5. Buyer Portal Agent

A grounded agent that answers order status, shipment, spec and reorder questions for global buyers around the clock, and raises the reorder before the buyer thinks to. Target: 70% of buyer status and spec queries answered without a merchant.

Quick winPilotNow

Business impact

84/100

Buyer queries auto-resolved

Build effort

34/100

Customer-facing

Projected lift

+21%vs pre-agent baseline

Buyer queries auto-resolved

Time to value

≤ 90 days

From kickoff

Interactions / month

3,948+16%

Agent-handled

Automation rate

73%-4% human touches

No human in loop

Net value / year

₹32L

Revenue + cost avoided − run cost

Payback

11 months

Cumulative net positive

Adoption curve

Agent-handled volume vs pre-agent baseline

Jul94 peakDec

Baseline (manual process)

JulAugSepOctNovDec

Readiness scorecard

Six delivery dimensions

DataModelIntegrationAdoptionComplianceOps readiness

Outcome funnel

How buyer queries auto-resolved is produced

  • Eligible interactions4,695
  • Agent-handled3,69579%
  • Qualified outcome1,54542%
  • Buyer queries auto-resolved won65542%

Usage by geography

-61
-101
-121
-131
-6
USAEUBangladeshVietnamIndia domestic

Cost mix

Share of run cost

34
  • Inference43%
  • Integration16%
  • Data prep15%
  • Change mgmt1%

Cohort retention

Share of users still using the product weekly

W1W2W3W4W6W8
Cohort 11059386797259
Cohort 21019480726353
Cohort 31059480716252
Cohort 41059583776656
Cohort 51079180726654

Performance by geography

Indexed 0–100 across five signals

VolumeAccuracyDeflectionCSATRevenue
USA43765710038
EU44100707245
Bangladesh6554449436
Vietnam7195725991

Model quality

Live vs target

MetricCurrentTarget30-dayStatus
Precision8590+3.1Closing gap
Recall7585+4.4Closing gap
Grounded-answer rate8495+1.8Closing gap
Escalation rate26-2.2Closing gap
P95 latency (s)03-0.4Closing gap
Hallucination flags / 1k-21-0.9Closing gap

Service levels

84%

Impact score

66%

Delivery ease

89%

Uptime (30 days)

Unit economics

₹ per year, indexed to pilot scale

  • Revenue influenced₹2.1Cr
  • Cost avoided₹-1L
  • Run cost₹3L
  • Net value₹32L

Cumulative ROI by quarter

Q187% peakQ4

Channel mix

  • Voice-75%
  • WhatsApp25%
  • Web50%
  • Field app50%
  • Email50%

Language mix

3
  • English7%
  • Hindi-8%
  • Gujarati-15%

Segment performance

Volume, automation and win rate per market

MarketInteractionsAutomationBuyer queries auto-resolvedValueTrend
USA99
54%
-3%₹9L
t0t1t2t3t4t5t6t7
EU-201
34%
7%₹14L
t0t1t2t3t4t5t6t7
Bangladesh99
44%
12%₹-23L
t0t1t2t3t4t5t6t7
Vietnam249
49%
15%₹-42L
t0t1t2t3t4t5t6t7
India domestic-126
52%
16%₹-51L
t0t1t2t3t4t5t6t7

Experiment log

What we tested on this product

ExperimentStatusResultOwner
Prompt v3 vs v2Won+6% buyer queries auto-resolvedAiera Labs
Voice-first vs chat-firstRunning+1% so farGrowth Pod
Retrieval window 8k → 32kWon-2% escalationsPlatform
Human review threshold 0.7Rolled backno liftOps
Vernacular tone packRunning+4 CSAT ptsCX

Risk register

Owned mitigations before scale

RiskSeverityMitigationOwner
Data freshness in source systemsMediumHourly sync + staleness alertsData Eng
Regulatory language in outputsHighBuyer NDACompliance
Field-team adoptionMediumIn-app nudges + weekly coaching drillsSales Ops
Model cost driftLowRouter to smaller models for routine intentsPlatform

Where it sits in the portfolio

Tap another dot to switch product

Quick winsBig bets
Impact ↑Effort →

Systems it touches

  • SAP SD

    Order, delivery, invoice and dispatch status

    Live

  • Freight forwarder APIs

    Container tracking and ETA changes

    In build

  • DeployOne Connect

    Voice and WhatsApp in buyer time zones

    Live

Product brief

Who uses it

Global buyers, merchants, customer service

Metric it moves

Buyer queries auto-resolved

Why Arvind wins

Answers from live ERP data, so it can be trusted with commitments

Grounded retrievalOrder-status toolsMultilingual voice + chat

Guardrails

Applied to every response

  • Buyer NDA
  • Certification claims (GOTS/BCI)
  • Export documentation accuracy

Delivery plan

14-week path to production

Weeks 1–3

Ground the data

Connect 3 product sets, historical conversations and CRM outcomes.

Weeks 4–6

Pilot with one team

Run in USA with human review on every output.

Weeks 7–10

Automate the loop

Hand buyer queries auto-resolved decisions to the agent with escalation thresholds.

Weeks 11–14

Scale across lines

Roll out to 3 languages and all geographies.

How it works

End-to-end, from signal to learned outcome

01

Capture the signal

Every Voice and WhatsApp interaction, plus CRM and order events, is streamed in and tagged to a customer, Denim and geography.

02

Ground the context

Grounded retrieval indexes the 3 Arvind product sets, pricing rules and policy language so nothing is answered from memory alone.

03

Reason and decide

Order-status tools scores the situation and picks the next best action — answer, quote, schedule, escalate — with a confidence threshold behind it.

04

Act in the workflow

Multilingual voice + chat writes back into CRM/ERP and fires the customer-facing action in English or Hindi, so the team sees it where they already work.

05

Learn from outcome

Won/lost and buyer queries auto-resolved outcomes flow back nightly; prompts, thresholds and routing are retuned every fortnight.

Data it runs on

Sources, contents and refresh cadence

  • Conversation history

    Voice, WhatsApp, Email-to-agent, Trade shows transcripts with intent, objection and sentiment labels.

    Streaming

  • Catalogue & pricing

    Denim · Wovens · Knits with specs, availability and approved commercial language.

    Hourly

  • CRM outcomes

    Stage moves, win/loss reasons and owner activity per enquiry.

    15 min

  • Operations systems

    ERP order, stock, dispatch and service records used to answer status questions.

    Hourly

  • Market signals

    Demand and competitor movement across 5 geographies via DeployOneSense.

    Daily

Customer & team journey

  1. 1Trigger

    Global buyers, merchants, customer service hits the moment this product exists for — 70% of buyer status and spec queries answered without a merchant

  2. 2Agent acts

    Aiera responds within seconds on Voice, grounded in the live catalogue and the customer's own history.

  3. 3Human in the loop

    Anything below the confidence threshold, or touching Buyer NDA, routes to a named owner with full context attached.

  4. 4Close the loop

    Outcome is written back, the customer gets a summary on WhatsApp, and buyer queries auto-resolved is attributed to this product.

Metric tree

What we steer and what we protect

North star

Buyer queries auto-resolved

Drivers

  • Agent-handled share of eligible interactions
  • First-response time and resolution time
  • Qualified outcomes per 100 conversations in USA
  • Repeat engagement within 30 days

Guardrail metrics

  • Grounded-answer rate above 95%
  • Escalation rate inside agreed band
  • Buyer NDA
  • Run cost per conversation below target

Squad to build it

Who is needed and for what

  • Product owner (Arvind)

    Scope, adoption and the business case

    0.3 FTE

  • Conversation designer

    Prompts, tone and 3-language scripts

    1 FTE

  • Data / integration engineer

    CRM, ERP and telephony wiring

    1 FTE

  • ML engineer

    Scoring models, evaluation harness, drift

    0.5 FTE

  • Ops champion per market

    Floor adoption across 5 geographies

    0.2 FTE each

Commercials

How the product pays for itself

Delivery model
Customer-facing
Cost drivers
Inference volume, telephony minutes, integration upkeep
Charge basis
Per handled conversation
Break-even
9 months at pilot volume
Scale unit
One geography × one Denim

Rollout waves

Scope and the exit test for each wave

Wave 1 — Prove

USA, Denim, human review on every output.

Exit test: Buyer queries auto-resolved beats manual baseline on 200+ conversations.

Wave 2 — Automate

Add EU and Hindi; agent acts without review above threshold.

Exit test: Escalation rate under 10% for four straight weeks.

Wave 3 — Extend

All 3 product sets and 4 channels, wired into CRM reporting.

Exit test: Adoption above 70% of eligible interactions.

Wave 4 — Compound

Roll across 5 geographies and share the model with sibling Arvind businesses.

Exit test: Net value positive at run-rate; owned by business-as-usual team.

Dependencies to unblock

Needed before wave 1 starts

  • API access to CRM and ERP for Arvind Textiles
  • Approved language pack for English, Hindi, Gujarati
  • Telephony numbers and consent records for outbound
  • Sign-off on Buyer NDA
  • Named market champions for floor adoption

Open questions for the business

Answer these to lock the scope

  • Which USA team runs the pilot and who signs off on outcomes?
  • How far back does usable conversation history go for Denim?
  • What confidence threshold is acceptable before the agent acts unattended?
  • Which existing report does buyer queries auto-resolved get compared against?
← Back to portfolio