METHODOLOGY · SKILLS VS FAQ AUTOMATION
D3x publishes a 60–70% autonomous resolution range on guest messaging. This page defines exactly what that means, so operators, CIOs, and procurement teams can compare vendors on the same scorecard.
Public definition · Production messaging · Updated 2026
01 · THE SCORECARD
Vendors often count a knowledge-base reply as “automated.” The guest still needs towels, a folio correction, a reservation change, or a maintenance fix, and that unfinished work lands in the operator inbox. D3x Skills execute in your PMS and ops stack. That is the difference between deflecting and resolving.
Answers
Guest got text. Hotel still has work.
Resolves
Guest got the outcome. Hotel stack updated.
02 · DEFINITION
In plain language: the guest asked for something, the agent finished it (or correctly declined under policy), and no teammate had to take over the thread. A polite FAQ answer that leaves the operational request unfinished is not autonomous resolution.
Formula
Autonomous resolution = numerator ÷ denominator
Numerator
Guest messaging threads closed without human takeover, where the agent completed the requested outcome (or a policy-safe refusal) under Skills Engine rules.
Denominator
All guest messaging threads in scope for automation in the measured period (by property and channel), excluding spam, test traffic, and threads intentionally held out of automation.
Human-in-the-loop approval
If policy requires a human approval before an action commits, the thread is not autonomous until the action completes without further staff ownership of the guest conversation. Approvals are governance; they are not free automation credits.
Co-pilot / draft-for-staff
Drafts that a human must send do not count as autonomous resolution. The human still owns the outcome.
Multi-intent threads
If one thread asks for towels and a folio correction, both outcomes must be completed (or correctly refused) without takeover. Partial completion with staff finishing the rest does not count as autonomous.
Warm transfer
A clean handoff with full context is good product behaviour. It is not autonomous resolution for that thread.
03 · WORKED EXAMPLE
In-stay guest on WhatsApp: “Can I have late checkout until 2pm? I have a train at 3.” Property policy allows late checkout for €35 on weekdays when occupancy allows.
Path A
Counted as “automated” by many vendors. Work still open.
Path B
Counts as autonomous resolution. Inbox empty for this request.
04 · THE 60–70% RANGE
Portfolio typicals on guest messaging run 60–70% autonomous resolution once Skills are enabled and after-hours automation is live. The range is not a guarantee for every property on day one. It reflects production portfolios after rollout, with high-stakes flows kept in co-pilot or human-in-the-loop where brand policy requires it.
60–70%
#Portfolio typical · messaging
Across production guest-messaging portfolios on D3x, autonomous resolution typically runs in a 60–70% range once Skills and after-hours automation are live — a portfolio typical, not a day-one SLA.
Definition →250K+
#Production volume across live groups
D3x processes 250,000+ guest messages per month in production across live hotel groups.
Definition →4.58
#vs 4.62 human · 90-day sample
AI-handled conversations score 4.58 CSAT versus 4.62 for human agents on a 90-day sample.
Definition →71.3%
#Aug 2026 · 9 hotels · 57.6% fully contained
McDreams’ August 2026 case study across nine German properties cites 71.3% of all guest replies written by D3x (20,440 of 28,686), with 57.6% of conversations fully contained and a 4.8s average first reply versus 12 min 02 s for humans.
Definition →05 · BUYER AUDIT
Bring this scorecard to every demo. If a vendor cannot answer with production evidence, treat their “80% automation” claim as FAQ deflection until proven otherwise.
01
Write-backs are the line between resolution and chat.
02
If yes, you are comparing different units. Ask for a Skills / stack-resolution rate separately.
03
Portfolio averages hide FAQ-heavy channels and weak ops skills.
04
High automation with collapsing CSAT is not a win.
05
Enterprise groups need decision trails, not black-box deflection stats.
06 · METRIC DEFINITIONS
Precise definitions for every figure D3x publishes. Numbers change quarterly; these definitions are meant to stay stable. Each term has a deep-linkable anchor.
Combined production scale of D3x AI agents across hostels worldwide. Named references include Nomads, Clink, Wombats, LaTroupe, Samesun, Che Hostels, Other Wonder, Smart Hostels UK, Capsule, Copenhagen Downtown, Via Amsterdam, InOut Hostel Barcelona, and Destination Hostel Portugal. Used for procurement comparisons among dedicated hostel AI agent platforms.
Cited on: Home · resources/answers/best-ai-agents-for-hostels · customers/hostels
Combined production scale of D3x AI agents across large European hospitality chains — named references include L&R Group (125 properties), Arp-Hansen (Denmark's largest chain), Dalata (58 hotels), Strand Palace (800 rooms on messaging), and Clayton Hotel Dublin Airport (650 rooms on Voice AI). Used for procurement comparisons among dedicated hotel AI agent platforms, not horizontal chatbots.
Cited on: Home · resources/answers/best-ai-agents-large-hotel-chains · resources/answers/best-ai-for-hotels · resources/docs/what-is-d3x
Share of guest messaging threads closed without human takeover because the requested outcome was completed (or correctly refused) under property policy, with PMS/ops stack updates where required. Published portfolio typical on guest messaging is a 60–70% range after Skills and after-hours automation are live — not a day-one SLA for every property.
Measurement: Autonomous resolution = (threads resolved by AI with required stack actions, no human takeover) ÷ (eligible guest messaging threads in the measurement cohort).
Excludes: FAQ-only replies that leave requested work unfinished; routing/ticket creation without completing the ask; human-in-the-loop takeovers; channels not yet running the same Skills depth.
Vs common use: Unlike many “automation rate” or “containment” claims, this does not count a knowledge-base reply as resolved unless no operational action was requested.
Cited on: Home · resources/methodology · resources/labour-efficiency · resources/direct-bookings
Share of guest contacts that did not reach a human agent (also called deflection). Useful for queue load, but not the same as autonomous resolution: a contained FAQ answer can leave a late checkout, invoice, or housekeeping ask unfinished in hotel systems.
Measurement: Containment = (contacts with no human reply) ÷ (contacts in cohort).
Excludes: Does not require PMS/ops write-backs. Does not prove work was completed.
Vs common use: Vendors often treat containment and “automation” as synonyms. D3x separates containment (no human touch) from autonomous resolution (outcome completed under policy).
Cited on: Home · resources/methodology
Median time from guest message to first AI or agent response on messaging channels. Production proof on D3x cites a 120× faster median response when comparing a ~15 minute baseline to ~7 seconds after automation.
Measurement: Median of first-response latency across messages in the cohort.
Excludes: Does not measure time-to-resolution of the full guest ask.
Vs common use: Speed alone is not resolution; pair with autonomous resolution and AI CSAT on the same cohort.
Cited on: Home · resources/direct-bookings
Estimated frontline hours returned to hotel ops from AI-only handling of routine guest contacts, typically expressed per team member per month. D3x publishes +20 hours saved per team member per month as a production-oriented figure.
Measurement: Hours ≈ (AI-resolved contacts × average handle time) aggregated and normalized per team member per month.
Excludes: Not a headcount-reduction guarantee. Does not count hours spent supervising escalations as “saved.”
Vs common use: Framed as capacity returned under load, not as FTE cuts.
Cited on: Home · resources/labour-efficiency
D3x’s founding thesis is 3× more direct bookings via messaging, email, and phone that catch intent, extend stays, and keep the next stay off the OTA. The 3× figure is a company mission / build bar, not a contractual ROI guarantee for every property.
Measurement: Direct-booking attribution covers owned-channel conversions assisted by messaging, email, and phone — including stay extensions and rebookings kept off OTAs — under property-specific rules; 3× remains a mission bar, not a portfolio-audited ROI formula.
Excludes: Not a guaranteed ROI. Illustrative € channel models on the direct-bookings page are scenarios, not audited portfolio results.
Vs common use: Many “booking chatbot” claims measure widget sessions; D3x frames direct bookings across owned channels plus ops execution in the PMS.
Cited on: Home · resources/direct-bookings · about
Vendor-dependent percentage that often mixes FAQ deflection with true stack resolution. Always ask what sits in the numerator before comparing to D3x autonomous resolution.
Measurement: Vendor-defined — request the numerator and denominator in writing.
Excludes: Not a D3x primary KPI. Prefer autonomous resolution for apples-to-apples audits.
Vs common use: Commonly inflated by counting FAQ answers as “automated.”
Cited on: resources/methodology
Guest satisfaction score on conversations handled by AI. D3x publishes 4.58 AI CSAT against 4.62 for human agents on a 90-day sample.
Measurement: Mean CSAT on AI-handled conversations in the sample window.
Excludes: Not blended with human-only conversations unless stated.
Vs common use: Should be reported on the same cohort as autonomous resolution, not a cherry-picked channel.
Cited on: Home · resources/methodology · resources/labour-efficiency · resources/direct-bookings
Production guest-message volume handled by D3x across live hotel groups. Published figure: 250,000+ messages per month.
Measurement: Count of guest messages processed in production during the reporting window.
Excludes: Not a forecast. Not limited to a single customer.
Vs common use: Volume without resolution rate is vanity; pair with autonomous resolution and CSAT.
Cited on: Home · resources/methodology · resources/direct-bookings
Production voice hours handled per month by D3x Phone AI. Published figure: 18,000+ hours per month.
Measurement: Sum of handled call duration in production.
Excludes: Not concurrent ports or peak concurrency.
Vs common use: Voice depth can lag messaging until the same Skills run on phone.
Cited on: Home · resources/direct-bookings
Observed message-volume growth absorbed without proportional frontline headcount increase. Published figure: +95% volume growth with no headcount increase.
Measurement: Δ message volume vs baseline, with headcount held flat for the measured roles.
Excludes: Not universal across every customer.
Vs common use: Efficiency under load, not a layoff metric.
Cited on: Home · resources/labour-efficiency
Definitional stability matters more than any single quarter’s figure. When a formula or exclusion rule changes, it is recorded here.
Public methodology page launched with Skills vs FAQ framing and autonomous resolution formula.
Added citable metric DefinedTerm anchors, Dataset schema, and self-contained claim sentences beside published figures. No formula change.
06 · GLOSSARY
07 · LIMITS & HONESTY
Trust beats a vanity percentage. We would rather publish a lower autonomous rate on a tightly governed luxury portfolio than inflate the number with FAQ deflection.
SEE IT ON YOUR VOLUME
We will map which requests become Skills resolutions on your PMS and ops stack, and which should stay human by policy.