist, BPMS components and architecture, task types in process automation, decision tables, 9 steps to executable processes, build/buy/outsource decision framework, and the "redesign before automating" principle from Dumas et al. (Springer, 2013) and Dennis, Wixom & Tegarden (Wiley, 2021)
- AI Economics Framework: See
references/ai-economics-framework.md for the core economics of AI adoption the prediction-as-cheap-input thesis, the seven-component anatomy of a decision (input data, training data, prediction, judgment, action, outcome, feedback), what becomes more valuable as prediction gets cheap (judgment, data, action capacity), the workflow redesign principle (reengineering before automating), the AI Canvas tool, tool vs. transformation modes of adoption, stakes-based risk framework (recoverable vs. catastrophic errors), the new human-machine division of labour (known knowns/unknowns taxonomy), first-mover vs. fast-follower timing strategy, Uganda/East Africa-specific AI tools table, and a full AI Strategy Checklist for business plan writers from Agrawal, Gans & Goldfarb, Prediction Machines (Harvard Business Review Press, 2022, updated edition); includes worked examples for customer service, inventory, marketing, finance, and quality control in the Uganda context
- Uganda ICT IP Guidelines: See
references/uganda-ict-ip-guidelines.md for the complete IP framework for technology and digital businesses in Uganda covering copyright duration (50 years for software), trademark registration (renewable in 10-year blocks), patent and utility model terms, trade secret protection for algorithms, data protection obligations under the Data Protection and Privacy Act 2019, open source licensing strategy, source code escrow, IP ownership in commissioned software contracts, URSB registration process, international protection via WIPO/ARIPO/Berne Convention, MoICT&NG's roles (IP guidance, incentivisation, commercialisation, monitoring, landscape analysis, capacity building), revenue sharing structure (25% to government for MoICT&NG-funded innovations), and NDA/assignment/revenue-sharing agreement templates from MoICT&NG, Uganda ICT IP Guidelines Version 1, January 2025
- Digital innovation strategy: See
references/digital-innovation-strategy.md for the 6D cyclical strategy model (Discover/Detect/Diagnose/Direct/Deliver/Dexterity), Strategy Nucleus (five-layer purpose architecture), Three Horizons multimodalism (H1 exploitative/H2 extending/H3 exploratory simultaneous not sequential), Doblin's 10 Types of Innovation (configuration/offering/experience domains), Blue Ocean Strategy five-step process, VRIN analysis framework, Design Thinking five steps, Transform Loop (Scan/Prioritise/Learn/Experiment/Plan/Build), Transformation Backlog priority matrix, Industry 4.0 maturity ladder (digitisation digitalisation digital transformation), and Uganda/EA application including East African AI ecosystem (Farmbetter, Apollo, JUMO, M-KOPA) from Digital Business Strategy (2024) and Palfreyman (2020).
Required Inputs
| Input artefact | Source/provider | Required | Behaviour when absent |
|---|
| Prioritised use cases, process baseline, data rights, architecture, vendor options, safety controls, capability evidence, and economics for 14 ai integration | Process owners, technical lead, legal/privacy review, vendors, and approved model | Yes | If absent, data rights, baseline performance, evaluation evidence, or human fallback is unavailable, hold the use case at discovery and return the control or test needed. |
| Finalised business brief, target reader, country, and stage | Client intake and engagement owner | Yes | Stop section decisions and route the missing context to client intake. |
| Reconciled upstream assumptions that this section consumes | Named pipeline owners | Conditional | Record the dependency, affected claim, owner, and recovery step; do not substitute an invented value. |
Outputs
| Artefact | Consumer | Observable acceptance condition |
|---|
| Staged AI or agent integration plan with use-case gates, controls, costs, and fallback operations | Plan author and target decision-maker | The artefact answers the section decision and traces each material conclusion to the supplied evidence. |
| 14 ai integration exception and handoff note | Downstream section owners | Every blocked or conditional item names its consequence, owner, evidence request, and restart condition. |
| 14 ai integration release record | Reviewer or plan assembler | Records the checks completed, failures, unassessed items, professional review required, and release state. |
Evidence Produced
| Evidence | Format | Acceptance condition |
|---|
| Use-case readiness score, build/buy decision, data/control review, cost model, evaluation plan, and rollback path | Source-linked table, calculation, or annotated prose | The evidence is reproducible from named inputs and distinguishes verified fact, management assumption, and inference. |
| 14 ai integration decision record | Decision note | States the selected action, rejected credible alternative, countercase, rationale, and risk accepted or avoided. |
| 14 ai integration review trace | Gate entry | Identifies the date, input versions, reviewer role, failed checks, recovery owner, and any check that remains not assessed. |
Capability Contract
For 14 ai integration, the controlling focus is process baseline, AI use-case value, data readiness, model or vendor choice, evaluation, human control, economics, and rollback. This skill may analyse and prototype within authorised data and tools; it may not expose client data, purchase services, deploy to production, grant autonomous permissions, or claim model capability without testing. Its normal mode is read-only analysis and drafting. Any mutation, external communication, spending, certification, or professional conclusion outside that boundary requires explicit authority and must remain traceable to the approving role.
Degraded Mode
For 14 ai integration, loss of evidence about 14 ai integration evidence, decisions, failure thresholds, ownership, and downstream handoffs activates degraded mode. If the controlling 14 ai integration evidence is unavailable, the same boundary applies. When data rights, baseline performance, evaluation evidence, or human fallback is unavailable, hold the use case at discovery and return the control or test needed. Return the verified subset, label the affected decision qualified or not assessed, explain the downstream consequence, and state the smallest evidence request or authorised action that permits recovery. Do not convert the missing check into a pass.
Decision Rules
| Choice or condition | Action | Failure or risk avoided |
|---|
| For 14 ai integration, automation value is plausible but error cost or action authority exceeds the tested control boundary | reduce autonomy, add human approval, narrow scope, or reject the use case | AI theatre or unsafe automation adds cost and operational liability without customer value |
| For 14 ai integration, A current legal, regulatory, tax, accounting, market, or platform claim controls the 14 ai integration decision | Verify the controlling source, effective date, jurisdiction, and reviewer status before release | Stale external facts become permanent plan assumptions |
| For 14 ai integration, The evidence reconciles with neighbouring sections and the countercase does not overturn the choice | Complete staged ai or agent integration plan with use-case gates, controls, costs, and fallback operations, attach the evidence and release record, and hand off named dependencies | Premature release and repeated downstream rework |
Workflow
- Define the exact 14 ai integration decision, intended reader, jurisdiction, business stage, and permission boundary.
- Collect prioritised use cases, process baseline, data rights, architecture, vendor options, safety controls, capability evidence, and economics and map each material conclusion to its source; stop the affected conclusion when an input could change it.
- Apply the specialist methods and directly linked references already contained in this skill, retaining its domain thresholds, calculations, and Uganda or East Africa context where applicable.
- Compare the credible alternatives, test the countercase and failure path, and apply the decision table rather than selecting a template default.
- Produce staged ai or agent integration plan with use-case gates, controls, costs, and fallback operations with the evidence, exception, and handoff records; reconcile every shared assumption with its owning section.
- Run the section quality checks, applicable finance or professional review, and anti-slop gate. If a gate fails, correct the evidence or decision and return to the responsible step.
Quality Standards
- Staged AI or agent integration plan with use-case gates, controls, costs, and fallback operations must answer a real decision for the named bank, investor, DFI, grant, board, or strategic-partner reader.
- Use-case readiness score, build/buy decision, data/control review, cost model, evaluation plan, and rollback path must be source-linked, dated where facts can change, and sufficient for another reviewer to reproduce the conclusion.
- The section exposes its countercase, stop condition, recovery action, and effect on neighbouring sections.
- No unavailable source, calculation, tool, or professional review is reported as passed; finance and statutory judgements follow the governing doctrine.
- Language remains specific to 14 ai integration, uses British English naturally, and passes the repository anti-slop gate without promotional filler.
Anti-Patterns
- In 14 ai integration, treating an unavailable prioritised use cases, process baseline, data rights, architecture, vendor options, safety controls, capability evidence, and economics as confirmed. Correction: qualify the affected conclusion and issue the named evidence request.
- Producing staged ai or agent integration plan with use-case gates, controls, costs, and fallback operations that restates the brief but makes no choice. Correction: record the choice, rejected alternative, rationale, countercase, and implication.
- Ignoring a conflicting upstream assumption. Correction: return it to its owning section and resume only from a reconciled version.
- Reporting an unavailable check as passed. Correction: mark it not assessed and narrow the release state.
- Claiming compliance, assurance, bankability, or investor readiness from narrative quality. Correction: run the applicable gate and retain its evidence.
- Copying the worked example into a client plan. Correction: use the method only and replace every fact with verified engagement evidence.
Worked Example
A forecasting assistant may reduce manual spreadsheet work, but source data has missing weeks and no baseline forecast. Repair the dataset, compare against the existing method, and retain manual planning until accuracy and decision-value gates pass.
For a foundation investment or data-product case, load AI data foundation investment case.
