The Iris blog

Field notes on calm support.

Playbooks and essays on AI-first customer support — for stores, startups, agencies, and creators who answer their own customers. Written by the team building Iris.

GuideJuly 6, 2026·6 min read

The best live chat software for a website in India (2026)

An honest buyer’s shortlist for Indian small businesses: the free-forever option, the polished paid widgets, the heavyweight, and the India-built desks — judged on the three things a “best of” list never checks: rupee billing, the help center and AI behind the bubble, and whether “free” stays free as you grow. We make one of these, and we’ll flag it plainly.

Read the essay →

More posts

GuideJuly 6, 2026

Free live chat software: what you actually get (and where it ends)

A clear-eyed guide to free live chat — what Tawk.to, Crisp, Tidio and Freshchat give away, the five places “free” quietly runs out, and how to tell a genuinely free plan from a free trial in disguise. Yes, we have a free plan too, and we’ll be honest about its edges.

GuideJuly 6, 2026

Zendesk alternatives for Indian teams (2026)

Zendesk is powerful and priced for teams with an ops person to run it. For an Indian small business the honest alternatives are fewer than the internet claims — Freshdesk, Zoho Desk, Intercom, and yes, Iris — compared on published prices, the AI resolution meter, GST invoicing, and who actually understands rupee billing.

GuideJuly 6, 2026

AI chatbot for customer support: what actually works in 2026

A practical guide for small businesses: the difference between a chatbot that helps and one that hurts, why a grounded AI beats a scripted flow, what share of your questions an AI can genuinely close, and the one design choice — silence when unsure — that separates a trustworthy support AI from a liability.

DataJuly 6, 2026

The customer support metrics that actually matter (and the ones that lie)

First response time, CSAT, resolution rate, deflection, re-contact — the five numbers worth tracking, honest benchmarks for each, and why every one of them can lie unless you read it next to a second number. A field guide to measuring support without fooling yourself.

PlaybookJuly 6, 2026

How to reduce your customer support workload (without hiring)

A practical playbook for cutting support volume before it reaches a person: the help-center-first loop, a grounded AI that absorbs the FAQ tail, proactive messages that stop tickets being born, and the honest math on how much of a typical queue can simply stop landing on a human.

DataJuly 6, 2026

The real math of Intercom + Fin pricing (worked examples)

Intercom bills you twice — once per human seat, once per AI resolution — and the second meter is the one that surprises people. Worked out end to end for three teams: a solo founder at ~300 conversations a month, a 5-person team at ~2,000, a 15-person team at ~8,000. True monthly bills at published prices, plus the mechanics almost everyone misses.

GuideJuly 6, 2026

Intercom alternatives for small teams in India (2026)

An honest field guide for Indian founders: seven support tools compared on the things that actually bite here — ₹ vs $ billing, UPI, GST invoicing, WhatsApp-adjacent workflows, Hindi and Hinglish support, and data residency. Global heavyweights, regional players, and yes, Iris — each with its real trade-offs, ours included.

DataJuly 6, 2026

What an AI support answer actually costs in 2026

The public math, computed at list prices you can verify: the token anatomy of one real support answer — system prompt, knowledge context, history in; a short reply out — priced across a frontier model, a value model, and a small one. Why a single answer runs from under ₹1 to a few ₹, why caching changes everything, and why 99-cent retail sits so far above raw model cost.

DataJuly 6, 2026

We analyzed tens of thousands of support conversations. Here’s what customers actually ask.

The real distribution of a support queue: about a third orders and access, a quarter how-do-I, a fifth refunds and billing, the rest a long tail. What AI closes alone, what needs a human, and the uncomfortable truth that most questions are the same thirty questions rephrased.

ResearchJuly 6, 2026

The 2 AM problem: what happens to your customers while you sleep

For a solo founder or creator, the most expensive support ticket of the week arrives at 2am with a payment question. Nobody is awake. Here is what that silence actually costs — the abandoned purchase, the refund request, the one-star review — the response-time math behind it, and the honest menu of ways to fix it.

PlaybookJuly 6, 2026

50 copy-paste support replies that actually calm angry customers

A real, usable library organized by situation — refund demands, late delivery, "this doesn’t work," "I want a human," price complaints, feature requests, escalations, and the review ask — each in a warm-professional voice, each with a one-line note on why it works. Plus a short Hinglish section. Bookmark it.

GuideJuly 6, 2026

How to write help articles an AI can actually answer from

The RAG-era rewrite guide: one question per article, the answer in the first sentence, exact numbers and policies stated not implied, no "see above," tables over prose for plans and pricing — and why vague articles make an AI hallucinate. With a before/after rewrite and a 10-point checklist.

DataJuly 6, 2026

What a support ticket actually costs you (the math nobody does)

Worked numbers for three businesses — a solo creator at ~300 tickets a month, a 3-person team at ~1,500, a 10-person team at ~8,000. Loaded cost per resolution in ₹ and $, the hidden costs founders never count, and the deflection math: what absorbing 40–60% of volume is actually worth every month. Plus the formula to compute your own number.

ProductJuly 5, 2026

Raise your agent

Most support AI is a switch you flip. Ours is closer to a teammate you train: give it a voice, exact answers, private knowledge and hard limits — then rehearse every change on your own past conversations before customers feel it. And if forms aren’t your thing, just tell it what to change.

ProductJuly 4, 2026

Your support AI can now act

Answering from articles is table stakes. Today Iris AI can look things up in your systems — orders, bookings, account state — through tools you define in a form or bring via MCP. Safely: verified identity, allowlisted responses, human-approved changes.

Customer zeroJuly 4, 2026

The day we turned off Intercom

This morning, every SuperProfile creator — a hundred-million-request-a-day platform — got Iris as their support bubble. Intercom and our old India chat widget went dark behind a comment. Here is what the first hour looked like.

ProductJuly 4, 2026

A room of your own

The inbox is where a support team lives for eight hours a day — so we built a library of 200 rooms, 8 materials and 4 desks, a mixer for making your own, and a Brand Studio that designs rooms from your logo and website.

EngineeringJuly 4, 2026

Why Iris feels instant

The local-first inbox: what WhatsApp Web can teach you about what NOT to store locally, the bounded cache Iris keeps on your device, delta polling, drafts that survive reloads — and the list of things we deliberately never cache.

MigrationsJuly 3, 2026

Leave Intercom without losing your help center

The importer walkthrough — token, preview, drafts, publish — plus an honest accounting of what carries over, what you rebuild, and why everything lands as a draft on purpose.

EssaysJuly 3, 2026

AI-first support means silent when unsure

Most support bots are optimized to always say something. That is the bug. The case for confidence-gating, grounding, and treating unanswered questions as the product.

SaaSJune 29, 2026

Ship a help center + AI agent before your seed round closes

A day-one support stack for a two-founder startup: one snippet, twenty articles, and deflection math that defers your first support hire by a year.

Support opsJune 24, 2026

The boring discipline that makes a great inbox

SLAs you can actually keep, routing that routes, macros that do the paperwork, snooze that isn’t a hiding place — and the three metrics worth reading.

AgenciesJune 19, 2026

Run support for every client — under their brand, not yours

The whitelabel playbook: a workspace per client, their domain on email and help center, your team behind the curtain — and how to price it into the retainer.

CreatorsJune 15, 2026

Your audience asks the same 12 questions. Answer them once.

Course access, payment confusion, refund policy — the dozen questions every creator answers daily, and how to make each answer the last one you type.

SMB & D2CJune 10, 2026

How a 3-person store answers support at 2am (without hiring)

The seed-a-help-center loop that turns WISMO, refunds, and COD questions into answers that send themselves — and a gaps list that writes your docs roadmap.