1. Vertical fit vs generic posture
A generic chatbot (Intercom Fin, Chatbase, Tidio Lyro) ships a single configuration that every customer adapts via prompt-engineering and a knowledge dump. A vertical-specialised chatbot (SLAtech AI Medical, ManyChat, Botpress) ships industry-tuned seed content, compliance posture and tone-of-voice presets out-of-the-box. Decision rule: if your industry has compliance constraints (medical, legal, education, payments), pick a vertical. If your traffic is mixed (coworking, generic SMB), pick a generic with graduation path.
2. Data residency
GDPR-residency is the single largest enterprise-procurement gate in Europe. US-hosted vendors (Intercom, Drift, Chatbase) require a supplementary transfer-impact assessment per Schrems II - bookable but time-consuming. SLAtech offers hosting in your own jurisdiction on request (Enterprise). Ask vendors three questions: (1) where is data hosted, (2) which sub-processors process my data and where, (3) what's your DPA's SCC version.
3. Compliance posture matrix
Match the vendor's compliance posture against your buyer-side regulatory exposure. HIPAA-covered entity → require BAA-eligible vendor with single-tenant option. EU enterprise → require GDPR-compliant DPA referencing SCC 2021/914. Payments-touching → confirm vendor doesn't store cardholder data (PCI scope reduction). Healthcare adjacent → PHI redactor at ingest is a material difference. SLAtech publishes a 14-framework compliance glossary at /en/compliance/ - request equivalent transparency from every shortlisted vendor.
4. Pricing models - flat vs per-conversation vs per-resolution
Flat-tier pricing (SLAtech, Tidio Lyro starter) lets you forecast cost; each tier includes a monthly allowance (SLAtech Pro: 5,000 resolutions), and growing past it means adding capacity or moving up a tier, not a per-resolution bill. Per-conversation pricing (older Drift tiers) penalises traffic growth. Per-resolution pricing (Intercom Fin) penalises successful AI deflection - every successful resolution is a billed event. Decision rule: model your expected conversation volume across a 12-month growth curve under each pricing model. Per-resolution favours low-traffic premium-vertical use cases; flat favours scale.
5. Channel coverage
Web widget is table-stakes. Telegram, WhatsApp Business (Cloud API), Instagram DM, email - each is a separate procurement decision. WhatsApp Business is a significant integration: Cloud API access requires Meta Business verification; vendors that ship native integration (SLAtech Pro tier, ManyChat) save weeks of vendor-side onboarding. Web-only vendors (Chatbase) are inappropriate if your buyers shop via messaging.
6. Evaluation methodology
A published quality score is only worth as much as the method behind it. Ask any vendor, including us: how many questions, who wrote them, who labelled the answers, what the comparison set was, and whether you can re-run it yourself. SLAtech does not publish scores today. What we do measure per answer, and what is still missing, is at /en/eval/.
7. Vendor longevity and concentration risk
AI chatbot vendor consolidation is accelerating. Buyer-side de-risking patterns: (a) avoid sole-channel-bet vendors (web-only chatbots are exposed if buyers shift to messaging), (b) avoid vendors with single LLM-provider exposure (you don't want a single LLM provider's rate-limits cascading to buyer-facing outage), (c) check vendor revenue runway if public. Procurement-side mitigation: contract export-data right, ensure portable knowledge format (Markdown / JSON), avoid proprietary lock-in formats.
8. Implementation timeline
Self-service onboarding (SLAtech, Tidio, Chatbase): 10-30 minutes to first widget live. Assisted onboarding (Intercom, Drift): 2-6 weeks with a customer-success manager. Implementation consulting (Ada, larger Drift): 6-12 weeks. Decision rule: if your team is ≤50 people, prefer self-service; the consulting-overhead value drops below the time cost. ≥500 people, prefer assisted (compliance approvals and cross-team handoffs benefit from a dedicated CSM).
9. Total cost of ownership
Beyond seat license / per-conversation cost: (a) implementation effort (consultant fees or internal eng hours), (b) integration cost (CRM webhooks, calendar sync - bundled vs add-on), (c) knowledge-management labour (uploading, refreshing FAQ content), (d) procurement / legal review (DPA negotiation, BAA execution). Aggregate the multi-year TCO before choosing - sticker price is rarely the differentiator.
10. Migration path and vendor lock-in
Knowledge format portability is critical. Vendors with proprietary content formats (Botpress' visual flows) impose a high migration cost when switching. SLAtech and Chatbase use Markdown + JSON for knowledge content - portable across vendors. Conversation history should be exportable as JSONL. Lead-capture webhook configuration should be vendor-agnostic (your CRM connection survives the vendor switch). Ask every shortlisted vendor: how do I export my data on day 1, day 30, day 365.
11. Procurement checklist
Twelve questions for every vendor RFP: (1) hosting region, (2) sub-processor list, (3) DPA SCC version, (4) BAA availability, (5) SOC 2 / ISO 27001 status, (6) PHI redaction policy, (7) eval methodology + score, (8) channel coverage, (9) pricing model + projected 12-month spend, (10) implementation timeline, (11) export-data right, (12) vendor longevity / financial transparency.
12. Decision matrix template
Score each shortlisted vendor across the 11 criteria above (1-5 scale, weighted by your priorities). Typical weights: vertical fit 3×, compliance 3×, pricing 2×, channels 2×, all others 1×. Total weighted score = the procurement-defensible recommendation. Document the weighting before scoring to avoid bias toward incumbent vendor. SLAtech ships a downloadable matrix template at /en/buyer-guide/matrix.csv - adapt the weights to your environment.
