~3 min
Typical self-serve setup
Train URL → configure agent → embed snippet (operator timing, not a lab claim).
AI customer support case studies and results
Documented FoundChat pilots with baselines, interventions, metrics, escalation rules, and what would falsify the result. Built for founders evaluating AI support outcomes they can reproduce.
Methodology · as of July 2026
What we measure and what we refuse to claim
In scope
- In-scope ticket or chat deflection for a named intent cluster (not all support)
- Website first-response time for in-scope questions
- Unanswered-question rate after weekly transcript review
- Escalation quality for refunds, disputes, and VIP paths
Not claimed
- Guaranteed deflection percentages for every business
- Chat volume as a success metric
- 100% automation or contact-center replacement
- Results without clean docs and clear escalation rules
Transparent product metrics
Public pricing and setup timing, plus ROI calculator defaults you can stress-test before presenting to finance.
From $9/mo
Starter plan floor
Public pricing; credits scale with AI message volume not seats.
94%
ROI model default
Default automation rate in the ROI calculator stress-test lower for finance reviews.
Case studies
Each study follows the same playbook: baseline → intervention → metrics → escalation → falsify.
SaaS case study
From ~200 weekly Tier-1 tickets to ~40 in-scope
Baseline
A lean B2B SaaS team (~founder-led support) faced repetitive billing and setup questions that consumed inbox time before product work.
Intervention
Trained a FoundChat agent on cleaned help docs and pricing pages. Week one scoped only billing FAQs and setup how-tos. Refunds and account disputes stayed human. Weekly transcript review rewrote conflicting articles.
-
~200 → ~40
Weekly in-scope tickets
-
4 weeks
Pilot window
-
Billing + setup only
Scope
Escalation
Money movement, legal, and VIP account changes always escalate.
What would falsify this
If unanswered-question rate stays flat after two review cycles, sources not the model are the bottleneck. Expand intents only after the log shrinks.
SaaS case study
Website first response cut ~80% for product FAQs
Baseline
Common product questions waited hours in a shared inbox while the founder shipped. Marketing-site visitors bounced before a human replied.
Intervention
Embedded FoundChat on pricing and docs first. Success measured as website first response and ticket deflection not chat vanity counts. Account-specific issues stayed in the existing inbox.
-
~80% faster
First response (in-scope)
-
Website + help widget
Channels
-
Founder-led + AI
Team model
Escalation
Billing disputes and custom contract questions go to humans.
What would falsify this
If CSAT on escalations drops, tighten handoff copy and intent boundaries before adding pages.
Ecommerce case study
Shipping & returns policy tickets compressed
Baseline
A storefront received repetitive shipping and returns questions already answered on policy pages, while damaged-order claims needed humans.
Intervention
Cleaned conflicting policy pages, trained the agent on those sources only, and escalated claims/goodwill refunds. Measured policy-ticket volume not total chat opens.
-
Shipping + returns
Intent focus
-
Human-only
Claims
-
Storefront widget
Channel
Escalation
Damaged orders, chargebacks, and goodwill refunds stay human.
What would falsify this
If the agent contradicts itself, pause expansion and reconcile policy pages before retraining.
Related reading