Early access — Chem IRL's own support runs on it first

Support email that answers itself, with your hand on the wheel.

Connect your Gmail or Outlook support mailbox. The agent reads each message against your knowledge base, drafts a reply with a confidence score, and routes it to your review queue. Auto-send only when you turn it on, above a threshold you set.

support@yourbrand.com
Read-only
  • Maria Chen

    2m

    Where is my order #A-2841?

    Hi, I ordered a few days ago and haven't received tracking yet…

    resolveddelighted
  • Devon Park

    14m

    Refund for damaged item

    The package arrived with a cracked screen. I'd like to return…

    escalatedfrustrated
  • Aiyana Ross

    31m

    Do you ship to Canada?

    Quick one — just want to check shipping options before I check…

    newneutral
  • Sam Okafor

    1h

    Still no response from your team

    This is the third time I'm reaching out about my subscription…

    unresolvedangry
  • Marcus Webb

    2h

    Re: Refund issued

    Thanks. Honestly though, four emails for a refund is a lot…

    resolvedfrustrated

Where is my order #A-2841?

Maria Chen · maria@hey.com · 2 messages

resolveddelighted
96%confidence
M

Maria

Today, 9:14 AM

Hi! I ordered a few days ago (#A-2841) and I still haven't received any tracking info. Could you let me know where it is?

C

Chem IRL Solutions

AIToday, 9:14 AM · auto-sent

Hi Maria — thanks for reaching out! Your order #A-2841 shipped yesterday via DHL Express and is currently in transit, expected to arrive by Thursday. Tracking: DHL-7741-22A. Let me know if anything looks off!

M

Maria

Today, 9:21 AM

Amazing, thank you! That was fast 🙏

Sources used: Order API, Shipping SOP v3
Auto-sent when confidence ≥ 85%Flagged for human review below thresholdEvery action is logged & reversible

Thread intelligence

Solving the issue isn't the same as keeping the customer.

Most support tools tag a thread as resolved or unresolvedand call it a day. Chem IRL Solutions tracks two independent axes — resolution (was the issue actually fixed?) and sentiment (how does the customer feel?) — so you catch the dangerous middle: the thread that's closed but the customer is on the way out.

newin progressescalatedresolvedunresolved
delightedsatisfiedneutralfrustratedangry
resolveddelighted
“Best night out in ages — Mini Golf was hilarious, food was unexpectedly great too.”

Issue answered, customer happy. Auto-close.

resolvedfrustratedWhy this matters
“OK, you sorted the refund. But four emails for a session that was clearly cut short? Probably won't book again.”

Technically resolved — but the customer is on the way out. Single-axis systems miss this entirely.

in progressneutral
“Got it, sending the photos over now.”

Mid-conversation, no emotional signal. No action needed.

unresolvedangry
“This is the SECOND time. I'm disputing the charge.”

Auto-escalated to a human within seconds. Don't let an AI try to fix this.

Resolution × sentiment, at a glance.

Every thread maps to exactly one cell. Filter your inbox by any combination — including the ones single-axis systems miss.

Last 7 days · 4,812 threads

delightedsatisfiedneutralfrustratedangry
new·1%12%2%1%
in progress·1%8%1%·
escalated··1%2%2%
resolved28%22%13%6%1%
unresolved···1%2%

The cell most ops dashboards never surface.

Roll it up into a real metric.

Tags become a churn-prevention signal. Track the share of threads that resolved but left the customer unhappy — usually the biggest gap between what your team thinks happened and how customers actually felt.

Resolution mix · last 7 days

7% resolved + unhappy

  • Resolved71%
  • In progress12%
  • Escalated9%
  • Unresolved5%
  • New3%

Confidence algorithm

We'd rather flag a draft than send a wrong answer.

Chem IRL Solutions scores every reply across five signals before it leaves your outbox. The score is transparent — you can see exactly why a draft was confident, or why it landed in the review queue.

  • Above your threshold — reply auto-sends and the thread is marked in progress.
  • Below the threshold — draft is queued for review with citations to the exact source passages it used.
  • No good source — Chem IRL Solutions refuses to guess. It pings a human and proposes adding the question to your SOP.

Composite confidence

87%

Threshold: 85% · auto-send

87%

Source coverage

weight 30% · 92

Did we find passages in your knowledge base that directly address the question?

Retrieval similarity

weight 25% · 88

How semantically close are the retrieved passages to the customer's wording?

Intent clarity

weight 20% · 84

Is the customer asking one clear thing, or three vague ones in a single email?

Historical resolution

weight 15% · 76

How often have similar past threads ended in resolved : happy?

Sentiment risk

weight 10% · 95

Is the customer already frustrated? Negative tone lowers confidence on purpose.

Voice of customer

Your inbox is your most valuable research tool.

Chem IRL Solutions reads every customer email, clusters recurring patterns, and turns them into actions you can ship — fixes for your website, your product, your SOPs, or your team. The same inbox that drains your support hours becomes your fastest source of business intelligence.

Common questionFrustrationEasy winProduct gapKnowledge gap
Common question58 mentions · +24%

Age policy after 6pm

By far the most-asked pre-booking question — customers don't realise the games go strictly 18+ after 6pm and turn up with kids.

Can we bring our 14-year-old for a 7pm Mini Golf slot? Couldn't see anywhere on the site.

Suggested action · Website

Make age policy a hard checkbox in the booking flow, not just a footer line

Could deflect ~46 tickets/mo · save ~6.1 hours/mo

Frustration21 mentions · +38%

Booking calendar shows slots that are actually full

21 customers booked online and were told on arrival the slot was double-booked. Driving angry sentiment and refund requests.

Drove an hour for our 8pm Shuffleboard slot and you said it was double-booked — your site let me book it.

Suggested action · Product

Open engineering ticket: investigate booking calendar overselling on shared bays

Could deflect ~17 tickets/mo · save ~4 hours/mo

Easy win33 mentions · +9%

Add 'How long is each game?' to product pages

Customers don't realise slots are 45–90 minutes depending on the game and email asking before booking.

How long does the Mini Golf actually take? Trying to plan a meal afterwards.

Suggested action · Website

Show slot length and group size on every game page above the booking button

Could deflect ~28 tickets/mo · save ~3.7 hours/mo

Product gap16 mentions · +50%

Daytime kids' birthday packages requested

16 enquiries this month for kids' birthday parties (12–15-year-olds) before 6pm. No package exists, agents reply ad hoc.

Looking to book a 13th birthday for ~10 kids on a Saturday afternoon — do you do a package?

Suggested action · Product

Build a 'Kids Party Package' (12–15s, 2 games + food bundle) for daytime slots

Demand signal — bring to your next product review.

124

Tickets / month deflectable

34h

Support hours saved

5

Categories of insight

Common questions become website FAQs. Frustrations become product tickets. Easy wins become weekend fixes. Product gaps become your next launch. Knowledge gaps make Chem IRL Solutions smarter automatically.

Explore the signals dashboard

Live demo with sample data — no signup required.

Integrations

Native in Gmail and Outlook. Sourced from everywhere.

Chem IRL Solutions sends from your real address, threads stay where your team already works, and your knowledge can live anywhere — websites, SOPs, docs, even shared folders.

Faster first response
12×
Auto-resolved threads
71%
Languages supported
30+
Setup time
< 10 min
  • Gmail

    Available now

  • Outlook

    Available now

  • Help Scout

    Coming soon

  • Zendesk

    Coming soon

  • Shopify

    Coming soon

  • Notion

    Available now

  • Google Drive

    Available now

  • Your website

    Available now