Most outsourcing is priced by the minute, the hour or the FTE, and your clients are weighing AI for the same queues. VoiceRun lets you bring it to them first: agents on client queues with your people as the escalation path, every call scored on that client’s own scorecard, and models trained for one client at a time, inside the contact center they already run.
Built for COOs, SVPs of client services, heads of quality and chief digital officers at BPOs and outsourcers.
Client program · utility outage line
Sat 6:52 AM<1%
of annual revenue, in aggregate, sits in performance bonus and penalty provisions at one of the largest outsourcers
Concentrix, Form 10-K, FY202557%
overall employee attrition at one large outsourcer in 2025, down from 64% the year before
TTEC Holdings, Form 10-K, FY202550%
of companies that cut customer service staff because of AI are predicted to rehire for similar work by 2027
Gartner, customer service predictions, 2026The calls
Quality evidence comes first, because it is what keeps a client at renewal. Agent programs follow on the queues where the client’s policy is clearest and the volume is highest.
Each program’s calls are graded against that client’s own scorecard, with the transcript quote behind every criterion, so internal QA and the client’s audit start from the same evidence.
Systems the agent works in: Recordings · Client scorecards
Your QA team and the client’s auditor review the same call, the same quote and the same rationale. Overrides tune the scoring model for that program, and contested scores follow a tracked review.
Systems the agent works in: Scorecard · Contest workflow
Every call from a new hire is scored against the program’s scorecard from day one, so a go-live gate rests on a score trend and coaching starts from the agent’s own calls.
Systems the agent works in: Recordings · Scorecard · Coaching
An agent takes the routine intents on a client’s queue under that client’s policies and writes the outcome to their system, while your people keep the calls that need judgment.
Systems the agent works in: Client CRM · Billing · Order systems
Agents answer in parallel, so a surge on one program becomes contained routine calls plus a short queue for your people, while staffing catches up to what the client actually sent.
Systems the agent works in: CCaaS queues · Client CRM
When a person is needed, the agent can pass who called, why and what was done with the transfer, so your team picks up mid-conversation and the caller never starts over.
Systems the agent works in: CCaaS · Agent desktop
Add a language to a program without a hiring wave. Speech recognition and voices are chosen per language from the providers that perform best on each, and the agent follows a caller who switches mid-call.
Systems the agent works in: CCaaS queues · Client CRM
Appointment reminders, renewal notices, payment-due and outage-update calls placed at volume, with consent and calling-window rules written in code and every answer written back to the client’s system.
Systems the agent works in: Dialer lists · Client CRM
Required disclosures, identity checks and prohibited statements are checked on every call against what each client’s contract requires, with alerts routed to the supervisor who owns that program.
Systems the agent works in: Recordings · Alerts
Why it's hard
The models are rarely the obstacle. The obstacles are the billing model, the client’s stack, the client’s data, and the scorecard the client will use to judge the result.
The reality
Most outsourcing is priced by the minute, the hour, the FTE or the transaction. An AI agent that removes handle time also removes billable time, and a client who builds it themselves removes the program.
How VoiceRun handles it
You sell and run the AI program. Agents work the client’s queues with your people as the escalation path, and VoiceRun meters usage against a committed minimum with no per-seat fee, so you have a clean cost to price a per-resolution or gain-share contract against.
Usage-based pricing
Most outsourcing is priced by the minute, the hour, the FTE or the transaction. An AI agent that removes handle time also removes billable time, and a client who builds it themselves removes the program.
You sell and run the AI program. Agents work the client’s queues with your people as the escalation path, and VoiceRun meters usage against a committed minimum with no per-seat fee, so you have a clean cost to price a per-resolution or gain-share contract against.
Usage-based pricing
Tested before launch
Every client has calls it will use to judge your AI before go-live. Write them as scripted callers and run them against every release, and the acceptance test becomes one you have already passed.
Simulated callers place real voice calls to the agent, evaluators grade every conversation against what should have happened, and a release that regresses never reaches your callers.
$ vr simulate staging --name client-acceptance --wait
7/7 simulated calls passed · evaluator: judge
The platform
Start with the quality evidence, which needs only a recording feed and a scorecard. Agent programs, the call record that spans them and models for clients that opt in follow on the same platform.
Your systems
Every program has its own contact center, CRM and policies, and most of them belong to the client. Agents work through the same systems your people do, and scoring needs only the recordings you already keep.
Client contact center
Stays the home for human-handled calls; the agent escalates into the queues and skills your team works, whichever platform the client runs.
Common Genesys Cloud CX, NICE CXone, Five9, Talkdesk
Client CRM & ticketing
Where the agent reads the account and case history and writes what happened, inside the client’s system and under the client’s access rules.
Common Salesforce Service Cloud, Zendesk, ServiceNow, Microsoft Dynamics 365
Client back office
Billing, orders, outages, appointments: whatever the routine intents on that program actually read or change.
Recordings
The feed VoiceScore scores from, by webhook or storage bucket, against the scorecard each client owns.
Common Platform recording exports, Cloud storage buckets
Telephony
Numbers provisioned through VoiceRun or brought from the client’s own carrier account.
Common Twilio, Telnyx, Bring your own carrier
Named products are examples of what’s common in each category; none of them is a prebuilt connector. VoiceRun agents call any API or database your team exposes, and every connection is code your team can read. Your carrier, numbers and contact center stay where they are.
Control
Each of these is a setting or a rule in code you own, and each leaves a record the client can inspect.
Recording, redaction and tracing are set per environment and stay off until you turn them on, so each program runs only what its client’s contract allows.
Under our enterprise terms, VoiceRun does not use customer data to train models for VoiceRun generally, another customer or a third party without express written authorization.
Disclosures, authority limits, required offers and prohibited statements live in code you own. The agent says what the program allows and nothing outside it.
When a call needs a payment, the agent hands off to the client’s existing payment system, so card numbers stay out of the conversation and the transcript.
VoiceScore limits who sees which calls by role, down to the team and campaign, and redacts personal data from transcripts before storage and before any model call.
Agents and supervisors can contest any score, and each step of the review is recorded. The score of record traces to a quote your QA team and the client’s auditor can both read.
Your code remains your code, your data remains your data, and the models we train for you serve only you.
Deployment options: our SOC 2 managed cloud, your VPC, your cloud, or on-premises with VoiceRun Edge, available under enterprise agreement.
FAQ
That is settled in the contract. VoiceRun has no per-seat fee and meters usage against a committed minimum, which gives you a clean cost to price a per-resolution, base-plus-usage or gain-share structure against. Your team runs the program, owns the escalation path and holds the quality evidence, so the client buys one service from you.
Scorecards are separate documents, versioned and routed by call type, and VoiceScore scopes visibility by role down to the team and campaign. Recording, redaction and tracing are per-environment settings, and VoiceScore redacts transcripts before storage and before any model call. How those boundaries map onto your client structure is an architecture question we settle with you, with the client’s security questionnaire on the table.
That is the client’s decision, and it usually turns on three questions: does it work with the contact center they already run, does it pass their security review, and does it give them more visibility than they have today. The Audio Control Plane keeps their contact center as the home for human-handled calls and gives them one record per call. Where a client has residency requirements, deployment options extend to a VPC or on-premises under enterprise agreement.
VoiceScore grades every call against the client’s own scorecard and attaches the quote and rationale behind each criterion, so the evidence an auditor asks for already exists for the whole program. Your QA team and the client’s auditor can calibrate on the same calls, and contested scores are tracked step by step. That record matters most when a client decides how to split volume between vendors.
Some clients will go direct, and a contract clause rarely stops them. The position that holds is running the AI and the humans as one program: agents on the client’s queues, your people as the escalation path, and one scorecard grading both. VoiceRun is built for that role, with agents written as code in your repo, every major speech and language model to choose from, and a call record the client can audit.
Start with the program where the client already cares most about quality evidence. VoiceScore needs only a recording feed and the client’s scorecard, so it can run on a program before any agent does. Then put an agent on one or two high-volume intents of that same program, with your people as the escalation path, and release it to a slice of traffic first.
Bring the scorecard and a few recordings from one program you’re cleared to share. We’ll score them live and map the queue an agent should take first, with your people as the escalation path.
