eBook · Human-Centric AI Council
AI Solution
Decision Framework
Introduction
Start with the problem,
not the product.
Every HR leader today feels the pressure to "do something with AI." Vendors pitch innovation; executives demand speed; employees expect smarter systems. But AI readiness isn't about adopting technology — it's about understanding what work needs to change and how that work should be enabled.
Too often, organizations start from the solution side: "What kind of AI should we use?" The better question is: "What kind of work problem are we solving, and what form of AI readiness do we actually have?"
"The costliest mistakes usually happen before implementation — when teams choose the wrong approach to acquire, design, and sustain the capability."
— Kyle Lagunas, Founder Kyle & Co · Executive Director, HCAICThis framework introduces a decision lens — Build, Buy, Borrow, or Bot — grounded in a five-level AI maturity model. By connecting the type of AI with the strategic intent, business problem, and organizational readiness, HR leaders can make disciplined, value-driven choices without over-engineering solutions or getting swept up in hype.
Chapter 1
Build, Buy, Borrow, or Bot —
mapped to AI maturity.
Not every AI solution demands the same approach. The right path depends on the type of AI capability you're acquiring. Select a maturity level to see the decision guidance.
If the process is unique to your HR workflows, consider using internal automation (RPA or scripting). Only when no vendor solution fits your proprietary process.
For standardized processes (e.g., onboarding checklists, document approvals), use existing SaaS tools with proven reliability. Faster implementation; vendor support included.
For small-scale needs, leverage shared services or BPO partners with built-in automation capabilities. Low-risk way to validate fit and adoption before committing.
Low ROI — building internally often costs more than the savings realized. Use bots for quick, contained automation where speed matters more than customization.
When your organization possesses clean, proprietary data that offers strategic differentiation — and the use case is unique enough that no vendor solution fits.
When use cases are generic (e.g., industry-standard writing assistance) or when your data maturity is low. Most organizations should start here.
Consider academic or vendor partnerships for co-development of context-specific assistants. Good for piloting before committing to procurement.
Automate model retraining and data refreshes rather than model creation. Use workflow bots to surface AI suggestions within existing tools your team already uses.
Only viable for enterprises with high data maturity, a dedicated ML team, and use cases where proprietary data creates a meaningful competitive advantage.
For standard predictive models (turnover risk, engagement scoring), proven SaaS vendors offer faster time-to-value. Evaluate model transparency carefully.
Partner with research institutions or analytics consultancies to co-develop models tailored to your workforce context. Ideal when internal capability is developing.
Automate data pipelines and model refresh cycles rather than the modeling itself. Bots can schedule reporting and flag anomalies without replacing analytical judgment.
Only appropriate for enterprises with mature data infrastructure, explainability requirements, and the capacity to monitor behavioral models continuously. High risk.
Most organizations should buy adaptive systems from vendors with proven personalization engines. Rigorously evaluate bias controls and override mechanisms.
Consider piloting with a vendor's sandbox or academic research partner before enterprise commitment. Adaptive AI is high-stakes — pilot scope and fallback plans matter.
Automate model monitoring, drift detection, and escalation alerts. Human oversight is non-negotiable at this level — bots assist governance, not replace it.
Reserved for organizations where AI is core to competitive strategy and where internal teams can responsibly govern systems that redefine roles and structures. Rare.
Partner with platforms offering catalytic AI — but subject them to rigorous ethics, governance, and executive oversight. Demand explainability and exit clauses.
Co-develop with trusted research or advisory partners who specialize in responsible AI at the catalytic level. Don't go it alone at this maturity stage.
Automate governance workflows — policy triggers, audit logging, ethics board alerts. Catalytic AI demands more human oversight, not less, as automation increases.
Chapter 2
Four things to know
before you begin.
Before evaluating any AI solution, HR teams must assess readiness across four core dimensions. If you can't answer these clearly, you're not ready to evaluate — you're shopping.
| Dimension | Key Question | Why It Matters |
|---|---|---|
| Business Problem Definition | What is the work we're trying to improve, augment, or transform? | If you can't articulate the problem, you can't design the right intelligence. |
| Data & Infrastructure | Do we have the data quality, volume, and integration required to train or enable AI? | Poor data is the single biggest constraint on AI success. |
| Governance & Ethics | Can we oversee, explain, and validate the AI's logic and outcomes? | Ensures compliance, trust, and accountability across the organization. |
| Ownership Strategy | How should we source and sustain this capability over time? | Determines ROI, agility, and long-term resilience of your AI investments. |
Chapter 3
Market scan guidelines
for HR teams.
A disciplined market scan isn't about collecting vendor names — it's about uncovering what truly works for your context. The goal is not to find more tools; it's to identify the right-fit solution that delivers measurable value without creating redundancy or integration chaos.
Check what you already have
Audit your current technology stack before scanning the market. Review contracts, product roadmaps, and release notes from current vendors. Check marketplaces (Workday Extend, ServiceNow Store, SuccessFactors Store) for available add-ons or AI copilots. Map your stack by function to identify overlaps and white spaces.
You may already own 80% of what you need — only the last 20% may require new spend.
Define the "why" before you search
Write a clear, shareable problem statement. Avoid starting with "We need an AI tool for..." Frame it as: "We need to..." (e.g., improve turnover prediction accuracy, personalize learning, automate compliance reporting). Define success metrics, clarify non-negotiables, and set a realistic timeline including pilot and full rollout.
If you can't clearly articulate the problem, you're not ready to evaluate solutions.
Build your AI solution tracker
Turn your market scan into a living knowledge base. Log discoveries, contacts, pricing, and lessons learned. Add fields for AI type (1–5), maturity score, and ethical/governance notes. Share findings across HR, IT, and Procurement to avoid duplicate evaluations and siloed decisions.
Establish your evaluation criteria
For each solution evaluated, apply consistent criteria: Business Fit (does this solve our specific problem?), AI Depth (what maturity level does it truly demonstrate?), Integration (can it connect with our existing systems?), ROI & Cost (what's the expected return vs. total cost of ownership?), and Governance (how does the vendor handle explainability, bias, and data privacy?).
Chapter 4
Solution assessment
by maturity level.
When evaluating a specific AI solution, score it across four dimensions against the maturity level it claims. Use this to separate genuine capability from vendor hype.
Dimension 1
Task-to-Technology Fit
Does the AI type match the nature of the task? Rule-based tasks need deterministic logic, not adaptive models. Ambiguous creative challenges require catalytic AI, not bots.
Dimension 2
Data Quality & Access
Does your data meet the requirements for the claimed maturity level? Predictive AI needs high-volume historical data. Adaptive AI needs real-time behavioral signals. Verify before buying.
Dimension 3
Explainability & Bias Controls
Can the vendor explain how the AI reached its conclusion? Are there fairness audits, auditability layers, and human escalation paths? Non-negotiable at Type 3 and above.
Dimension 4
Human Impact Assessment
How does this AI change roles, workflows, and employee experience? Map the downstream effects before deployment — especially for adaptive or catalytic systems that reshape how work is done.
Chapter 5
Measuring what
actually matters.
Track AI performance across four categories. Operational metrics tell you if it's being used. Accuracy metrics tell you if it's working. Trust metrics tell you if people believe in it. Business impact tells you if it's worth it.
Operational Efficiency & Adoption
- Adoption Rate — % of intended users actively engaging
- Task Completion Time — AI-assisted vs. manual baseline
- Frequency of Use — daily/weekly active sessions
- Cost per Task — total cost of AI use / tasks completed
- Time-to-Deploy — days from contract signature to go-live
Accuracy, Quality & Model Performance
- Prediction Accuracy — correctness of AI classifications
- Precision & Recall — true positives vs. false positives
- Bias/Equity Metrics — accuracy across demographic groups
- Model Drift — accuracy variance between baseline and current
- Hallucination/Error Rate — incorrect outputs per total outputs
User Experience & Trust
- User Satisfaction Score — CSAT or NPS survey results
- Trust Score — "I trust this AI's recommendations" (1–5 scale)
- Time-to-Insight — query to usable insight
- Explainability — % of users who understand why AI decided what it did
- Decision Accuracy — % of AI suggestions accepted by users
Business Impact & ROI
- Return on Investment — (Total Benefits − Total Costs) / Total Costs
- Cost Savings — baseline spend vs. AI-enabled spend
- Quality of Outcomes — accuracy of hires, engagement lift, retention improvement
- Strategic Alignment — % of AI use cases tied to stated HR priorities
- Risk Reduction — compliance incidents, bias flags, governance escalations
Conclusion
Progress that is
responsible and repeatable.
AI has crept into everything HR touches — platform roadmaps, process design, data strategy. HR leaders aren't just shopping for "AI tools" anymore. You're making architectural decisions about what you'll own, what you'll outsource, how you'll govern it, and how work will run going forward.
The Build, Buy, Borrow, or Bot framework isn't a checklist. It's a shared language — one that lets HR, IT, Finance, and Procurement have the same conversation about the same problem at the same level of rigor. Use it as a reality check so progress is responsible, measurable, and repeatable.
The question is never which AI tool is best. The question is: which approach is right for this problem, this organization, at this level of readiness — and how will we know if it's working?
Want the rest
of the toolkit?
This framework is part of the HCAIC Cohort Zero toolkit — four practitioner-built resources covering AI literacy, responsible AI governance, change management, and solution decision-making. Free to access with a one-time signup.
About Kyle & Co
Kyle & Co is a modern research and advisory firm helping HR and talent technology leaders make smarter decisions, faster. The team behind the Human-Centric AI Council, the Transformation Realness podcast, and industry-shaping reports on workforce strategy and responsible AI.
About the Author
Lydia Wu is the lead contributor on this framework, bringing practitioner expertise in AI solution strategy and HR technology evaluation. This guide draws on her experience alongside several members of the Human-Centric AI Council.