White-Label Agency Services
Bots that deflect the boring 40% — and know when to stop
A good bot resolves order-status, FAQs and qualification instantly, then hands humans a transcript instead of a furious customer. We build that kind: scoped honestly, trained on your content, measured on deflection.
Overview
Chatbot development at Aadhya
We build three kinds of bots: AI support assistants trained on your docs and macros (RAG on GPT-class models) for web chat; WhatsApp Business API bots for commerce, notifications and support in WhatsApp-first markets; and structured lead-qualification flows that book meetings straight into calendars. Built, integrated into your stack, and maintained — because an unmaintained bot decays into a complaint generator.
The honest design rule: bots take the deflectable 30–50% (status checks, FAQs, routing, qualification) and escalate the rest with full context. Our own chat teams work behind the bots we build, so we live with our own designs.
At a glance
- Types
- AI support (RAG), WhatsApp API, lead-qual flows
- Stack
- GPT-class models, your helpdesk, your CRM, WhatsApp BSP
- Timeline
- 2–4 weeks per build
- Measured on
- Deflection rate + CSAT, not message counts
- Maintenance
- Monthly retraining + transcript reviews
- Pricing
- Fixed build + monthly care plan
What you get
Built for accountability, not activity
- Honest scoping first: transcript analysis decides what the bot should handle; anything below ~85% projected accuracy stays human.
- RAG on your real content: docs, macros and policy pages indexed, with source citations in answers and hallucination guardrails.
- Escalation with context: handoffs deliver the transcript, the customer data and the bot’s best guess — no "please repeat your issue".
- WhatsApp done properly: BSP setup, template approvals, opt-in flows, catalogue/commerce messages and support routing under Meta’s rules.
- Deflection dashboards: resolved-by-bot rate, containment, CSAT-after-bot, escalation reasons — monthly, with retraining actions.
How it starts
- Scope
- Transcript/FAQ analysis → target intents and a projected deflection rate you approve.
- Build
- 2–4 weeks: flows, RAG index, integrations, tone pass.
- Pilot
- Two weeks shadow/partial traffic; failure review before full cutover.
- Maintain
- Monthly: transcript reviews, index refresh, new-intent additions.
Pricing shape
How this service is priced
Exact rates live on the pricing page — published, because serious buyers filter on it.
- Fixed build
- Scoped bot with integrations, one price.
- Care plan
- Monthly retraining, reviews and improvements.
- Bot + human desk
- Build plus our chat team behind it — one accountable vendor.
Fit check
Built for some teams. Wrong for others.
Honest scoping saves both sides a month. This desk fits when:
- Support teams drowning in the same twenty questions
- WhatsApp-first markets needing commerce and notification flows
- Sites leaking after-hours leads a qualification bot would catch
Probably the wrong desk if: Replacing support entirely — deflection has honest ceilings; One-shot builds nobody will maintain (decay is measurable within a quarter); Grey-market bulk messaging dressed as automation — declined on principle.
The Aadhya way
The best bot compliment is silence: customers get answers and never think about the machinery. That requires restraint in scoping (bots do the deflectable tier, humans keep the judgment), citations against hallucination, and maintenance as a scheduled habit — because a bot is software pretending to be staff, and staff get reviewed.
Questions buyers ask
Lead-qual flow bots from ~$1,200; RAG support assistants typically $2,500–$6,000 depending on integrations; WhatsApp commerce bots quoted per flow set. Care plans from $250/month. Fixed quotes after scoping — see pricing.
GPT-class APIs (OpenAI/Anthropic) with retrieval over your indexed content — the model answers only from your sources, cites them, and defers to humans off-script. Model choice follows your data-residency and cost constraints, not fashion.
Guardrails specifically against it: retrieval-only answering, confidence thresholds, restricted topics, and "I’ll get a human" as the trained fallback. Pilot-phase failure reviews catch the residue before full traffic does.
30–50% of inbound volume for typical support desks (status, FAQs, routing), higher for transactional WhatsApp flows. We project a number during scoping from your actual transcripts and are measured against it — not against vendor-slide fantasies of 80%.
Yes: Zendesk, Freshdesk, Intercom, Gorgias handoffs with full context; HubSpot/Salesforce lead-writes; Shopify order lookups; calendar booking (Calendly/HubSpot Meetings) inside the conversation.
Automation on the official WhatsApp API (via a BSP): approved message templates, opt-in management, commerce catalogues, order updates and support flows — compliant with Meta’s rules, unlike gray-market "bulk WhatsApp" tooling, which we neither build nor recommend.
2–4 weeks from scoping sign-off for most builds; WhatsApp template approvals add Meta’s review time (days, occasionally weeks). Pilots run two weeks before full cutover.
Ideally us, on the care plan: monthly transcript reviews, index refreshes as your docs change, new intents added, drift caught. Unmaintained bots decay measurably within a quarter — if you insist on self-maintenance, we hand over runbooks.
Yes — qualification questions, routing rules, and calendar booking in-flow. Paired with the appointment-setting desk, the bot catches the after-hours leads your SDRs call back in the morning.
Yes — multilingual flows (English, Hindi, Hinglish) are standard for India-market WhatsApp builds, with language detection routing and templates approved per language. For other languages we scope honestly against model quality in that language.
Both, in sequence: humans first if you have no support baseline (bots automate patterns, and you need patterns to learn from), bot second to deflect the repetitive tier. We sell both, so the recommendation follows your transcripts, not our margin.
Related services
White-Label Agency Services
Agencies die of two things: turning down work they cannot staff, and margins eaten by staffing it.
SEO Content Writing
The internet does not need more words; it needs pages that answer a search better than the current top result.
White-Label PPC
Ad accounts rot from neglect: stale negatives, drifting budgets, ads nobody tested this quarter.
Next step
Start with a pilot, not a contract.
Describe the queue, the list or the workload. You get a written pilot plan and a fixed quote within 48 hours — and the pilot itself proves us before you commit to anything longer.