# Conversational AI for phone calls: How AI voice agents handle long, complex conversations

By Samuel Oram-Jones

Conversational AI is software that understands what a person says or types and answers in natural language, keeping track of the conversation so far. A phone call is the hardest place to do that, because there are no buttons to tap, no scrollback to reread, and a caller who can change their mind mid-sentence.

Bland builds AI phone agents that answer inbound calls and place outbound calls. Your team builds and edits them in plain English with Norm or in Bland's visual builder, with no engineers needed on your side. Enterprise customers also get Bland's forward-deployed engineers.

<!-- S8-SLOT lead-claim-sec11 -->

## What is conversational AI, and how does a phone call run through it?

On a phone call, conversational AI is a loop of four jobs: speech in, understanding, a decision about what to do next, and speech out. Speech recognition turns the caller's voice into text.

A language model works out what the caller means. A dialogue manager decides the next step, which might be asking a question, looking something up, or calling a tool such as a calendar. Text to speech then says the reply aloud.

Chat forgives slow software and phone calls do not. A two-second pause in a chat window is a loading indicator, and on a phone it sounds like a dropped line. Bland runs at sub-200ms latency. For more on how turns stack up, see [multi-turn conversation](/blog/multi-turn-conversation), or start with [what conversational AI is](/blog/conversational-ai).

Buyers also ask whether a conversational AI platform is a thin layer over a third-party model such as GPT or Claude. Bland self-hosts its models and uses no third-party model providers. Bland operates a closed-source model it developed itself.

## How do Bland's agents hold a long, complex phone conversation?

Bland's agents hold a long, complex call by building it from steps called nodes, and a node can hold the call on its step until it has what it needs.

A long insurance intake or a multi-step booking breaks in the same few places: the caller asks something off-script, changes an answer from earlier in the call, asks the agent to hold, or calls back next week.

### Conversational Pathways give the call a structure

Conversational Pathways are node-based conversation flows that give an agent structured control over dialogue, branching, and actions. Most nodes represent a phase of conversation with multiple back and forth dialogues, such as collecting a policy number or asking for a date. [Prompt-based agents vs Conversational Pathways](/blog/prompt-based-vs-conversational-pathways) covers when each approach fits.

### A node holds the step until it has what it needs

A node can carry a condition. If the condition is not met, the agent stays on the same node until it is fulfilled. In Bland's reservation example, the condition on the first node is "You must get the date, time, and number of guests for this reservation", so the agent stays on that node until the caller has given all three.

### An off-topic question gets answered, then the call picks up where it left off

Any node can be set as a global node, which every other node can reach. When a caller asks something unrelated to the current step and it matches a global node's label, the agent moves to that global node, answers, and then returns to the node it was at before.

In the reservation example, a caller asks for opening hours partway through booking, the agent gives the hours, and the agent goes back to asking for the reservation details.

### Flex Mode lets the agent backtrack, jump ahead, or stay put

By default a pathway is strict, and the agent follows the edges between nodes. In Flex Mode the edges are suggestions, not commands, and the agent can backtrack, jump ahead, or stay put.

That covers a caller who says *Wait, can I come back to the email question? I gave you the wrong one.* Strict mode stays available for auth flows, scripted disclosures, ordered data collection, and compliance, and Flex Groups let you flex some sections while keeping others strict.

### The Wait for Response node holds while a caller looks for a card

The Wait for Response node works like a default node, and it can also wait if a caller needs time to respond or needs to hold for a moment. On that node, when a caller says *give me a second, let me find my card*, the agent waits instead of pressing on.

### Webhooks run in the middle of a call

A webhook node executes webhooks at any point during the conversation, and it can send speech during or after the webhook.

### Memory covers the callback

Memory belongs to the contact, not to any one conversation. When memory is on, Bland keeps facts, a rolling summary, open items such as *Call back Friday*, entities, and a sliding window of recent messages for each contact.

A contact reached over SMS, iMessage, or voice keeps one shared history, and facts, summaries, and open items carry across channels. Bland offers native iMessage agents on Enterprise. Bland's [pathways](https://docs.bland.ai/tutorials/pathways), [Flex Mode](https://docs.bland.ai/tutorials/flex-mode), and [memory](https://docs.bland.ai/tutorials/memories) pages hold the full mechanics.

## Does Bland's conversational AI handle inbound and outbound calls?

Yes, Bland's conversational AI handles both: its agents answer inbound calls and place outbound calls, and Bland is not only for outbound. The same pathways, personas, tools, memory, and analytics power every channel, so a flow built for voice can also drive text over SMS, RCS, and iMessage without a rebuild.

MyPlanAdvocate (healthcare) shows what the inbound side looks like: $40M added in five months, with one Bland agent qualifying every inbound call. On the outbound side, American Way Health (insurance) shows $430M added annual revenue from automated outbound qualification. Bland has 250+ enterprise customers.

Another customer result:

Idaho Housing and Finance Association saves $750,000 a year by replacing its IVR with Bland, which handles 4,000 inbound calls a day and cut average call time by 20%.

> "We save thousands of dollars a day."

Greg Blake, Chief Information Officer, Idaho Housing and Finance Association

## Can your team build conversational AI agents without engineers?

Yes. Norm, Bland's AI assistant, builds voice agents using natural language descriptions, and Bland's dashboard includes a visual builder for non-technical users. You describe what you want in plain English, and Norm builds, edits, tests, and debugs your agents. When a call goes wrong, you flag it, hit Norm Fix, and Norm reproduces the failure, applies a fix, and verifies it against the original call. Read more about [Norm](/norm).

Salesforce, HubSpot, Cal.com, Calendly, Notion, and Slack connect from Bland's dashboard. Anything else connects through Bland's REST API and webhooks, which Bland's forward-deployed engineers can help with on Enterprise. A new agent can be tested with Talk in browser or Call me, and nothing in that first test needs a phone number, a plan, or a purchase.

## How do enterprises govern and secure conversational AI on Bland?

Every agent has development, staging, and production environments, and live calls run the pinned production version. When you edit a pathway you are working on a draft, and live calls are not affected until you publish it. Evaluations score an agent's calls, and a passing run can be required before a version is promoted.

Two further tools check the call logic itself. Bland's node-level regression test framework, called Standards, is built to catch behavior regressions if a prompt changes. Bland's testbed takes one turn from a real conversation, lets you change what the agent was told, and runs it again to see whether the answer changes.

Every enterprise customer runs on its own isolated infrastructure, run by Bland: dedicated compute, containers, and resources, with releases the customer chooses when to adopt and canary deployments on a share of real calls. Dedicated GPUs run by Bland are available. Bland carries a 99.9% uptime SLA.

On compliance, Bland is SOC 2 Type II, HIPAA compliant with a BAA, GDPR compliant with a DPA, and PCI DSS v4.0.1 self-attested. Bland holds FedRAMP 20x Class C (Moderate), certified 17 November 2026, FedRAMP Marketplace FR2628647242.

Bland's Trust Portal holds its SOC 2 Type II report, a penetration test summary, its Data Processing Agreement, and its Business Associate Agreement, and security teams can file a security review request or ask for customer references. Start at [Bland's trust and security page](/trust-security).

If your contact center runs on Five9 or NICE CXone, Bland works alongside it. Bland is an accredited Five9 AI Agent Connect partner and a NICE DEVone technology partner.

## What does conversational AI on Bland cost?

Bland's [pricing](/pricing) is $0.14 a minute on the self-serve Start plan. Enterprise pricing is custom, set with Bland's sales team. Phone is billed separately, on a customer's own carrier or Bland's at pass-through cost.

## How does conversational AI on Bland compare with an IVR, a chatbot, or a do-it-yourself stack?

An IVR or scripted chatbot follows a fixed script, and a do-it-yourself stack joins separate speech, model, and voice vendors that your own engineers connect and maintain. Bland's conversational AI sits between them: it is built and edited without code, and it still gives the call structure. For the IVR case in depth, read [AI voice agent vs IVR](/blog/conversational-ai-to-replace-ivr-phone-trees).

| | Conversational AI on Bland | Scripted IVR or chatbot | Do-it-yourself stack |
|---|---|---|---|
| Who builds it | Your team, in plain English with Norm or in Bland's visual builder; Enterprise adds Bland's forward-deployed engineers | A team that writes the menu or script | Engineers who connect and maintain each vendor |
| How a step stays on its goal | A node's condition keeps the agent on the step until it has what it needs | The script's fixed path | Dialogue logic your team writes |
| An off-topic question | A global node answers it, then the agent returns to the step it was on | Falls outside the script | Depends on logic your team builds |
| Memory across calls and texts | One shared history across voice, SMS, and iMessage when memory is on | Only what the script stores | Whatever your team stores |
| Models | A closed-source model Bland developed itself, with no third-party model providers | Fixed scripts and rules | Third-party models your team chooses |
| Inbound and outbound | Both | Set by the system it is built on | Set by the telephony your team connects |
| Governance | Development, staging, and production environments, evaluations, and isolated enterprise infrastructure | Set by whoever runs the system | Built and audited by your team |

<details><summary>What these rows mean</summary>

- **Who builds it:** Bland's column follows its label and the Norm documentation; the other two columns describe the category in general terms.
- **How a step stays on its goal:** Bland's column describes a node condition as Bland's pathways documentation states it.
- **An off-topic question:** Bland's column describes a global node as Bland's pathways documentation states it.
- **Memory across calls and texts:** Bland's column applies when memory is turned on for an agent.
- **Models:** Bland's column is Bland's own statement about its models.
- **Inbound and outbound:** Bland answers inbound calls and places outbound calls.
- **Governance:** Bland's column lists its environments, evaluations, and enterprise isolation as Bland documents them.

</details>

## Which industry guides to conversational AI does Bland publish?

- [Conversational AI for airlines: Rebooking, baggage, and status calls](/blog/conversational-ai-for-airlines)
- [Conversational AI for BPO teams: AI phone agents for each client's calls](/blog/conversational-ai-for-bpo)
- [Conversational AI for dealerships: Service booking and sales follow-up calls](/blog/conversational-ai-for-dealerships)
- [Conversational AI for hiring: Applicant screening and interview scheduling calls](/blog/conversational-ai-for-hiring)
- [Conversational AI for marketing: Inbound lead calls and outbound follow-up](/blog/conversational-ai-for-marketing)
- [Conversational AI for real estate: Lead, showing, and tenant calls](/blog/conversational-ai-for-real-estate)
- [Conversational AI for retail: Order, return, and store calls](/blog/conversational-ai-for-retail)

## Frequently asked questions

### What is conversational AI?

Conversational AI is software that understands what a person says or types and answers in natural language, keeping track of the conversation so far. On a phone call it combines speech recognition, a language model, a dialogue manager, and text to speech.

### What is a conversational AI platform?

A conversational AI platform is the software that lets a team build, run, and manage agents that hold conversations. On a phone, that means a way to design the call flow, connect tools and systems, test the agent, and put it on live calls.

Bland's agents answer inbound calls and place outbound calls, and your team builds them in plain English with Norm or in Bland's visual builder. For a ranked comparison, read about the [best conversational AI platforms](/blog/best-conversational-ai).

### Is conversational AI the same as a chatbot?

No. A scripted chatbot follows a fixed set of replies, while conversational AI works out what a person means and answers in natural language, keeping track of what was said so far. Some chatbots use conversational AI, and scripted ones do not.

### Can conversational AI hold a long phone conversation?

It can when the agent has a structure to follow. Bland's agents build a call from nodes that hold a step until its condition is met, answer an off-topic question through a global node and return to the step, let Flex Mode backtrack or jump ahead, wait through a Wait for Response node when a caller needs a moment, and keep memory per contact when memory is on.

### Is Bland only for outbound calls?

No. Bland's agents answer inbound calls and place outbound calls, and on Enterprise the same pathways run text channels.

### Does Bland have a no-code option?

Yes. Norm, Bland's AI assistant, builds voice agents using natural language descriptions, and Bland's dashboard includes a visual builder for non-technical users.

### What models does Bland use?

Bland self-hosts its models and uses no third-party model providers. Bland operates a closed-source model it developed itself.

### How was this page put together?

This page draws on Bland's documentation as captured on 20 November 2026 and on Bland's fact file. It describes how Bland's features work as those sources state them. No calls were tested or timed for it.
