The honest answer depends on deal size and complexity. For SMB-targeting offers under roughly $5k ACV with a defined buying motion, our agents close deals end-to-end in production today. Above that, the right pattern is AI for qualification and booking, with humans for negotiation and close. We're explicit about which model fits your funnel during the discovery call so you don't pay for a feature you won't use.

We also recommend a 2-week 'agent vs human' bake-off during the pilot so you have real numbers, not vendor claims, before deciding. In these tests, 73% of Kallix customers found AI conversion rates within 10% of their best human SDR on equivalent lists — the other 27% saw the AI outperform.

These numbers come from anonymised production data across 80+ customers over the last 12 months. Three things drive variance more than anything else: list quality, offer-market fit, and the script's first 8 seconds. We tune the opener heavily during the pilot because it's where 60% of cold-call abandonment happens.

We don't promise specific numbers in marketing copy; we share benchmarks for your industry during discovery and write the success metric into the pilot agreement. Industry variance is significant: SaaS outbound typically lands at the lower end of the cold range (3–4%); real estate warm re-engagement regularly hits 25–30% because of the high-intent nature of portal enquiries.

Objection handling is the single biggest tuning area during pilot. We listen to real call recordings, tag every objection that occurred, build a response library with your sales lead, and stress-test it in roleplay sessions before the agent goes live.

The agent also learns. Every production call feeds back into a weekly tuning cycle where we measure which response variant performs best for which objection, and update the live agent's playbook. For the most common deflection — 'send me an email' — our best-performing response branch acknowledges the request, delivers a 30-second value hook, and offers two specific callback times rather than simply agreeing to email.

Compliance is the customer's responsibility — you own the list and the consent — but we make compliant operation the default path. The agent identifies itself as an AI when directly asked (never deceptive), includes call-recording disclosure in the opener for relevant regions, and stops outreach immediately when a caller asks to be removed; that removal propagates to your CRM in real time.

We publish a region-by-region compliance guide covering the US (TCPA, state-level mini-TCPAs), EU (GDPR), India (DPDP 2023, TRAI regulations), and our legal team reviews every custom script before go-live. For Indian customers, we hold a dedicated compliance walkthrough covering the Digital Personal Data Protection Act 2023 requirements.

We deliberately don't sell leads because the unit economics work better when you control list quality. Most customers use Apollo, ZoomInfo or in-house lists from their CRM. We help with list hygiene: deduplication, DNC scrubbing, timezone segmentation before the first call.

If you don't have a list, we can recommend list-building partners we've worked with for specific industries. For Indian customers, we support direct integration with lead exports from 99acres, MagicBricks and JustDial so portal enquiries can flow directly into outbound sequences without manual CSV handling.

Integrations are set up during onboarding week 1, not post-launch. We connect via native OAuth for HubSpot, Salesforce and Zoho, and via REST API for other platforms. The bidirectional sync means the agent enters each call informed — company, role, prior interactions, open opportunities — and exits each call having written structured output: disposition, qualification signals, next action, and lead score delta.

For CRMs not on the native list, we use Zapier or Make as a relay layer. We have pre-built relay integrations for Monday.com, Airtable, Notion (used as a lightweight CRM by many SMBs) and several industry-specific tools including property CRMs for real estate customers. The integration spec is reviewed and agreed in week 1.

Calendar integration eliminates the scheduling loop that kills conversion between the call and the actual meeting. The agent proposes two time slots based on live rep availability, the prospect confirms verbally, and the booking is created — all within the same call. No action required from the rep until they walk into the meeting.

We support round-robin distribution across multiple reps, territory-based routing (the Mumbai rep gets Maharashtra leads), and priority rules (senior AEs get high-score leads). Automated no-show reminders go out 24 hours and 1 hour before the meeting via SMS or WhatsApp. The reminder cadence alone recovers an average of 12% of booked meetings in production.

Voice naturalness has advanced significantly since 2023. We use a combination of ElevenLabs and proprietary fine-tuned voices depending on the use case and language. The two variables that matter most for perceived naturalness are latency (how fast the AI responds after the human stops speaking) and prosody (whether pauses, emphasis and pacing feel human). We target sub-400ms response latency, which falls within normal human conversational response time.

For outbound sales, a confident and direct voice outperforms an overly warm one — it matches the professional tone prospects expect from a B2B SDR call. You choose the voice persona during onboarding and can A/B test two voice variants simultaneously to find what resonates with your specific audience.

Setup speed depends on two variables: how quickly your team reviews and approves the script, and how complex your CRM integration is. HubSpot or Salesforce customers with a clear ICP and an existing call script go live at 2 weeks. Customers with custom CRM setups, multi-language requirements or a need to build the sales playbook from scratch land at 4 weeks.

Pilot calls start in week 3 on a small segment — typically 100–200 leads — so you have real performance data before full rollout. We don't consider setup complete until the success metric written into the pilot agreement has been validated. Compare this to a DIY build on Vapi or Bland: most engineering teams estimate 2–4 months to reach a production-grade agent.

Language support is especially critical for the Indian market where buyers frequently switch between Hindi and English mid-call — a pattern we call code-switching. Our Hindi and Hinglish models are trained on real Indian B2B call data, not translated from US English, which means they handle common Indian sales objections ('abhi budget nahi hai', 'sir please email karo') naturally and respond with culturally appropriate phrasing.

For international customers, we work with a partner network of voice actors and linguistic consultants to develop voices that are culturally accurate, not just grammatically correct. We are explicit in discovery about which languages are production-grade versus beta, and we won't deploy a language we can't support reliably.

Warm handoff is one of the features we engineered most carefully because a botched transfer kills the deal. The sequence: the agent detects a handoff trigger (configured by you during onboarding), places the prospect on a brief hold, dials the assigned rep, delivers the briefing — 'I have Priya from ABC Corp, interested in the Growth plan, main concern is Salesforce integration' — then bridges the call. The full transcript is already in the CRM before the rep picks up.

If no rep is available, the agent schedules a specific callback time rather than leaving the prospect in limbo. Transfer success rate — rep available and call connected within 45 seconds — is above 92% in production. Missed transfers automatically create a high-priority CRM task for the rep.

Transcription runs in parallel with the live call. Each transcript is automatically tagged with: call outcome (meeting booked, not interested, callback scheduled, DNC), key qualification signals, objections raised, and a 3–5 sentence AI-generated summary. This structured output writes directly to the CRM record, so reps have full call context without listening to a recording.

For data compliance, recordings are stored in encrypted AWS S3 — us-east-1 by default, with EU (eu-central-1) and India (ap-south-1) region options for GDPR and DPDP customers respectively. Call-recording disclosure is included in every opener script by default. You can delete individual recordings on demand to respond to data subject access requests.

This is the question we answer most carefully. For repetitive, high-volume top-of-funnel calls, the economics of AI are clear: if you're paying $60k/year for an SDR to dial 80 contacts a day, and the AI dials 800 contacts a day at $1,499/month, the cost comparison is straightforward. But most of our best customers don't use this as a replacement play.

They use the AI to expand the volume of qualified pipeline handed to closers, and redeploy SDRs into roles that require human judgment: high-value enterprise outreach, complex multi-stakeholder deals, relationships with named accounts. That model grows total revenue rather than just cutting cost. The better question isn't 'replace or not' — it's 'what is the highest-leverage work my humans should be doing?'

We structure the pilot around a specific, written success hypothesis: 'If the agent achieves X connect-to-meeting rate on Y leads in Z weeks, it earns a production rollout.' That metric is agreed before work begins so there's no ambiguity at the end — and no pressure to continue if the numbers don't hold up.

Week 1 is setup and alignment: ICP definition, objection library, integration wiring. Weeks 2–3 are live calling with daily reporting so you can see the funnel build in real time. Week 4 is performance-based tuning and a final review. The pilot fee is credited in full toward month 1 of production if you continue.

Performance correlates with three structural features: call volume (AI adds the most leverage when you have hundreds of contacts to reach daily), qualification clarity (the more objective your ICP criteria, the more accurately the AI qualifies), and a defined next step — booking a visit, scheduling a demo, generating a quote, not 'stay vaguely interested'.

Verticals that don't work well: high-touch enterprise sales where buying decisions are driven by relationships built over many months, and categories with prohibitive compliance burdens for AI calling. We're direct about fit in discovery — we decline engagements where we don't see a clear performance case, because a poorly-fit deployment hurts both parties.

Multi-touch follow-up is where the AI's consistency advantage is most visible. Human SDRs typically stop following up after 2–3 attempts; research shows the average B2B sale requires 8 or more touchpoints. The AI executes every touch in the sequence — without fatigue, reprioritisation or forgetting — and every touch is logged to the CRM.

Sequences use conditional branching: if the day-1 call results in 'call me Thursday', the Thursday call is automatically scheduled and a WhatsApp reminder goes out Wednesday evening. If it results in 'not interested', the contact moves to a 90-day re-engagement cadence rather than being dropped entirely. The full sequence is approved in the Kallix dashboard before any contact is made.

Inbound and outbound require different agent configurations. Inbound agents are optimised for qualification speed: the caller is already interested, so the goal is to capture intent and book the meeting as fast as possible — average handle time is 3–4 minutes. Outbound agents are optimised for opener effectiveness and patience: the contact was not expecting the call.

For most SMB customers, inbound ROI is faster and easier to prove — the immediate value is recovering the 35–50% of inbound calls that currently go to voicemail outside business hours. Outbound ROI builds more slowly but scales without a natural ceiling. We typically recommend starting with inbound if you have significant call volume, then layering outbound in month 2.

Each Kallix customer operates in an isolated data environment. Your call recordings, transcripts, contact data and CRM mappings are not accessible to other tenants, and they are not used as training data for models shared across customers — a common concern with AI vendors that use customer interactions to improve base models.

For Indian customers under the Digital Personal Data Protection Act 2023, we offer India-region data residency with all processing within AWS ap-south-1 (Mumbai). Our Data Processing Agreement is available before contract signature and covers the standard clauses required for DPDP compliance. We are targeting SOC 2 Type II certification in Q3 2026.

Personalisation depth scales directly with the richness of your CRM data. At minimum the agent uses name and company. With richer data — job role, company size, industry vertical, recent activity, prior objections — the opener becomes genuinely specific: 'I saw you downloaded our SDR ramp-time guide last Tuesday — that's exactly the problem I wanted to talk about.'

The key technical constraint is data freshness at call initiation: the agent reads the CRM record at the moment of dialling. For inbound use cases where the caller just submitted a form, we use a webhook trigger that passes the form payload directly to the agent before the first ring fires — so the agent opens with the exact context from the form submission.

Speed-to-lead is the single biggest driver of inbound conversion rate, and it is where AI has a structural advantage over any human team. An SDR team working 9-to-5 misses every lead that comes in outside business hours. The AI responds within the configured window at any hour.

Harvard Business Review research, replicated by InsideSales across 100,000+ leads, shows that responding within 5 minutes makes a prospect 9x more likely to engage versus a 30-minute response. Responding within 1 minute increases that likelihood to 22x. Our default configuration is a 60-second response window for inbound leads; this is configurable from 30 seconds to 24 hours depending on your use case and applicable calling regulations.

Concurrency is where AI has its most structurally obvious advantage over human teams. An SDR makes one call at a time; a Kallix agent fleet can work an entire list simultaneously. For time-sensitive campaigns — trade show follow-up, event registrant outreach, promotional windows — the ability to call everyone within a tight window is the critical capability.

Concurrency is configured at the campaign level. For cold calling, we recommend moderate concurrency aligned to expected answer rates to avoid telephony congestion. For inbound callback queues, full concurrency is appropriate — no caller waits. The Kallix dashboard shows live concurrent call count, active call durations, and queue depth in real time.

Sales intelligence is a first-class feature in Kallix, not an afterthought. Every call generates structured data: outcome, talk time, hold time, objections raised, qualification signals captured, call sentiment and final disposition. These roll up into campaign dashboards that identify which list segments, scripts and time slots are performing best.

The most actionable report is the objection frequency table — it surfaces which objections appeared most in production that week, enabling your sales lead to prioritise the tuning cycle. We also generate anomaly alerts: if a region shows a spike in DNC requests, or a script variant has an above-threshold early-hang-up rate, it appears on the dashboard the same day.

Knowledge ingestion happens in two forms. Structured data — product catalogue, pricing tiers, feature specifications, integration docs, case studies and FAQ content — loads into the agent's retrieval layer so it answers specific questions accurately. Procedural knowledge — how you open, qualify, handle objections, position against competitors and close for next steps — is encoded in the agent's conversational flow during script development.

The most valuable training input is call recordings from your best-performing human SDR. We transcribe them, identify what makes them effective, and encode those patterns. If recordings don't exist, we run a structured 2-hour interview with your top performer as a substitute. The knowledge base is updated continuously based on production call data — questions the agent could not answer accurately become knowledge base additions in the next weekly tuning cycle.

We are direct about failure modes because they determine whether Kallix is the right fit. The biggest single performance killer is a poorly defined ICP: if it is not clear who you are trying to reach and what a qualified conversation looks like, the AI optimises for the wrong signals. The second most common failure mode is list quality — a stale, mis-targeted list produces poor results whether a human or AI makes the calls, but AI makes the problem more visible by scaling it.

There are also category-level mismatches: if your sales motion is fundamentally relationship-driven over a 6-month cycle with no clear early qualification step, the AI has limited leverage. We identify these mismatches in discovery before a pilot is scoped and will not take on an engagement where we do not see a credible performance path.

The comparison is more nuanced than a single cost number. An SDR brings judgment, relationship presence and brand tone that AI does not yet fully replicate — particularly for high-value accounts where a genuine conversation matters. For high-volume, clearly defined top-of-funnel work, AI wins on economics, consistency and availability.

The frame most customers land on is not 'which one' but 'what is the right mix.' A common model: AI handles all cold outreach and warm follow-up volume previously done by 2 SDRs; those 2 SDRs are redeployed to higher-value pipeline work or upmarket accounts. Total pipeline grows; headcount cost is flat or lower. The productivity gain from redeploying humans to higher-leverage work is often as large as the AI cost saving.

Auto-dialers and power dialers solve the 'how do I dial faster' problem. They are telephony infrastructure. Once a call connects, the rep still qualifies, handles objections and books. Kallix is the conversational layer above telephony that removes the need for a rep on the call entirely.

The meaningful comparison is not 'AI agent vs dialler' — a dialler is a tool; Kallix is the SDR. We can integrate with telephony infrastructure you already have, or provide it end-to-end. If you are already using a power dialler and want to reduce rep time spent on repetitive qualification calls, Kallix sits naturally above it in the stack.

Brand consistency in outbound matters more than most teams expect. When the agent introduces itself as 'Hi, I'm Aryan from [Your Company]', the persona must match your brand tone — the same way you would brief a new hire on communication style. We define the agent's name, style and approach during onboarding, and it stays consistent across every single call.

Voice cloning uses a specific recorded voice with written consent from the voice donor and is subject to our acceptable use policy. In practice most customers find that our library voices — 40+ options across genders, accents and styles, tested extensively across thousands of calls — perform as well as or better than custom clones. We recommend A/B testing two voice styles in the first month to find the best fit for your specific audience.

WhatsApp follow-up is particularly high-performing for the Indian market, where message open rates exceed 90% compared to 15–20% for email. We configure a post-call WhatsApp template for each call disposition: 'Meeting booked' sends a calendar confirmation with a join link; 'Send collateral' delivers the relevant brochure or case study; 'Callback requested' sends a reminder 30 minutes before the scheduled time.

All WhatsApp messages go through the WhatsApp Business API registered to your brand's number — the prospect sees your company name, not an unknown number. We handle the WhatsApp Business API setup and Meta template approval as part of onboarding. Template changes go through a 24-hour Meta approval cycle, so we build and submit templates in week 1 to avoid delays.

India's B2B calling environment has specific characteristics that generic US-built AI platforms don't handle well: code-switching between Hindi and English mid-call, objection phrasing that requires culturally appropriate responses ('abhi budget nahi hai' needs a different response than the American 'not in the budget right now'), portal-specific lead structures from 99acres and MagicBricks, and a strong buyer preference for WhatsApp over email for follow-up communication.

Kallix has a dedicated India delivery team. Our Hindi/Hinglish voice models are trained on real Indian B2B call data, not transliterated from American English. Our DPDP compliance documentation is current, updated for the 2023 Act. India pricing is available in INR, and our India-market playbooks are built from live production deployments — not adapted from US templates.

The full cost picture: plan subscription plus per-minute overage if applicable, plus a one-time onboarding fee ($999 for Growth, waived for Scale and Enterprise). A team running 5,000 minutes/month on Growth with 3,000 overage minutes at $0.18/min has an all-in monthly cost of approximately $2,039 — compared to one US SDR at $5,000–$7,000/month fully loaded.

ROI calculation uses your own funnel numbers: if the AI books 30 meetings per month and your close rate is 20% at a $5,000 average deal size, the AI generates $30,000 in closed pipeline for roughly $2,000 in monthly cost. We run this exact model for your specific funnel during the discovery call, using your historical SDR data as the baseline so the comparison is credible, not illustrative.