Build vs. buy and vendor selection framework

Use this before the vendor intake checklist, not instead of it. That checklist assumes you’ve already picked a vendor and are getting ready to sign. This framework is for the stage before that: deciding whether to build at all, and if not, which vendor to run through intake.

Not legal advice. Pair with Legal and Procurement, especially on contract terms and data ownership.

Step 1: Build vs. buy

Most HR AI use cases should be bought, not built. The exceptions are narrower than most HR Tech teams assume.

Factor Favors build Favors buy
Uniqueness of the need No vendor solves this well; it’s specific to your org’s process or data A mature vendor category already exists (ATS scoring, HR chatbots, comp benchmarking)
Internal AI/ML capability You have engineers who can own model development and maintenance long-term HR Tech is 1-3 people with no dedicated ML capacity
Time to value You can tolerate 6+ months to a working v1 You need something live in weeks, not quarters
Data sensitivity Keeping data in-house is a hard requirement (rare, but real for some comp and health-adjacent use cases) Standard vendor DPA and security review is sufficient
Differentiation value This is a genuine competitive differentiator for how you do HR This is table-stakes HR infrastructure everyone needs
Total cost of ownership You’ve actually costed engineering time, monitoring, retraining, and governance overhead, not just the initial build Vendor pricing is transparent and predictable at your scale
Maintenance burden You have a named owner for the model’s entire lifecycle, not just the launch You’d rather pay for someone else to own model drift, retraining, and support

Default assumption: build only if you scored “favors build” on uniqueness of need AND internal capability. Everything else is a secondary consideration. Teams that build because of data sensitivity or differentiation ambition alone, without the capability to back it up, are the most common source of abandoned internal AI tools.

A note on the notebooks in this playbook: the attrition risk model and HR Q&A agent demos are working sketches meant to show you what a well-governed build looks like, not a recommendation to build these specific things in-house. Most orgs are better served buying an attrition or Q&A product and applying this playbook’s governance to it.

Step 2: Vendor comparison scorecard

If you’re buying, compare 2-4 real vendors before you pick one to run through the intake checklist. Score each 1-5.

Criterion Weight What to evaluate
Capability fit 25% Does it solve your actual use case, not an adjacent one the sales deck implied?
Governance readiness 25% Can they answer the vendor intake checklist categories today, or will you be waiting months for documentation?
Security and data posture 20% SOC 2, data residency options, DPA terms, sub-processor list
Integration complexity 10% Does it work with your actual HRIS/ATS/LMS, or does the demo only show a clean sandbox?
Pricing transparency and scalability 10% Is pricing predictable as you scale usage, or does it have hidden per-seat/per-call cliffs?
References in an HR context specifically 10% Not general enterprise AI references, HR-specific customers you can actually call

Weighted score = sum of (criterion score × weight). Score every finalist independently before comparing, same anti-anchoring principle as the use case prioritization matrix.

Red flags at the evaluation stage

Catch these before you’ve invested in a full intake process with a vendor who was never going to pass it:

Any of these doesn’t automatically disqualify a vendor, but it means slow down and verify before starting the formal intake.