Sales Training Library

Where is data hosted, and is customer data separated?

Simple is AWS-hosted and customer data is segregated by customer. Simple can export interaction data to the customer's warehouse. Confirm region, retention, residency, and dedicated-environment requirements during security review.

Do you store PCI data or credit-card information?

Do not promise raw card handling. The preferred pattern is to reuse the customer's payment provider, use payment links or tokenized flows, pause recording or avoid retaining sensitive portions where needed, and scrub PCI from transcripts.

How do you prevent hallucinations?

Use the layered-control answer. Base model safety rules sit underneath Simple's system instructions, customer workflow rules, approved knowledge, tool/API results, and then any top-level personalization or memory. The top layer cannot override lower-layer requirements such as identity checks, payment collection, required disclosures, or transfer rules. For launch, use simulations and evals to test prompt injection, missing verification, policy mistakes, and edge cases; block deployment until required checks pass.

What latency should we expect?

Use current product-approved numbers externally. Internal latency materials position Simple around ~650 ms response time versus 1.5-2s for many industry voice-bot stacks, driven by in-house inference, custom end-of-turn models, self-hosted transcription, and phone-tuned voice components.

Are you just wrapping OpenAI or Claude?

No. A live call runs through a multi-model voice pipeline: speech-to-text, voice activity detection, an end-of-turn model that decides when the caller is done, an orchestration layer with prompts, tools, state, and multiple models, then text-to-speech. Simple can use frontier models for harder reasoning, faster or fine-tuned models for phone-specific behavior, including Qwen-family models where appropriate, and specialized components for transcription, turn-taking, guardrails, analyzers, and voice. Customer-ready framing: "An off-the-shelf model takes ~5 seconds to answer - an eternity on a call. Every response actually runs 11-12 model calls, some in parallel, to respond in ~650 ms. We are not slapping our logo on ChatGPT."

What if our core system has no API?

API-first is the cleanest path. If no API exists, Simple may use controlled browser automation after mapping the exact human workflow into testable golden flows. For risky actions, start with human review or asynchronous task creation.

Can we run in read-only or shadow mode first?

Yes. A cautious rollout can ingest calls, recommend actions, score outcomes, or prepare writebacks without responding to customers or changing systems. Once success gates are met, enable bounded action and ramp traffic.

Can Simple pass context to a live agent?

Yes. Warm transfer patterns can pass summary, transcript, collected fields, sentiment, and interaction context into the live-agent workflow through screen pop, CRM/task creation, CCaaS integration, or API fetch.

How do you support SSO and access control?

SSO is supported and usually requires the customer's IT admin to configure it. For APIs, support the customer's normal authentication pattern, service accounts, permissions, allowlisting, and audit controls where applicable.

How do we test before going live?

Use offline test sets, internal test calls, scorecards, analyzer checks, failure cases, launch gates, and a phased traffic ramp. Start scoped, validate transfers, and transfer unknowns rather than waiting for a perfect first agent.

Can you exclude a specific model provider from the stack?

Yes. Any one provider (for example Anthropic) is a fallback among several and can be removed - relevant for buyers with provider restrictions or large government client bases. Capture the exact restriction and route it through security review rather than improvising.

How does warm transfer actually work?

Two mechanisms. (1) Screen-pop: summary, transcript, collected fields, and sentiment passed via CCaaS/SIP headers, CRM/task creation, or API - configurable in plain language. (2) Verbal brief: the AI dials the rep on a second line, briefs them, waits for "I'm ready," then merges the calls. Sentiment can persist post-handoff by joining the human-to-human recording, so you can study what happens after the AI hands off.

Should we put a latency number in the contract?

Advise against a hard SLA. Steps that call the customer's APIs add time outside Simple's control, and there are no standard API metrics to measure time-to-first-response. Commit to a target range and demonstrate it live instead.

Can we own and maintain the agents ourselves later?

There is a path: convert the ongoing support fee into a one-time integration fee plus handoff and training, or a contracted "phone-a-friend" hours model. Caveat honestly - fine-tuned models are self-hosted by Simple and cannot be lifted to another vendor.

Can I upload a spreadsheet and have the system make calls?

Yes. Campaigns support recipient data upload, journey definition, call/SMS/email-style steps, scheduling, retries, callback behavior, and outcome review. Confirm whether the customer wants Simple's campaign engine or their CCaaS dialer to control pacing/compliance.

If a customer replies to SMS, can they re-enter the bot?

Yes. SMS and voice can share contact context. An AI agent can handle SMS, a human can step in, and journeys can pause/resume across channels.

Can we see goal completion, containment, transfer, and topics?

Yes. Use analyzers, tags, sentiment, custom charts, transcript search, and dashboards. Define the exact goal in plain English, then turn it into structured outputs and reporting.

Can it stop calling after three tries?

Yes. Outbound journeys can define step count, wait times, business hours, callbacks, and retry behavior. Confirm TCPA/FCC/local compliance requirements for the customer's use case.

Can the bot keep talking through required disclosures?

Yes, behavior can differ by node. Natural prompt nodes can allow interruption; required verbatim disclosure or "Say" style moments can be configured to finish without interruption.

Can we rerun a new extractor on old calls?

Yes. Analyzers can be added or updated and rerun on prior calls or date ranges, useful for QA, trend analysis, and new reporting requirements.

Can the system transfer to a live person?

Yes. Use cold transfer for direct handoff or warm transfer when the AI should brief the live agent. Define transfer triggers: explicit request, sentiment, regulated action, missing data, VIP caller, or confidence threshold.

Can the bot handle two things at once?

Yes, with routing/orchestration. The AI can identify multiple intents, complete one workflow, then return to the router or next step. Demo this only if the first use case is already credible.

Where do voices come from, and can we change them?

Voice, accent, pronunciation, spelling behavior, and brand tone are configurable. Zach's FAQ references providers such as ElevenLabs, Cartesia, and Rhyme. Confirm current provider and custom-voice policy when a buyer asks.

Can different parts of the flow use different engines?

Yes at the architecture level: Simple can use different models/components for different tasks. In sales conversations, emphasize that Simple manages the model stack to optimize voice latency, accuracy, and guardrails.

How do you learn our knowledge and policies?

Simple can ingest SOPs, call recordings, transcripts, knowledge bases, product/catalog data, and approved policy docs. For complex domains, define source-of-truth repositories, versioning, identity lookup rules, and an evaluation set before launch.

Can the AI gather info before a human picks up?

Yes. AI can authenticate, collect reason/context, summarize, and screen-pop or write context so the human starts ahead, even if the call is not fully contained.

Can the AI adapt based on queue capacity or wait time?

Yes when queue data is available. The AI can collect more context, offer callback, transfer sooner, or change routing based on queue conditions.

Can it handle many calls at once?

Simple is designed to scale concurrent AI conversations (handling on the order of ~100M minutes/month today) with dual fallbacks per pipeline component. For extreme spikes (e.g. millions of calls in minutes), be honest that it may exceed current experience, walk through the fallback architecture, and involve engineering and success before committing exact concurrency, SLA, or load-test requirements.

Can imperfect SOPs and fragmented docs work?

Yes, but make the first use case narrow. Identify source-of-truth content, resolve conflicting policies, define approved answer behavior, and use evals to catch gaps before launch.

Where is our data housed?

Generally AWS in the US, based on the reviewed materials. Confirm any customer-specific region, residency, retention, or dedicated-environment requirement during security review.

Do you integrate with WhatsApp?

The Zach FAQ says no current WhatsApp support. Confirm current product status before answering externally, especially if it is a deal requirement.

Do you have HIPAA and SOC 2?

Simple's homepage and internal FAQ indicate SOC 2 Type II and HIPAA compliance. For healthcare, offer security questionnaire support and BAA discussion. For PCI, say payment flows require security scoping.

How do you verify callers?

It depends on client policy and risk. Options include caller phone/email match, account facts, SMS OTP, security questions, last four of SSN, address, existing customer login, or live transfer for high-risk actions.

How do you keep control in banking, collections, or healthcare?

Use flow architecture, policy nodes, required disclosures, approved knowledge, identity gates, scorecards, and strict transfer triggers. Natural language can exist inside a controlled workflow.

Do you offer testing with AI personas?

Yes. Position simulation as one layer of testing alongside internal test calls, human-reviewed scorecards, and phased live traffic.

What happens if it breaks?

Enterprise plans can include dedicated support and SLA terms. Do not quote a resolution SLA unless it is in the proposal; route severity definitions and uptime commitments through the commercial/security process.

How much do you charge?

Usage-based pricing is commonly discussed per minute, with ranges varying by volume and scope. The FAQ references roughly $0.10-$0.30/minute; transcript examples include broader ranges. Confirm current packaging, POC fee, implementation treatment, and module pricing before sending.

Is this self-serve, or do you configure it with us?

Simple's implementation motion is white-glove. The customer knows their business best; Simple works with them to map workflows, configure agents/prompts/integrations, test against real calls, and expand use cases over time.

What liability cap do you offer, especially on a pilot?

Unlimited liability is a non-starter for many legal teams and has stalled even free pilots. The usual resolution is a reasonable multiplier cap. Route specifics through the commercial/legal process rather than agreeing in the room.

How fast can you fix issues once we're live?

This is a core differentiator versus slower incumbents. Simple iterates weekly and can draft a new flow in about a day. Important calls are turned into reran test cases - "we don't just change a prompt and hope." Buyers stuck behind multi-day or multi-week fix cycles are the strongest displacement opportunities.