Have you heard of Zanus AI for AI voice agents? If you are evaluating outbound calling platforms, you will likely encounter it alongside others. This article is an independent buyer's guide from Sysevo. Sysevo is not affiliated with Zanus, and you should confirm all current details directly with the vendor.
The term "AI voice agent" covers a specific subset of technology: software that places or receives calls, understands the caller's intent from speech, makes decisions based on that understanding, and logs the outcome to a system of record. Not all platforms that use voice cost the same, integrate the same way, or solve the same problems. Before you commit budget, you need to understand what you are buying, where the gaps are, and which questions will get you the honest answers.
What an AI Voice Agent Actually Does
An AI voice agent is not a recording or a simple IVR menu. It listens in real time, recognises intent from natural speech, and adapts its responses based on what it hears. When a prospect calls about a car rental, the agent hears the intent (availability inquiry), checks a calendar or booking system, provides available slots, and moves toward a booking or a handoff to a human. If the caller is angry about a previous reservation, the agent detects tone and frustration, retrieves the booking from a database, and either resolves the issue or escalates it.
The mechanism is speech recognition fed into a language model, which generates a response, which is then synthesised back into speech. That loop repeats until a termination condition is met (booking confirmed, human transfer initiated, or call ended). For this to work in a business context, the agent must write outcomes somewhere. Most platforms integrate with a CRM, database, or webhook endpoint. The data written depends on what was captured during the call: caller identity, intent, resolution status, and next steps.
The difference between a mediocre deployment and a useful one is captured in the details. Does the platform write data in real time or only after the call ends? Can it retrieve context from your CRM mid-call, or does it start each conversation cold? Does it handle silences, overlapping speech, or accents reliably? How long is the latency between the caller speaking and the agent responding? A 2-second delay becomes obvious and frustrating; most professional deployments aim for under 1 second. These are the questions that separate a working system from one that frustrates customers and wastes your time.
Have You Heard of Zanus AI for AI Voice Agents in Outbound Campaigns
Outbound calling is where most AI voice agents are deployed. A business uploads a list of contacts (prospect names, phone numbers, and attributes), the platform dials them, and the AI handles the conversation. Typical use cases include appointment setting for home services (HVAC, plumbing, solar), debt collection follow-up, customer satisfaction surveys, and reactivation campaigns for subscription businesses. In these scenarios, the agent's job is narrower than inbound reception: make the call, deliver a scripted introduction, qualify the lead or gather the required information, and log the outcome.
The math on outbound work is clearer than inbound. If your business normally makes 50 outbound calls a day with two full-time people, and each call takes 6 minutes average, you are consuming 10 hours of labour. An AI agent can make the same calls in about 2 hours of parallel execution (dialling multiple prospects simultaneously), writing results to your CRM in real time. The outcome is not that you eliminate the two people; it is that they stop making calls and start working the qualified leads the AI has already sorted into buckets.
Pricing for outbound platforms typically ranges from £0.03 to £0.15 per minute of call time, though some vendors charge per call attempt rather than per minute. A campaign that dials 200 prospects, with an average call duration of 90 seconds, would consume 300 minutes and cost £9 to £45 per batch, depending on the vendor and plan. Committed volume contracts can reduce per-minute rates by 20 to 40 percent. The real cost driver is not the per-minute rate; it is the volume you dial and the false-positive rate of your list, since the platform charges for calls to wrong numbers, voicemails, and disconnects.
How Voice AI Design and AI Personality Affect Outcomes
Two platforms with identical technical capabilities can produce vastly different results because of how they handle voice AI design and personality. A voice that sounds robotic, speaks too fast, or pauses at unnatural moments creates listener discomfort. The caller may hang up immediately, or they may answer every question with monosyllables, resisting engagement. A voice that sounds natural, pauses slightly before responding, and adapts its pace to the caller's speech patterns gets longer conversations and higher qualification rates.
Branded voice AI is a feature offered by some platforms: the ability to select or create a voice that matches your brand identity. A financial services company might choose a professional, measured tone; a fitness app might choose an upbeat, motivational voice. Some vendors offer pre-built voices (often 10 to 50 options); others allow custom voice creation using voice cloning, where a few minutes of recorded speech from a real person is synthesised into a consistent AI voice. The trade-off is that custom voices cost more (typically £200 to £1,000 per voice) and require longer setup time.
The agent's behaviour under pressure also depends on design choices. When a caller is confused or angry, does the agent repeat itself robotically, or does it acknowledge the confusion and offer alternative phrasing? When a caller interrupts, can the agent handle it, or does it pause and lose the thread? When there is background noise, does the agent ask for clarification or make incorrect guesses? These capabilities are not guaranteed just because you bought an AI platform. They are the result of careful prompt engineering and testing, and they vary significantly between vendors.
Where Voice AI Platforms Struggle and When They Are the Wrong Choice
AI voice agents are not universal problem-solvers, and honest evaluation requires understanding their limits. Complex troubleshooting conversations (IT support, medical advice, legal questions) are outside the safe zone. When a customer's issue is novel or multi-step, and a wrong answer creates liability or frustration, a human is still required. Most voice AI platforms are built for high-volume, narrow-scope interactions: appointment booking, simple qualification, survey collection, and status updates. If your business's core work is complex consultation, a voice agent is an inefficient first step and may damage your brand.
Accent and speech pattern recognition remains inconsistent across platforms. Non-native English speakers, regional accents, and speech disabilities are handled reliably by some systems and poorly by others. Before committing, test the platform's speech recognition against recordings of your actual customer base. If 20 percent of your inbound calls are from non-English speakers, and the platform misunderstands half of those, the system will frustrate both your customers and your team. This is verifiable during a trial, and it should be a kill criterion if the platform fails.
Integration friction is underestimated as a cost. A platform that promises CRM integration but requires manual API key setup, does not auto-map fields, or requires a developer to validate data flow will sit in your environment creating friction for weeks. During your evaluation, ask the vendor for a written integration specification, and ask for the name of someone in your industry who has implemented it. Call that reference and ask specifically about integration time and ongoing maintenance. If the reference says three months, budget for it. If they say they gave up, that is your answer.
Testing a Platform Trial and Measuring What Matters
A trial should not be a demo. A demo shows you what the vendor wants you to see; a trial shows you what the platform does when you use it unsupervised. Structure your trial in three phases: technical integration, outbound campaign, and measurement. Technical integration means connecting the platform to your CRM or database and verifying that data flows in real time and in the format you expected. If it takes longer than four hours, or if you need a vendor engineer to complete it, the integration is too hard for your team to maintain independently.
Run a small outbound campaign (50 to 100 dials) during week two of your trial. Use real contact data and real scripts from your business, not the vendor's sample scripts. Have the platform dial at your normal pace and time of day. Log outcomes into your CRM and run a quick report at the end. You are measuring three things: how many calls were completed versus dropped, what percentage of conversations reached a qualification decision, and whether the logged outcomes are accurate and actionable. If fewer than 60 percent of calls complete, or if only 30 percent of calls result in a clear next step, the platform is not a fit for your use case.
Measurement during live use is harder but more important. Pick a specific metric before the trial: maybe it is qualified leads per 100 dials, or average call duration, or callers who agreed to a follow-up. Compare that metric to your current baseline (human agents doing the same task, or a competing platform if you are comparing two). A good AI voice agent should outperform a human on speed and consistency, but it will rarely beat a human on complex problem-solving or relationship-building. Know what you are optimising for, and measure whether the platform achieves it.
Security, Compliance, and Data Storage
Any platform that records calls or stores caller information needs to be evaluated on security and compliance. Start with the vendor's public security page. Look for SOC 2 Type II certification, which indicates annual third-party audits of access controls, encryption, and incident response. Not all vendors publish this; if it is not listed, ask them in writing and get their answer in writing. The same applies to data storage: where are calls recorded, where is the transcript stored, and who has access? If the vendor's data centre is outside your regulatory region (GDPR in the EU, CCPA in California, HIPAA for healthcare), you need explicit confirmation that they comply with your jurisdiction's requirements.
Call recording retention policies vary. Some platforms delete recordings after 30 days; others retain them indefinitely unless you request deletion. For outbound calling, shorter retention is usually fine and reduces your data liability. For inbound customer service, longer retention may be useful for training and dispute resolution. During your evaluation, confirm the default retention period and whether it is configurable. If you are in a regulated industry (healthcare, financial services, law), confirm in writing that the platform can support your retention and deletion policies, and ask for a copy of their Data Processing Agreement.
Caller consent is a legal requirement in many jurisdictions. Some platforms include a consent prompt in their call flow, while others leave it to you. If you are responsible for consent, you must document it. Ask the vendor how they recommend handling TCPA compliance (in the US) or GDPR consent (in the EU), and whether they provide templates or guidance. Do not assume the platform is compliant just because it can record audio; compliance is a business and legal process, not a technology feature.
Comparing Platforms and the Questions to Ask in Writing
When you are ready to compare options, do not rely on sales conversations. Send a standardised questionnaire to each vendor in writing, and require written answers. Include questions about pricing (per-minute rates, setup fees, minimum commitments), integration (supported CRMs, API documentation, integration support cost), and support (response time for technical issues, whether support is included or charged separately). Written answers create accountability and give you a document to reference if circumstances change or promises are not kept.
Ask each vendor for a list of customers in your industry, and insist on speaking to at least two. Do not accept the sales team's curated references; ask for the full customer list and pick one yourself. Ask that reference: How long did integration take? What was not included in the sales pitch? Would you use this vendor again? How responsive is support when you have a problem? These conversations are where you will discover the platform's true cost of ownership, which is often higher than the per-minute rate suggests.
Request a written service level agreement (SLA) that specifies uptime guarantees, support response times, and remedies if the platform fails. If the vendor does not offer one, that is a signal. Ask for CRM integration documentation and ask whether the vendor will support custom integrations if your CRM is not on their supported list. Request a test environment where you can run campaigns with real data before going live. A vendor that refuses any of these is asking you to trust without verification, which is a risk you should not take.
Implementation Timeline and True Total Cost
Implementing a voice AI platform is not a one-week project. Budget 4 to 8 weeks from contract signature to first live campaign. Weeks one and two cover security review, vendor setup, and CRM integration. Week three involves testing call flow logic and voice personality, and getting internal stakeholder sign-off on how the agent will represent your business. Weeks four to six cover small pilot campaigns, measurement, and tuning. Week seven is for full deployment and staff training. If your integration is complex or your CRM is custom-built, add another two to four weeks.
Total cost of ownership includes more than per-minute call charges. Budget for setup and integration (often £1,000 to £5,000 depending on complexity), monthly platform fees (typically £300 to £2,000 depending on call volume and included features), and your internal labour for campaign setup, results review, and ongoing tuning. A business making 1,000 outbound calls per month at £0.10 per minute (assuming 90-second average duration) would spend roughly £1,500 on call charges, plus platform fees, plus internal labour. Over 12 months, that is £20,000 to £30,000 all-in. If that business was previously paying £3,000 per month in outbound calling labour, the AI investment pays for itself in six months.
One option is Sysevo's tiered plans, which combine voice calling with a built-in CRM, eliminating the integration step. Sysevo charges per call minute and a flat monthly fee; there are no separate CRM fees or integration costs. Other platforms charge separately for calling, CRM, and integration support. Evaluate based on your actual usage and your tolerance for integration complexity. A business that needs fast time-to-value and minimal technical overhead may prefer an all-in-one platform. A business with an existing CRM investment may prefer a calling-only platform and integrate it themselves.
Frequently Asked Questions
What is the difference between inbound and outbound AI voice agents?
Inbound agents answer calls from customers; outbound agents initiate calls to prospects or existing customers. Inbound requires live availability and context retrieval from your systems. Outbound requires list management and campaign scheduling. Most platforms support both, but they are tuned differently and priced differently. Outbound campaigns typically cost less per minute because they are scheduled and parallelised.
Can an AI voice agent replace my receptionist?
Partially. An AI agent can handle call triage, voicemail transcription, and appointment booking without human involvement. But customer service calls involving complex problems, billing disputes, or complaints still need human judgment. Most effective deployments use AI to filter and prioritise calls, then route qualified or escalated calls to humans. This makes receptionists more productive, not redundant.
How accurate is speech recognition for accented English or regional dialects?
Accuracy varies by platform. Most modern systems handle standard English dialects well but struggle with heavy accents, non-native speakers, or low-quality audio. Test the platform with audio samples from your actual customer base before committing. If your customers are geographically diverse or multilingual, accuracy may be a blocker.
What happens if the AI voice agent does not understand what the caller is saying?
Platforms handle this differently. Some repeat the question or offer a menu of options. Others escalate to a human agent. Some simply end the call. During your trial, test this scenario explicitly. Listen to recordings of failed interactions. If the agent's recovery is poor, customers will hang up and call back, which defeats the purpose.
Do I need to re-record my existing scripts for an AI voice agent?
You need to adapt your scripts. Written scripts for humans (with conversational hesitations and asides) do not work well with AI. Scripts for AI should be direct, avoid ambiguous pronouns, and include alternative phrasings for key questions. Most vendors provide script templates. Budget time for testing and refinement with actual AI voice before going live with real prospects.
Can an AI voice agent handle multiple languages?
Most platforms support multiple languages, but quality varies. English is universally supported. Spanish, French, and German are widely supported. Less common languages may not be available or may have lower accuracy. Check the vendor's language list explicitly. Switching languages mid-call is rare and usually requires separate agents.
How do I ensure GDPR or TCPA compliance when using an AI voice agent?
Compliance is not automatic. You must have caller consent (GDPR/CCPA), and you must manage contact lists to exclude opted-out numbers (TCPA). Work with the vendor's legal team to document your compliance process. Get a Data Processing Agreement in writing. Do not assume the platform handles compliance just because it can record calls.
Independent buyer's guide published by Sysevo. Sysevo is not affiliated with, endorsed by, or partnered with Zanus, and Zanus is the trademark of its owner. Product details change often, so confirm anything that matters to your decision with the vendor directly before you buy.
