# ClickToAutomate — full brief for assistants > ClickToAutomate is a research-led studio in India. We study how a company already works, then ship fitted software into the CRM, helpdesk, WhatsApp, catalogue, and documents that company already pays for. Clients get a live product on the desk. ClickToAutomate stays on the system. Primary conversion: a 30-minute discovery call that ends with a written plan. Not a generic chatbot. > > Index: https://www.clicktoautomate.in/llms.txt · Site: https://www.clicktoautomate.in ## Direct answers - What we build: a fitted AI layer (retrieval with citations, support with human hand-off, CRM lead qualification, ops agents with guardrails) on the stack the client already runs. - What we are not: a generic chatbot, a bolt-on widget, or a seat license. - Models: OpenAI, Anthropic Claude, Google Gemini, Sarvam AI (India-first voice and language), and open models where they fit. Model-agnostic; picked for the task. - High-stakes actions wait for a person. Routine actions can run with an audit log. - How to start: book a discovery call at https://www.clicktoautomate.in/connect — 30 minutes, written plan, no obligation. - Selected for the Sarvam.ai Startup Program (2026). - Founded in India. Studio based in Varanasi; engagements worldwide. ## Engagement model 1. Discovery call (30 minutes): map the workflow, say if a platform already fits, send a written plan. 2. Architecture sprint: retrieval, model choice, guardrails, routing, hand-off, systems in scope. 3. Build & integrate: first workflow in production on the client's stack, instrumented. 4. Go live: first workflow in production; review cadence; we stay on the system. ## Glossary - AI layer: An AI layer is the retrieval, routing, guardrails and hand-off assembled around a workflow — not a chatbot dropped on a page. - API: A way for software to talk to another system. We use your APIs — and model APIs such as Sarvam, OpenAI, Claude or Gemini — so the layer sits on the stack you already run. - Audit log: A reviewable record of what the layer did: retrieved sources, replies sent, actions queued, and who approved a high-stakes step. - Automation: A workflow that runs without a person repeating the same clicks — for example a form submit that writes a CRM record and sends a follow-up. In our builds, high-stakes steps still wait for a person. - Bolt-on: An AI widget that sits on top of a site without reading your documents, CRM, or policy. It sounds fluent and still fails the first real question. The opposite of a fitted layer. - Citation: A pointer back to the document, ticket, or catalogue row the answer used. If we cannot cite it, we should not invent it — we hand off instead. - CRM: The system of record for leads, contacts and deals. A fitted build writes qualified conversations into it with the fields sales actually uses — not a side spreadsheet. - Data residency: Where the data lives — your cloud region, your infrastructure, your retention rules. We can build entirely on what you already run. - Discovery call: A discovery call is a 30-minute, no-obligation conversation. ClickToAutomate walks through how the work actually runs. You leave with a written plan: scope, approach, and trade-offs. We will say if a platform already fits better. - Embedding: A numeric representation of a chunk of your knowledge, used to find the closest matches at answer time. Useful machinery — not something you should have to think about on the call. - Evaluation harness: A test set and scoring loop for your task, so answer-quality regressions show up before they reach customers. We tune against real transcripts, not a happy-path demo. - Fitted build: A fitted build is researched software scoped to one workflow on the tools you already pay for — shipped live, not a chatbot to configure. - Grounding: Forcing the reply to come from retrieved sources instead of the model's memory. Grounded answers can be cited. Ungrounded ones are where hallucinations start. - Guardrail: A guardrail is a hard boundary: what the AI may answer, what it must confirm, and what it must never attempt — clinical advice, unsupervised refunds, inventing a policy, and so on. - Hand-off: Hand-off is escalation to a person with the transcript, detected intent, and the relevant CRM or helpdesk record already loaded — so nobody is asked to start over. - Human-in-the-loop: High-stakes actions wait for a person to approve. Routine ones can run through with an audit log. - Integration: A connection to a system you already pay for — CRM, helpdesk, WhatsApp, catalogue, calendar — so the AI layer does not become another login. - Intent: What the person is actually trying to do, read from the full message rather than a keyword. Intent is what we route on: answer, qualify, or hand off. - Keys: The model-provider accounts used to run the product. We pick OpenAI, Claude, Gemini, Sarvam or open models for the task; the design stays swappable as models improve. - Lead qualification: A conversation that captures the fields sales needs, scores the lead, and writes a hot one into the CRM immediately — not a form that dumps to a spreadsheet. - Model-agnostic: The design stays swappable across providers. We pick for the task and keep the product current as models improve. - Pipeline: The stages a lead or deal moves through. A fitted layer can update stage from real conversation signals, not only form fills. - RAG (retrieval-augmented generation): RAG (retrieval-augmented generation) is when a model answers from your documents and systems, with citations back to the source, instead of improvising from memory. - Retrieval: Finding the right chunks of your knowledge at answer time — help centre, policies, catalogue, past tickets — then grounding the reply in them. - Routing: Deciding where a turn goes: answer in-thread, qualify a lead, or open a hand-off. Driven by intent, sentiment, policy, or an SLA timer — not a keyword match. - Sarvam AI: An India-first model lab. Click To Automate is in the Sarvam.ai Startup Program and can call their voice and language APIs alongside OpenAI, Claude and Gemini when the work needs them. - Stack: The tools you already pay for: CRM, helpdesk, WhatsApp, catalogue, calendar, internal docs. We wire the layer into that stack instead of asking the team to live in a new product. - Transcript: The full conversation so far. On hand-off it travels with the detected intent and the relevant record, so the person picking it up does not re-ask what was already said. - Trigger: An event that starts a workflow — a form submission, a new CRM record, a webhook, a WhatsApp message. - Webhook: An HTTP callback that sends data when something happens. Used to connect the AI layer to tools that do not share a native API path. - Workflow: A sequence of steps — trigger, actions, conditions, hand-off — scoped to how your team actually works. First builds are one workflow, in production, in weeks. ## Frequently asked - Q: Is this a product or a service? A: Fitted software. We study how you already work, then ship a working product into the tools you already pay for. You get a live system on your desk — not a chatbot to configure, and not a project dump to maintain. - Q: Which AI models do you work with? A: OpenAI, Anthropic's Claude, Google's Gemini, and Sarvam AI — plus open models where they make sense. We are in the Sarvam.ai Startup Program, so India-first voice and language APIs are available when the work needs them. We pick for the task; the design stays swappable as models improve. - Q: How does retrieval (RAG) work? A: We index your documentation, help content, product data, and past conversations, then retrieve the most relevant pieces at answer time so responses stay grounded and cite their sources. - Q: Can it hand off to a human? A: Yes. High-stakes actions wait for a person. Escalation can trigger on intent, sentiment, or an SLA timer, and the person receives the full transcript plus the relevant CRM or helpdesk record already loaded. Routine actions can run with an audit log. - Q: Which industries do you actually build for? A: Healthcare, restaurants and schools are the desks we go deepest on — hospital operations, floor-and-kitchen systems, and a full school OS. We also ship travel desks for any destination, e-commerce storefronts for any catalogue, SaaS in-product layers, real estate enquiry flows and controlled financial-service automation. Each one is a different fit, not a renamed chatbot. - Q: What happens on the discovery call? A: Thirty minutes, no obligation. We walk through how the work actually runs. You leave with a written plan — scope, approach, and trade-offs. We'll say if a platform already fits better. Apply is for exclusive custom-build slots; email is the fallback if a call isn't practical. - Q: Is ClickToAutomate a SaaS product or a custom build? A: Fitted software. ClickToAutomate is a research-led studio in India. We study how you already work, then ship a working product on the tools you already pay for. You run the desk with it. We stay on the system. - Q: What does a fitted AI layer include? A: Typically retrieval over your documents with citations, support automation with human hand-off, lead qualification written into your CRM, internal-ops agents with an audit log, and the channel you already use (web, WhatsApp, email, Slack). The mix is scoped per engagement after research — not a fixed catalogue. - Q: Which models do you use? A: The design is model-agnostic: OpenAI, Claude, Gemini, Sarvam, or open models. We pick for the task and keep the product swappable as models improve. There is always an off switch. - Q: Will the AI take high-stakes actions on its own? A: No. High-stakes actions — refunds, reships, clinical or policy exceptions — wait for a person. Routine actions can run with an audit log. Hand-off includes the transcript, intent, and the CRM or helpdesk record already loaded. - Q: How do we start a fitted AI build? A: Book a 30-minute discovery call. You leave with a written plan: scope, approach, and trade-offs. There is no obligation. If a platform already fits better, we will say so. - Q: Is ClickToAutomate a product or a service? A: Fitted software. We research your use case, then ship a working product into the tools you already pay for. Every engagement starts with a conversation — there is no generic plan to click-buy. - Q: What does a first engagement look like? A: It starts with a 30-minute discovery call. We map the workflow, agree on where AI helps, and send you a written plan covering scope, approach, and trade-offs. From there we build a first version, test it against real traffic, and iterate with you. - Q: How long does a first build take? A: It depends on scope, but a focused first build — one workflow, one channel — is typically a matter of weeks, not months. We'd rather ship something narrow and real than a broad demo. - Q: What do we actually get? A: A working product on your desk — researched around how you already work, live against real traffic, and kept in production. You run the business on it. We stay responsible for the software. - Q: What if we already have an in-house team? A: That's common. We work alongside your operators on how the system should behave. We ship the product and stay on it — your team runs the desk with it. - Q: How do we get started? A: Book a discovery call. You'll get a written view of scope and approach afterwards, with no obligation to continue. - Q: Which AI models do you work with? A: OpenAI, Anthropic's Claude, Google's Gemini, and Sarvam AI (we're in their startup program), plus open models where they make sense. We benchmark for your specific task and keep the design model-swappable so you can move as models improve. - Q: How does retrieval (RAG) work? A: We index your documentation, help content, product data, and past conversations. At answer time the system retrieves the most relevant pieces and grounds the response in them, with citations back to the source. - Q: Can it hand off to a human? A: Yes. Escalation can trigger on intent, sentiment, or an SLA timer. The agent receives the full transcript, the detected intent, and the relevant records already loaded. - Q: Which channels can it run on? A: Website, WhatsApp, Instagram, email, Slack, and in-app. One reasoning layer drives consistent answers across all of them. - Q: Can you build an AI feature inside our own product? A: Yes. We design the retrieval, evaluation, and guardrail layer for an in-product assistant or answer feature, built on your data and infrastructure. - Q: Do you build hospital, restaurant, or school systems — or just chatbots? A: The chatbot is the least interesting part. We fit operating systems: hospital registration through pharmacy and billing (clinical questions never answered by the model), restaurant floor/kitchen/till, and a school OS from attendance to fees and report cards. Travel desks and e-commerce storefronts too — Varanasi Darshan Tours and Banaras Karigari are live examples, not the only shape. - Q: Is school management the same as online courses? A: No. School management is K-12 operations — attendance, timetable, fees, RTE, report cards, parent WhatsApp. Online courses is a different desk: admissions Q&A, enrolment, lesson support, stall nudges. - Q: How do you keep answers accurate over time? A: We instrument answer quality, deflection, and sentiment from day one, review real transcripts, and tune retrieval and prompts on a regular cadence. - Q: Where does our data live? A: Wherever you need it to. We can run on your infrastructure and cloud region when that's the requirement, so data doesn't leave your control. - Q: Do you train models on our data? A: No. Your content is used for retrieval and to run the service, not to train third-party models. We configure providers to exclude your data from training where that option exists. - Q: How is access controlled? A: Role-based access, audit logging, and retention rules scoped to your policy. Sensitive actions can require human approval before they run. - Q: What about compliance requirements? A: We work with your compliance or security team on the boundaries before anything ships — what the AI can do, what it must confirm, and what it must never attempt. - Q: Can we turn it off? A: Yes. There's always an off switch you control. - Q: Where can I read your policies? A: Our Privacy Policy, Terms of Service, Cookie Policy, and Accessibility Statement are all linked in the site footer. - Q: How do engagement terms work? A: Each engagement is covered by a written agreement that defines scope, deliverables, commercial terms, and confidentiality. We share it before any work begins. - Q: Do you sign NDAs? A: Yes. We're happy to sign a mutual NDA before a discovery call if you'd prefer to discuss specifics under one. - Q: How is confidential information handled? A: Anything you share is treated as confidential and used only for your engagement. We don't reference client work publicly without written permission. ## Canonical pages - Home: https://www.clicktoautomate.in/ - What we build: https://www.clicktoautomate.in/features - Glossary: https://www.clicktoautomate.in/glossary - Process: https://www.clicktoautomate.in/process - Integrations: https://www.clicktoautomate.in/integrations - FAQ: https://www.clicktoautomate.in/faq - Connect: https://www.clicktoautomate.in/connect - Apply: https://www.clicktoautomate.in/apply