Relevance AI

Relevance AI for telecom customer operations

Relevance AI sells a platform on which teams with little or no code build single agents and whole teams of them¹. Its home page brands the product the home of the AI Workforce², and the product page compresses the promise into building, running and managing agents at scale³. The draw is reach and speed: a very large integration directory, a no code builder, and prices published openly enough that a small team can begin without a sales cycle.

For a telecom head of customer operations, this assessment ranks Relevance AI seventh of seven. That is a judgment about fit with this buyer. Its pages are written with go to market teams in view. Pro targets a single GTM operator working alone⁴, and the product page speaks to the ops and revenue staff nearest the work⁵; a care floor of hundreds of advisers is a different audience. Its pages name no operator among customers, no contact center platform and no way to host inside an operator's own estate. Nexus, ranked first, publishes a telco page with billing, network and customer systems wired in from day one⁶, and its consultants put the first agent onto real contacts inside a month, as one step of a 3 month proof of concept⁷.

Background and customer evidence

Relevance AI counts its customers in the thousands of teams⁸ and displays a G2 score of 4.5 stars⁹. For a care team, Invent is the entry point: a tool meant to surface which tasks are worth automating and to convert what experienced staff know into running systems¹⁰. How old the company is, how many people it employs and how it is funded are questions its pages, as far as we opened them, leave unanswered.

The named results come from commercial and back office work. Send Payments saves 40 hours a week across its global operations¹¹. Zembl reports conversions up 30 percent with AI sales cover around the clock¹². Qualified credits more than 35 agents with $7M of pipeline¹³. A senior marketing manager at Autodesk gives a testimonial¹⁴. Customer Support does appear among the use cases on the home page¹⁵, but none of the named examples is an operator or a contact center deployment.

Building an agent, step by step

Relevance AI's documentation sets out five stages. A team starts an agent with Invent, copies one from the Marketplace or builds it fresh¹⁶. It equips the agent with tools from a no code tool builder¹⁷. It feeds in knowledge, either uploaded or synced from SharePoint, Notion, Google Drive or a website¹⁸. Several agents then become one workforce, arranged on a visual canvas¹⁹. Last come the guardrails: approval workflows, escalations and triggers²⁰.

Behind that canvas sit node and edge graphs that orchestrate the agents²¹. Python and JavaScript run sandboxed inside tools²². Quality rests on evaluations. No change publishes until the agent's eval suite has run again²³, and once agents are live, Relevance goes on scoring a sample of production runs so drift is caught²⁴. In a contact center, where answers lean on tariffs and policies that change, that live scoring is the feature to examine.

How fast, and with what help

Speed is the headline. An agent can be live within minutes, the documentation says, using the no code builder or a template²⁵. Describe the agent in plain words and Invent produces its prompt, its tools and its evals²⁶. Some use cases, the agents page adds, stand up in as little as a few hours²⁷, with agents written by the customer's own people in plain language²⁸.

There is a service layer as well. Relevance AI also offers people. Before any build, Relevance will sit with an operator to map workflows and pick the best agent use cases²⁹. For the launch itself it embeds a deployment team, which then trains the operator's people to keep building quickly on their own³⁰. Who staffs that team, what it costs and which plans include it go unexplained on the pages we opened.

An operator's own coders are catered for too. Relevance names Cursor, Codex and Claude Code as ways in³¹, and one MCP command, "mcp add relevance-ai", connects them³².

Integrations and how agents start

Relevance AI's integration figures are large and vary by page. The integrations page's own numbers do not agree. Its banner claims upward of 2,000 integrations³³; the listing under it shows 2,728³⁴. The home page speaks of more than 1,000 apps³⁵; the product page of above 1,000 native connectors, MCP support and custom connectors³⁶; the Enterprise tier of over 2,000³⁷. Because the counts differ, check each system you need by name.

Agents can be started on a schedule, by an app event, through a webhook or by a custom signal³⁸. In customer operations, the likely first trigger is a webhook from the ticketing system. We found no telecom billing system, OSS or BSS platform, or contact center platform among the named integrations on the pages we opened.

Voice comes with the Team plan, which adds calling and meeting agents to what Pro includes³⁹. Whether those calling agents could carry inbound subscriber traffic at an operator's scale is a question the pages we opened do not answer.

Security for subscriber conversations

Relevance AI states SOC 2 Type II compliance⁴⁰, and the enterprise page carries SOC 2 Type II and GDPR badges^41,42. Data is encrypted at rest⁴³, and customer data is excluded from training⁴⁴. Relevance AI's documentation adds a stronger promise: the vendor's own staff cannot see a customer's project information⁴⁵. Emails, phone numbers and similar PII can be masked or blocked⁴⁶, which matters in telecom conversations thick with account numbers and contact details.

On availability, the enterprise page commits to 99.9 percent uptime under an SLA⁴⁷. The facts gathered for this site include an uptime percentage.

Approvals and oversight

Relevance AI's governance features are practical. At Enterprise level, an operator's identity team would find SAML single sign on⁴⁸ and access governed by role⁴⁹. Where an action carries high stakes, the agent waits for a person to approve, at thresholds the customer sets⁵⁰. The product page lists approval rules set per tool⁵¹ and conditional approval rules⁵², escalation by Slack or email⁵³, and agent versions held as draft, active or pinned⁵⁴. Access control, permissions and data security share a single pane⁵⁵, and the enterprise page promises quality standards, guardrails and monitoring on every agent⁵⁶.

For an operator, per tool approval would let an agent look up an account freely while holding back a credit or refund until a supervisor signs.

Where data and logs live

Data residency⁵⁷ and audit logging⁵⁸ are both on the enterprise list. An operator's SOC could receive every audit entry and execution trace in its own store⁵⁹, formatted as OpenTelemetry⁶⁰, and Delta Share is a second export route beside OTEL⁶¹. An operator whose security team runs its own monitoring stack will value that log stream.

Hosting inside the customer's estate, whether on premises, in the customer's VPC or self hosted, is not offered on the pages we opened. Operators whose subscriber conversations must stay on their own infrastructure should raise this first.

Published prices

Relevance AI is one of only two vendors here to publish prices. Pro costs $19 a month on annual billing or $29 on monthly billing⁶². Team costs $234 a month annually or $349 monthly⁶³. Paying annually is 33 percent cheaper than paying monthly⁶⁴. Pro comes with 30,000 Actions a year⁶⁵; additional Actions are $80 for each 1,000⁶⁶, and additional Vendor Credits $20 for each 10,000⁶⁷. Model and tool usage paid through Vendor Credits passes through at wholesale cost, unmarked⁶⁸, and Pro lets customers bring their own API keys⁶⁹.

The Enterprise tier, which an operator would need, carries no figure on the pricing page, and its customers are assigned their own account manager^70,71. Relevance aims it at companies that want growth without matching growth in headcount, using an AI Workforce⁷².

For a telecom contact center, Actions are the unit to model. A busy queue could use them quickly, top ups come only in blocks of 1,000⁷³ and only on an active paid plan⁷⁴. Ask for an Enterprise quote built from your own contact volumes.

Support terms and stated limits

The support documentation is frank. Dedicated staff calls are reserved for Enterprise customers⁷⁵. Free customers get no SLA⁷⁶. Relevance's own help desk keeps Sydney office hours, nine to five, Monday to Friday, and is closed on weekends⁷⁷. Engineering receives bug reports, though Relevance says it will not reply to them⁷⁸. Its Harley AI support agent is a beta and may contain bugs or errors⁷⁹.

A contact center open seven days a week would need to negotiate past that weekday window in any Enterprise contract. Other limits: Pro allows 2 build users and 1 project⁸⁰, a subscription cannot move from one organization to another⁸¹, and the Free plan is retired⁸². The old Business plan was sunset and its customers moved onto Team⁸³. Enterprise customers are told to check rollover terms with their account manager⁸⁴.

Where it could fit in an operator

The pages we opened point to a few places where Relevance AI could earn a role at an operator, even if it is not the first choice for the inbound queue.

The commercial side of customer operations is one. Plans are written for GTM operators and for teams building at scale^4,85, and Zembl's case is about round the clock AI sales cover¹². An operator with an outbound retention or upgrade team could pilot there, using the Team plan's calling agents³⁹.

Back office work behind the contact center is another. Schedule, app event and webhook triggers³⁸ suit tasks that follow a contact, such as chasing a missing document or updating the CRM after a call. Approval rules per tool⁵¹ keep a person in the loop for any step that moves money.

A sandbox for the operator's own team is a third. With entry prices public⁶² and an MCP route for engineers^31,32, a small group can learn how agents behave before committing to a bigger platform.

Each of these is a supporting role. For the core job of a head of customer operations, taking inbound contacts off the queue at an operator's scale with agents tied into billing and network systems, the pages we opened give Relevance AI less to stand on than the vendors above it.

Strengths for an operator

1. Public prices from $19 a month⁶², with model usage at wholesale⁶⁸. 2. A directory listing 2,728 integrations³⁴. 3. Evals re-run on every change and sampled scoring of production runs^23,24. 4. Approval rules per tool and conditional approvals^51,52. 5. A 99.9 percent uptime SLA⁴⁷. 6. Log streaming in OpenTelemetry format to destinations the customer owns^59,60.

Gaps for an operator

1. No operator, telecom integration or contact center platform named on the pages we opened. 2. Its clearest audience is GTM, ops and revenue teams^4,5. 3. No on premises or customer VPC hosting. 4. Weekday support in Sydney hours, nothing at weekends⁷⁷. 5. Integration counts that change from page to page^33,35.

Verdict

Relevance AI gives a team a quick, priced and well governed way to build agents, and its evaluation and approval features are strong. For a telecom head of customer operations it ranks seventh: its pages speak to go to market work, name no operator or contact center deployment, and offer no hosting inside the operator's estate. Nexus, ranked first, brings an offer written for operators and experienced consultants on the care floor, gets the first agent answering inside the opening month of a 3 month proof of concept, and charges per role automated, never for consultant time⁷.