TL;DR
Inbound qualification is a different build from outbound dialing. The agent answers rather than calls, has to route rather than pitch, and its success metric is speed to a qualified human rather than dials per hour. Do not buy an outbound platform and configure it backwards. Best for: scoping the project correctly.
The capability that decides everything is whether the agent reads and writes CRM records mid-conversation or posts a transcript afterwards. Retell's own documentation is blunt that a connector posting a transcript an hour after the call is fine for compliance archiving and useless for personalisation. Best for: the one question to ask every vendor.
Synthflow publishes the most transparent inbound-relevant pricing: flat per-minute from $0.08 with enterprise as low as $0.07, no licence fees, no required contracts and explicitly no charges for failed calls, with speech recognition, model, voices, recording, transcription, CRM integration and API included. Best for: no-code inbound reception at predictable cost.
Bland should not be described as FedRAMP Class C certified. Its published assessment page explicitly states that it is not Class C certified. For a regulated deployment, evaluate the current assurance evidence, deployment scope and required agreements rather than a blanket certification list.
Speed to lead is the whole business case. Published research indicates response within five minutes materially increases conversion, and an always-on agent that qualifies and books in one call is competing against a human callback queue rather than against another agent. Best for: justifying the spend.
Contents
- What inbound qualification actually requires
- The handoff problem
- 1. Synthflow
- 2. Retell AI
- 3. Bland
- 4. ElevenLabs Agents
- 5. Twilio Conversation Relay
- Side by side summary
- Designing the qualification logic
- FAQ: Inbound Voice Qualification
What inbound qualification actually requires
An inbound qualification agent has five jobs and they are all different from outbound.
Answer immediately, every time. The entire economic case rests on this. Published guidance from a major CRM vendor states sales teams should contact new leads within five minutes of submission, citing research that rapid response significantly increases conversion. An agent that answers on the first ring at 2am is not competing with a better agent, it is competing with a callback queue.
Establish who is calling and why. Inbound callers are self-selected but heterogeneous. Some are buyers, some are candidates, some are existing customers with a support issue, and routing them apart is most of the value.
Apply qualification logic. Not a script, a decision tree with fallbacks. The agent needs to ask enough to route correctly and stop before it becomes an interrogation.
Schedule or transfer. Either book directly against a rep calendar or transfer live while the caller is still on the line. This is the step that separates a deployed system from a demo.
Write the outcome where a human will find it. Qualification result, conversation summary and next step, on the right record.
The fifth is where most projects fail, and it is the one buyers evaluate last.
The handoff problem
There are two architectures for CRM integration in this category and they perform completely differently.
Post-call sync waits until the call ends, then writes a transcript and a few fields to an activity record. It is simple, it is what most integrations mean, and it is adequate for reporting.
Mid-call read and write means the agent queries and updates records during the conversation. Retell documents this pattern explicitly for the enterprise CRM case: OAuth-authenticated API calls the agent makes mid-conversation, with lead lookup, contact updates, opportunity stage changes and case creation happening during the call rather than as a delayed post-call sync. Its own commentary is the clearest statement of why it matters, noting that real-time matters more than people assume and that most integrations miss the distinction, because a connector posting a transcript to an activity record an hour after the call ends is sufficient for a compliance archive and useless for personalisation.
For inbound qualification specifically, mid-call is not a nice-to-have. An agent that cannot look up whether the caller is already an open opportunity will re-qualify a live deal, which is worse than not answering.
The second half of the handoff problem is transfer quality. Category guidance is consistent that when the AI routes to a human, the full conversation context should transfer with it, and that integration depth varies sharply, with some platforms requiring middleware that introduces sync delays and maintenance overhead rather than offering native two-way sync.
1. Synthflow
Layer: No-code voice agent platform with flat per-minute pricing
Best for: Inbound reception and qualification without an engineering team
Synthflow is the most straightforward inbound build in this list and the most transparently priced. It is a no-code platform for building voice agents handling inbound and outbound calls through a drag-and-drop flow builder, with appointment booking, lead qualification and reception named as primary use cases. Documented capabilities include contextual routing, appointment booking, voicemail detection and SMS follow-ups out of the box.
Synthflow is a candidate for a managed voice workflow. Confirm the current per-minute package, included providers, failed-call treatment, support and deployment requirements. For sensitive data, request the current contractual and assurance evidence for that configuration.
For an inbound workload the no-charge-on-failed-calls policy matters less than it does outbound, but the inclusion of recording, transcription and compliance in the base rate matters a great deal, because those are the three items most commonly unbundled elsewhere.
The published customer evidence is public-sector and operational rather than sales-specific: a municipal deployment reporting three working AI agents, a 100% reduction in missed calls after hours, an 80% reduction in time spent on requests and fully automated data collection, described as successful enough to roll out citywide. A separate enterprise reference cites 65% voice automation across customer experience workflows.
For Synthflow inbound work, quote the expected minutes, concurrent calls, connected systems and support requirements. Confirm how incomplete calls are handled commercially.
2. Retell AI
Layer: Voice agent platform with real-time CRM reads and writes
Best for: Inbound qualification that must respect existing CRM state
Retell earns its place on integration architecture rather than price. Its documented integration surface covers a major CRM, an enterprise CRM, two support platforms, two contact-centre platforms, a document platform and custom API stacks, and lead qualification for inbound demand is a named use case alongside cold calling for outbound.
The enterprise CRM pattern is the reference implementation for this article's job. OAuth-authenticated API calls made mid-conversation, with lead lookup, contact updates, opportunity stage changes and case creation happening during the call. For inbound that means the agent can determine within the first fifteen seconds whether the caller is a net-new lead, an open opportunity or an existing customer, and route accordingly. No amount of post-call syncing produces that behaviour.
Retell also states the correct operating model plainly rather than overselling autonomy: AI phone agents handle repetitive high-volume work like appointment booking, order status and qualification, freeing the team for calls needing judgement or empathy, and most deployments run a hybrid where the agent takes routine volume and hands off to a person when the conversation calls for it. Documented agent capabilities include answering inbound calls and placing outbound ones, qualifying leads, booking appointments, checking order status, and routing or transferring to a live agent when needed.
Pricing: Voice AI pay-as-you-go at $0.07 to $0.31 per minute with 20 concurrent calls included and more available on demand, plus custom enterprise pricing. The base voice engine rate is $0.07 per minute covering the conversation layer including speech-to-text processing and real-time handling.
Where it falls short: A published range spanning more than four times from floor to ceiling makes early budgeting imprecise until model and voice choices are fixed. Concurrency above 20 is on demand rather than published, which matters for inbound specifically because inbound volume arrives in bursts you do not control. And Retell is a platform rather than a delivered solution, so the qualification logic, the routing tree and the calendar integration are your build.
Verdict: The right default when your CRM is the system of record and the agent must respect its state during the call. Size concurrency against your peak hour, not your average.
3. Bland
Layer: Fully bundled voice platform with enterprise compliance and delivered engineering
Best for: Regulated inbound and teams with no voice engineering capacity
Bland's relevance to inbound is compliance and delivery rather than any inbound-specific feature.
One per-minute rate covers the language model, speech-to-text, text-to-speech and telephony, with no per-token charges, no per-feature surcharges and no separate vendor invoices, and enterprise plans contracted on volume, dedicated infrastructure and compliance requirements. Independent worked modelling puts the flat rate at $0.09 rising to $0.13 per minute on a four-minute call, the lowest fully loaded figure among the major platforms.
For inbound qualification, the advanced feature list maps closely to what the job needs: warm transfers, live transfers, guardrails described as protected calls, alarm and monitoring, knowledge base gaps, citations, outcomes, custom dialing and custom code extraction. Knowledge base gaps and citations are the two most useful and least discussed, because an inbound agent that cannot say where an answer came from, or flag that it did not have one, will confidently mislead callers.
The delivery model is the differentiator for teams without engineers. Bland states most production agents go live in two to six weeks depending on conversation complexity, and that its forward-deployed engineer team builds the customer's first agent end to end, owning the pathway, the integrations and the test loop with the customer's operations team. Getting started is free: the Start plan provides two credits plus an inbound number valued at $15 per month with no card required, which is a genuinely usable inbound pilot.
Pricing: Bundled per-minute covering model, speech-to-text, text-to-speech and telephony, independently modelled at $0.09 to $0.13 all-in. Free Start plan with two credits and an inbound number valued at $15 per month. Warm transfer billing charges the proxy agent for talk time while active at the plan's per-minute rate. Enterprise contracted on volume, infrastructure and compliance.
Where it falls short: Bundling removes the ability to swap a cheaper or better provider into any single layer. Published per-tier rates are thinner than the feature matrix, so your actual rate is a sales conversation. And Bland's own analysis of the category is the best available caution on its own model, stating that published list prices tell enterprise buyers roughly 30% of their real bill with the rest in per-minute add-ons, concurrency overage, compliance surcharges and implementation.
Verdict: The right answer when the inbound line touches regulated data or when nobody internally can build the agent. Use the free Start plan to test the actual conversation before entering procurement.
4. ElevenLabs Agents
Layer: Voice-first agent platform with published minute allowances
Best for: Low to moderate inbound volume where voice quality is the differentiator
ElevenLabs approaches this from the voice layer upward rather than the orchestration layer down, and for an inbound line, which is frequently a caller's first experience of the company, that ordering is defensible.
The published tiers are unusually concrete for this category. Free at $0 per month includes 15 minutes of calls, 4 concurrent calls, the workflow builder, a knowledge base, multilingual support and a widget. Starter at $6 per month adds 75 minutes of calls, 6 concurrent calls, text messages and a commercial licence. Creator at $11 per month, or $22 with a first-month discount, adds 275 minutes and 10 concurrent calls plus additional minutes. Pro is $99 per month, with Scale at $299 and Business at $990 on the wider platform ladder and Enterprise custom.
Billing mechanics are documented clearly. There is no cost to create an agent, and calls made during setup and prompt testing are billed at half cost, which is a meaningful detail during a build. Voice-only calls are charged on duration with a 95% discount for silence periods longer than 10 seconds, which suits inbound patterns where callers pause. Language model costs pass through separately. In 2026 the company reduced agent pricing, with Starter moving to $0.08 per minute, alongside cuts of 55% to text-to-speech and 45% to speech-to-text on named models, and introduced pay-as-you-go for new self-serve users.
Pricing: Free $0 with 15 minutes and 4 concurrent calls. Starter $6 per month with 75 minutes and 6 concurrent calls. Creator $11 per month with 275 minutes and 10 concurrent calls. Pro $99 per month. Agent rate from $0.08 per minute on Starter with model costs passed through. Setup and testing calls billed at half cost. 95% discount on silence over 10 seconds.
Where it falls short: The included-minute allowances are small for a real inbound line. Two hundred and seventy-five minutes on Creator is roughly 90 three-minute calls a month, which is a pilot rather than a reception function, so most inbound deployments will be paying overage rates from the start. The credit system spans multiple products and can be confusing to forecast, with overages billed per model rather than per plan tier. And CRM depth is thinner than Retell's; this is a voice platform with agent capabilities rather than a sales platform.
Verdict: Strong voice quality, honest billing mechanics, and allowances sized for evaluation rather than production. Best where the inbound line is low volume and the caller experience is the point.
5. Twilio Conversation Relay
Layer: Carrier-native orchestration primitive for building your own agent
Best for: Teams with engineers who want the agent inside infrastructure they already run
Conversation Relay is not a product you configure, it is a primitive you build on, and for organisations already running their telephony on the same carrier it removes an entire vendor from the diagram.
The documented architecture combines fast speech-to-text and text-to-speech with the model of your choice, orchestrated through a WebSocket API. There is a dedicated TwiML noun with published reference documentation, configurable text-to-speech providers including three major voice vendors selectable by setting provider and voice attributes, automatic language detection and markup tags for speech control. Supporting documentation covers an onboarding guide, webhook security, voice AI attributes, an agent-to-human handoff blueprint, a conversational agent blueprint and a human agent augmentation blueprint. Reference architectures are published for integrating major model providers and for deployment on a large cloud platform.
The carrier's own technical framing of the latency approach is that raw audio packets stream directly from the cellular network to the model application layer, keeping the round-trip loop under 300 milliseconds so conversation flows naturally. That is a vendor claim about its own product and the measurement conditions are not stated.
Pricing sits on top of standard voice rates at $0.07 per minute for the relay, against published carrier rates of $0.014 per minute for outbound US and Canada voice, $0.0085 for inbound local and $0.0220 for inbound toll-free, with US local numbers at $1.15 per month and toll-free at $2.15. Model and voice costs are your own vendor relationships. Adjacent line items relevant to an inbound build include recording at $0.0025 per minute, storage from $0.0005 per minute per month tiering down, transcription from $0.024 to $0.05 per minute depending on mode, and text-to-speech from $0.0008 to $0.0130 per 100 characters by voice class.
Pricing: $0.07 per minute for the relay on top of standard voice rates. Inbound local $0.0085 per minute, inbound toll-free $0.0220. Numbers $1.15 and $2.15 per month. Model, voice and transcription priced separately at published rates.
Where it falls short: Everything above the audio stream is your build: qualification logic, state, CRM integration, testing, evaluation and monitoring. There is no drag-and-drop builder, no included knowledge base and no forward-deployed engineering team. Third-party testing vendors exist precisely because evaluating agents built this way is its own discipline. For a sales organisation without engineering capacity this is not a shortlist item.
Verdict: The most controllable and cheapest per minute at the infrastructure layer, and the most expensive in engineering time. Correct when telephony is already yours and an engineer owns the roadmap.
Side by side summary
| Tool | Layer | Best for | Entry price |
|---|---|---|---|
| Synthflow | No-code inbound and outbound voice agents | Reception and qualification without engineers | From $0.08 per minute, no contract |
| Retell AI | Voice platform with mid-call CRM reads and writes | Qualification that respects CRM state | $0.07 to $0.31 per minute |
| Bland | Bundled platform with compliance and delivery | Regulated inbound, no internal build capacity | Free Start plan, bundled per-minute thereafter |
| ElevenLabs Agents | Voice-first agent platform with minute allowances | Low volume where caller experience leads | Free with 15 minutes; Starter $6 per month |
| Twilio Conversation Relay | Carrier-native orchestration primitive | Engineering teams on existing telephony | $0.07 per minute plus standard voice rates |
Designing the qualification logic
The platform choice is the smaller half of this project. The qualification design is the larger half and it is where deployments succeed or embarrass you.
Identify before you qualify. The first action should be a CRM lookup, not a question. If the caller is an open opportunity, route to the owner. If they are an existing customer, route to support. Only net-new callers should enter qualification. This requires mid-call CRM access, which is why that capability leads this comparison.
Ask three questions, not eight. Category guidance is consistent that the agent should understand the product well enough to answer pricing questions, compare plans and address objections rather than just collecting form data. An inbound caller has volunteered their time and will forgive two or three qualifying questions in exchange for a fast answer. They will not forgive a form read aloud.
Define the escalation trigger explicitly. Decide in advance what causes a live transfer: a named account, a stated budget, a competitor mention, or three consecutive turns where the agent cannot answer. Bland's knowledge base gaps and citations features exist for precisely this failure mode, and any platform without an equivalent needs a hard fallback rule instead.
Write the outcome as structured data, not prose. Qualification result, reason, next step and owner as fields, with the transcript attached rather than substituted. A summary paragraph in an activity record is not routable.
Measure speed to qualified human, not call volume. The metric that justifies the spend is the interval from inbound call to a qualified prospect reaching a rep, compared with your current callback queue. Published guidance putting the response window at five minutes is the benchmark to beat, and an always-on agent beats it structurally rather than incrementally.
One honest caution on expectations. Vendor-published outcomes in this category are case studies rather than benchmarks: one platform cites a customer generating over $1 million in pipeline in three months at five times return, and another citing $500,000 in pipeline with a 3.2 times increase in email reply rates, with speed to lead and 24/7 coverage named as the drivers. Those are real numbers from real deployments and they are selected ones. Model your own case on your current inbound volume, your current callback latency and your current qualified rate, and treat vendor figures as evidence the category works rather than as a forecast.
FAQ: Inbound Voice Qualification
How is an inbound qualification agent different from an outbound AI dialer?
The agent answers rather than initiates, which changes the economics and the design. Outbound platforms bill per attempt including failures, with one vendor documenting a $0.015 minimum per outbound attempt and the same charge on failed calls, and their value is dials per hour. Inbound agents are billed on connected minutes because every call connects by definition, and their value is speed to a qualified human. Inbound also requires identification before qualification, since the caller may be an existing customer or an open opportunity, which outbound never faces.
Can the agent write to Salesforce or HubSpot during the call?
On some platforms yes, and it is the capability to test first. Retell documents OAuth-authenticated API calls made mid-conversation with lead lookup, contact updates, opportunity stage changes and case creation happening during the call rather than as a delayed sync, and states plainly that a connector posting a transcript an hour after the call is adequate for a compliance archive and useless for personalisation. Category guidance also warns that some platforms require middleware to reach the CRM, which introduces sync delays and maintenance overhead. Ask for a live demonstration of a mid-call record update rather than accepting an integration logo.
What does an inbound voice agent actually cost per month?
Model connected minutes, peak concurrency, numbers and optional services using the chosen package. Add implementation and support separately, and account for unsuccessful or repeated calls according to the provider’s current terms.
Do we still need a human on the phone line?
Yes, for escalation, and the vendors say so. Retell's own position is that AI does not replace call centre agents, that phone agents handle repetitive high-volume work such as appointment booking, order status and qualification while freeing the team for calls needing judgement or empathy, and that most deployments run a hybrid where the agent takes routine volume and hands off when the conversation calls for it. Design the transfer path before the qualification script, because an agent that cannot escalate turns a warm inbound caller into a complaint.
How long does it take to deploy one?
Two to six weeks for a production agent is the published figure from the vendor that also supplies the engineering, with the range depending on conversation complexity and its forward-deployed engineer team building the first agent end to end including the pathway, the integrations and the test loop. A no-code platform can produce a working agent in an afternoon and a trustworthy one in considerably longer, because the time is spent on edge cases, escalation rules and CRM mapping rather than on the conversation flow. Budget the integration work, not the build.
What compliance obligations apply to recording inbound calls?
Call handling also needs a review process. Our guide to conversation intelligence tools covers the adjacent options.
Work with Forma Nôrden
We build signal based outbound systems for B2B companies selling into the enterprise and upper mid market. Inbound qualification is a routing and CRM-state problem wearing a conversational interface, and teams that treat it as a voice project ship agents that talk well and route badly. Explore how we work.
For enquiries about this article: partnerships@formanorden.com
