Methodology

CCaaS 60-Capability Benchmark

Methodology, capability catalogue and public-evidence matrix. Companion to CCaaS Convergence Is Real. But It Is Not Capability Parity.

Research snapshot	24 July 2026
Platforms	Microsoft Dynamics 365 Contact Center, Salesforce Agentforce Contact Center, Genesys Cloud CX, NiCE CXone Mpower, Amazon Connect

1. Purpose

This benchmark is intended to improve the quality of a contact-centre shortlist. It is not intended to declare a universal winner. It separates three questions that are often collapsed into one:

Coverage: Is the capability publicly documented?
Depth: Does it satisfy the organisation’s operational scenario?
Fit: Is it the right architectural and commercial choice for this buyer?

A public product page can help answer the first question. Only detailed documentation, contractual confirmation and representative testing can answer the second. The third requires buyer-specific priorities. For that reason, the accompanying article does not publish a single total score.

2. Scope and exclusions

Included

Native capabilities in the named contact-centre product.
First-party modules or editions sold as part of the vendor’s contact-centre proposition.
Documented first-party integrations where they are a standard part of the operating model.
Generally available, limited-availability and preview capabilities, provided their status is labelled.
Voice capabilities and the adjacent operational controls required to run voice at enterprise scale.

Excluded

Roadmap promises without public documentation.
Bespoke professional-services work presented as a standard product feature.
Marketplace products unless the platform explicitly depends on a partner model.
Analyst rankings, vendor market share and unverified sales claims.
Pricing comparisons. Public list prices rarely represent the complete configured service.
Performance claims that have not been tested using the same workload and conditions.

Deferred to a future revision

Secure payment capture (PCI DSS pause-and-capture, DTMF suppression) beyond the masking controls in capability 33.
Voice biometrics and fraud detection beyond the identification controls in capability 16.

These are deliberate deferrals, not judgements of unimportance.

3. Evidence hierarchy

Use the highest available evidence and record the access date.

Grade	Evidence	Appropriate use
A	Current administrator/developer documentation, release note or published technical limit	Can support a documented-coverage judgement
B	Current official product or packaging page	Can establish vendor positioning or packaging, but not operational depth
C	Written response from the vendor tied to product version, edition and region	Can support a procurement judgement when incorporated into the contract
D	Demonstration using the buyer’s scripted scenario and representative data	Can support a functional-depth judgement
E	Production proof of concept with agreed success criteria	Strongest evidence before selection

Marketing copy alone must not receive a depth score of 3. Grade B evidence can support an M status where it documents a packaging, edition or licensing condition, but a D status requires Grade A evidence. A release plan documents intent, not current availability, and cannot support any status on its own.

4. Functional scoring scale

The public-evidence matrix later in this document is not itself a functional score. When the benchmark is used in procurement, score each capability as follows:

Score	Meaning	Required evidence
0	Not available or no acceptable evidence	Vendor confirms absence, or cannot demonstrate it
1	Possible through custom build, partner product or material workaround	Architecture and ownership documented
2	Available with a material limitation	Limitation may be edition, add-on, preview status, region, scale or reduced control
3	Available and demonstrated against the required scenario	Product version, edition, region and test result recorded

Important rule. “Documented” is not synonymous with “score 3”. A capability reaches 3 only after the vendor demonstrates it under the buyer’s acceptance scenario.

5. Weighting

Do not assign every buyer the same weights. For each capability, the buyer assigns:

Weight	Meaning
0	Out of scope
1	Useful
2	Important
3	Mandatory

weighted coverage = SUM(functional score × buyer weight)
                    ÷ SUM(3 × buyer weight)
                    × 100

Report mandatory-capability failures separately. A high aggregate score must never conceal a zero or one on a mandatory requirement.

6. Public-evidence status used in this matrix

The matrix below records only what the first-party source set was sufficient to establish. This is a public-documentation review, not a controlled product test: public documentation is uneven, and a vendor with more detailed documentation can appear stronger even when products are functionally similar. The matrix records evidence conservatively and still requires vendor review and scripted demonstration before any selection decision.

Code	Meaning
D	Documented as a native or standard first-party capability
M	Documented, but a module, add-on, edition, region, provider or material condition applies
I	Primarily dependent on integration, partner telephony or customer implementation
U	Not confirmed in the reviewed first-party source set

Assessed product baseline

Statuses are recorded against the most complete first-party contact-centre package each vendor sells, not the entry edition:

Platform	Assessed baseline
MS	Dynamics 365 Contact Center (standalone licence) with Copilot Studio
SF	Agentforce Contact Center Plus on a supported Agentforce Service edition
GC	Genesys Cloud CX 3 with Workforce Engagement and AI Experience capabilities
NC	NiCE CXone Mpower suite including workforce engagement management
AC	Amazon Connect with first-party optional features enabled

Edition or tier gating within the assessed baseline does not by itself trigger M. M applies when a capability requires a product outside the assessed baseline or another product family, carries a region, preview or limited-availability restriction, or has a vendor-documented material limit. D implies the capability is generally available in at least one region under the assessed baseline.

These codes must not be added together to produce a winner. They say nothing about comparative depth, ease of use, scale, implementation effort or price. A count of qualified cells reflects documentation style and product age as much as capability: Salesforce’s column carries more M and U statuses than any other platform largely because its contact-centre packaging is the newest and its published limits documentation is unusually explicit. That is a property of the evidence, not a finding of inferiority.

7. The 60 capabilities

MS: Microsoft Dynamics 365 Contact Center · SF: Salesforce Agentforce Contact Center · GC: Genesys Cloud CX · NC: NiCE CXone Mpower · AC: Amazon Connect

Domain 1 — Voice connectivity

#	Capability	Minimum acceptance question	MS	SF	GC	NC	AC
1	Inbound public telephony	Can the service receive production PSTN calls in every required country?	D	D	D	D	D
2	Agent-initiated outbound calling	Can an agent place a compliant outbound call from the service desktop?	D	D	D	D	D
3	Number acquisition and management	Can administrators acquire, port, assign and retire required number types?	D	M	D	D	D
4	Bring-your-own-carrier or SIP connectivity	Can the buyer retain its carrier and routing strategy without losing core controls?	M	M	D	D	I
5	Geographic and regional coverage	Are telephony, media, data and support available in every required region?	M	M	D	D	D
6	Voice quality diagnostics	Can operations diagnose jitter, latency, packet loss, device and carrier problems?	D	M	D	D	D

Interpretation note: Amazon Connect’s first-party documentation offers number porting and call forwarding as the routes to number retention; native carrier-replacement SIP trunking is not documented. Capability 4 is therefore recorded as I for Amazon Connect — the pattern depends on forwarding, partner services or customer implementation, not on a documented product control.

Domain 2 — Routing and queue control

#	Capability	Minimum acceptance question	MS	SF	GC	NC	AC
7	Skills-based routing	Can work be allocated using multiple skills and operational constraints?	D	D	D	D	D
8	Queue, priority and service-level routing	Can administrators manage priority, ageing and service objectives?	D	D	D	D	D
9	Proficiency, presence and capacity	Does routing account for proficiency, availability and concurrent workload?	D	D	D	D	D
10	Conditional and policy-based routing	Can business data and policy determine the route without bespoke code?	D	D	D	D	D
11	Overflow, callback and voicemail	Can the service preserve intent and queue position when agents are unavailable?	D	D	D	D	D
12	External or extensible routing	Can an external decision service influence or control routing safely?	D	D	D	D	D

Domain 3 — IVR and self-service

#	Capability	Minimum acceptance question	MS	SF	GC	NC	AC
13	DTMF IVR	Can flows capture keypad input, validate it and recover from failure?	D	D	D	D	D
14	Speech recognition and conversational IVR	Can callers use natural speech across required languages and accents?	D	D	D	D	D
15	AI voice self-service	Can a conversational AI agent manage a complete bounded use case?	D	D	D	D	D
16	Identity and authentication steps	Can the flow apply risk-based identification and authentication?	M	M	M	M	M
17	Transaction and API actions	Can self-service read and update systems through governed actions?	D	D	D	D	D
18	Contextual escalation within an interaction	When self-service escalates mid-interaction, do identity, intent, transcript and completed steps transfer to the human in the same session?	D	D	D	D	D

Domain 4 — Outbound engagement

#	Capability	Minimum acceptance question	MS	SF	GC	NC	AC
19	Preview or agent-controlled dialling	Can the agent review context before accepting or placing the call?	D	D	D	D	D
20	Progressive dialling	Can the service place one controlled call for available capacity?	D	M	D	D	D
21	Predictive dialling	Can pacing be demonstrated with required staffing and abandonment controls?	M	M	D	D	D
22	Power or agentless voice campaigns	Can high-volume automated or paced voice outreach be governed?	D	M	D	D	D
23	Answering-machine detection	Can live people, voicemail and uncertain results follow different policies?	D	U	D	D	D
24	Campaign, list, consent and retry management	Are eligibility, suppression, time zones, retries and outcomes governed?	D	M	D	D	D

Interpretation note: Salesforce’s help documentation for automated outbound campaigns confirms progressive and predictive dialling (“uses AI to automatically manage the calling rate based on real-time rep availability”), delivered through Amazon Connect Outbound Campaigns as an add-on to Salesforce Voice with Amazon Connect. Capabilities 20 and 21 are therefore M for Salesforce: documented, with add-on and telephony-model conditions. The same page states that campaigns connect reps “only to live, human-answered calls”, but does not document differentiated policies for voicemail and uncertain results, so capability 23 remains U — a verification item, not a claim of absence. Under other partner telephony, Salesforce’s developer guide delegates dialler implementation to the vendor telephony system. Microsoft’s proactive voice dial modes reached general availability in February 2026, including answering-machine detection and predictive pacing; however, Microsoft’s own documentation states that call connection time in representative-led progressive and predictive modes is not compliant with TCPA and directs buyers to review applicable telecommunications regulation. Capability 21 is therefore recorded as M for Microsoft, and representative-led outbound compliance must be validated per region on every platform.

Domain 5 — Agent experience

#	Capability	Minimum acceptance question	MS	SF	GC	NC	AC
25	Unified voice and digital desktop	Can the agent handle required channels without fragmented state?	D	D	D	D	D
26	Core call controls	Are answer, hold, mute, resume and disconnect controls complete?	D	D	D	D	D
27	Consult, transfer and conference	Can agents transfer internally and externally with preserved context?	D	D	D	D	D
28	Customer identification and screen pop	Does the correct customer and case context appear reliably?	D	D	M	M	M
29	Disposition and after-contact work	Can outcomes, notes, codes and wrap time be governed and reported?	D	D	D	D	D
30	Real-time assistance and knowledge	Can the desktop provide contextual guidance, knowledge and next actions?	D	D	D	D	D

Domain 6 — Recording, security and compliance

#	Capability	Minimum acceptance question	MS	SF	GC	NC	AC
31	Voice recording	Can policy determine which calls and segments are recorded?	D	D	D	D	D
32	Agent screen recording	Can authorised reviewers replay screens alongside the interaction?	D	I	D	D	D
33	Pause, resume and sensitive-data masking	Can recording and display of protected data be controlled reliably?	M	I	D	D	M
34	Retention and legal-hold policy	Can different interaction types follow controlled retention rules?	M	M	D	D	M
35	Encryption and role-based access	Are recordings, transcripts and keys protected by least privilege?	D	D	D	D	D
36	Consent, policy and audit evidence	Can the buyer prove which policy ran, who accessed data and what changed?	D	M	D	D	D

Interpretation note: Salesforce voice can use native or partner telephony. Recording-dependent capabilities must therefore be confirmed against the chosen telephony model rather than inferred from Salesforce CRM functionality alone. Salesforce’s published limits are unusually explicit here: recording access is restricted to the owner of the Voice Call Recording record, a paused recording does not resume visibly after a page refresh, and without Contact Lens only the first 15 minutes of a call are transcribed. These documented limits support the M statuses in this domain and are exactly the class of constraint a scripted demonstration should target.

Domain 7 — Analytics and quality management

#	Capability	Minimum acceptance question	MS	SF	GC	NC	AC
37	Voice transcription	Are required languages, latency and accuracy supported?	D	D	D	D	D
38	Sentiment, topic and interaction analytics	Can the system find operational patterns across interactions?	D	D	D	D	D
39	Automated summarisation	Can summaries be generated, corrected and written to the right record?	D	D	D	D	D
40	Manual quality evaluation	Can teams define forms, calibrate scoring and manage review workflows?	D	D	D	D	D
41	AI-assisted or automated evaluation	Can AI score defined criteria with evidence and human review?	D	M	D	D	D
42	Live supervisor monitoring and intervention	Can supervisors listen, coach, barge or take action with audit controls?	D	D	D	D	D

Domain 8 — Workforce management

#	Capability	Minimum acceptance question	MS	SF	GC	NC	AC
43	Demand forecasting	Can the engine forecast the buyer’s channels, queues and planning horizons?	D	M	D	D	D
44	Capacity planning	Can planners model staffing, shrinkage, skills and alternative scenarios?	D	M	M	D	D
45	Schedule generation and optimisation	Can schedules respect contracts, skills, rules and service objectives?	D	M	D	D	D
46	Real-time adherence	Can actual state be compared with schedule quickly and accurately?	D	M	D	D	D
47	Intraday management and replanning	Can operations identify gaps and make controlled same-day changes?	D	M	D	D	M
48	Employee schedule self-service	Can employees view schedules and manage time off or shift changes?	D	M	D	D	D

Interpretation note: Workforce management is packaging-sensitive on every platform: Genesys sells it within CX 3 and Workforce Engagement tiers, NiCE licenses it as part of the CXone Mpower suite, Amazon prices it as an optional feature, and Salesforce sells it in the Agentforce Contact Center Plus package or as a separate workforce management add-on. Under the assessed baselines in section 6, packaging alone therefore does not differentiate these cells. Salesforce remains M because its documentation publishes material limits, including forecast horizons and planning constraints, and because Agentforce Contact Center Workforce Engagement Management reached general availability on 22 June 2026, inside this snapshot window. First-party workforce functionality on any platform should be tested against the buyer’s workforce rather than converted into a generic maturity ranking.

Domain 9 — Administration, resilience and extensibility

#	Capability	Minimum acceptance question	MS	SF	GC	NC	AC
49	Role-based administration	Can operational, security and technical duties be separated?	D	D	D	D	D
50	Administrative audit logs	Can configuration and access changes be reconstructed?	D	D	D	D	D
51	APIs, events and webhooks	Can interactions and configuration participate in enterprise integration?	D	D	D	D	D
52	Observability and real-time dashboards	Can teams detect service, queue, media and workflow degradation?	D	D	D	D	D
53	Availability, resilience and disaster recovery	Are failure domains, service levels and recovery behaviours documented?	M	M	D	D	D
54	Configuration promotion and change control	Can tested configuration move between environments with rollback and audit?	D	D	M	M	D

Domain 10 — AI and orchestration

#	Capability	Minimum acceptance question	MS	SF	GC	NC	AC
55	Generative knowledge responses	Can answers be grounded, cited, evaluated and constrained?	D	D	D	D	D
56	Autonomous conversational voice agent	Can an AI agent own a bounded voice journey in production?	D	D	D	D	D
57	Governed tool and action execution	Can actions be authorised, logged, retried and reversed?	D	D	D	D	D
58	Workflow and journey orchestration	Can state and tasks persist beyond a single interaction or channel?	D	D	D	D	D
59	Human handoff with journey context	When a journey spans channels, sessions or systems, does the receiving person get objective, history, evidence and open work from persistent journey state?	D	D	D	D	D
60	AI governance, evaluation and guardrails	Can teams test, monitor, restrict and audit AI behaviour and cost?	D	D	M	U	D

Interpretation note: NiCE documents Autopilot voice deployment as requiring NiCE assistance to set up, and Genesys AI Studio capabilities consume AI Experience tokens under a usage-based model; both are commercial and operating conditions to confirm, not availability gaps. AI governance, evaluation and guardrail documentation was not confirmed for NiCE in the reviewed source set, so capability 60 is recorded as U for NiCE — a verification item, not a claim of absence.

8. How to turn each capability into a test

Every mandatory capability should have:

A scenario: what the customer, agent, supervisor or administrator is trying to achieve.
Representative conditions: region, language, volume, concurrency, integration and security constraints.
Expected evidence: screen, log, API event, recording, report or exported configuration.
Acceptance threshold: a measurable pass condition.
Failure test: what happens when a dependency, model, tool or user action fails.
Product coordinates: edition, licence, region, release status and required adjacent products.

Capability: Human handoff with journey context

Scenario:
A customer authenticates with an AI voice agent, explains a billing dispute,
uploads evidence through messaging and is escalated to a human.

Pass:
The human receives verified identity, intent, transcript, uploaded evidence,
actions already completed, policy result and recommended next step without
asking the customer to repeat information.

Failure test:
The billing API fails immediately before handoff. The journey remains active,
the failure is visible, and the human receives a safe recovery action.

9. Bias controls

The following controls apply to future revisions:

Do not infer platform depth from vendor age or market category.
Do not assign Microsoft or Salesforce an ecosystem advantage unless the buyer actually uses that ecosystem.
Do not describe Salesforce using an older Service Cloud Voice baseline when assessing Agentforce Contact Center.
Do not treat Genesys or NiCE as legacy merely because they began as contact-centre specialists.
Do not describe Amazon Connect as purely build-it-yourself without checking its current packaged capabilities.
Do not use “native” as a proxy for better. Native capability can still be shallow; an integration can still be operationally superior.
Do not score roadmap and generally available capability equally.
Do not convert missing public evidence into a definitive claim of absence.
Freeze the evidence date and refresh it before publication or procurement.

10. First-party source register

The following sources were used to establish the July 2026 public-evidence baseline. Several were updated during 2026; recheck them before relying on any individual status.

Microsoft

Salesforce

Genesys

NiCE

Amazon Web Services

11. Change log

Date	Platform	Capability	Old	New	Basis
2026-07-24	Salesforce	20	U	M	First-party help documentation confirms progressive dialling through Amazon Connect Outbound Campaigns, sold as an add-on to Salesforce Voice with Amazon Connect (source register, Salesforce item 10, accessed 24 Jul 2026)
2026-07-24	Salesforce	21	U	M	Same source documents AI-managed predictive pacing based on real-time rep availability; add-on and telephony-model conditions apply (source register, Salesforce item 10, accessed 24 Jul 2026)
2026-07-24	Microsoft	21	D	M	Vendor dial-modes documentation states representative-led progressive and predictive call connection time is not TCPA-compliant (source register, Microsoft item 3, accessed 24 Jul 2026)
2026-07-24	NiCE	60	D	U	AI governance, evaluation and guardrail capability not confirmed in the reviewed first-party source set; verification item, not a claim of absence
2026-07-24	Amazon	4	M	I	First-party documentation offers number porting and forwarding, not native carrier-replacement SIP trunking (source register, AWS item 9)

Non-matrix revisions, 24 July 2026: assessed-baseline table and M tie-break rule added to section 6, together with a note on interpreting qualified-cell counts; evidence-grade rule clarified in section 3; capabilities 18 and 59 acceptance questions sharpened to remove overlap; deferred-capability note added to section 2; interpretation notes expanded in Domains 1, 4, 6, 8 and 10; source register expanded with Genesys, NiCE and AWS AI documentation and two Salesforce outbound sources, the Microsoft screen-recording citation corrected from a release plan to current documentation, and the primary Salesforce reference corrected; NiCE CXone Mpower naming adopted; capability-table formatting corrected. Capability 23 remains U for Salesforce: the outbound campaigns documentation describes connecting reps only to live, human-answered calls, but does not document differentiated policies for voicemail and uncertain results.

Change-control note: Record every future matrix change with the date, platform, capability number, old status, new status and supporting first-party URL. This prevents a fast-moving release from silently changing the conclusion.

Read the companion article

This methodology supports the analysis in the convergence article. Neither publishes a total score, and neither should be used without vendor confirmation and scripted demonstration.

CCaaS Convergence Is Real

Reference hub