2. Fabric Match & Sampling Engine

Matches a buyer's swatch, spec or reference to the closest developed quality in the archive, so sampling rounds collapse from three to one. Target: Sampling rounds per enquiry down from three to one.

Core buildPrototypeNext

Business impact

85/100

Sampling rounds per order

Build effort

50/100

Customer-facing

Projected lift

+21%vs pre-agent baseline

Sampling rounds per order

Time to value

2 quarters

From kickoff

Interactions / month

3,612+21%

Agent-handled

Automation rate

68%-4% human touches

No human in loop

Net value / year

₹1.4Cr

Revenue + cost avoided − run cost

Payback

9 months

Cumulative net positive

Adoption curve

Agent-handled volume vs pre-agent baseline

Jul86 peakDec

Baseline (manual process)

JulAugSepOctNovDec

Readiness scorecard

Six delivery dimensions

DataModelIntegrationAdoptionComplianceOps readiness

Outcome funnel

How sampling rounds per order is produced

  • Eligible interactions4,810
  • Agent-handled3,21067%
  • Qualified outcome1,76055%
  • Sampling rounds per order won83047%

Usage by geography

34
-53
33
76
-32
USAEUBangladeshVietnamIndia domestic

Cost mix

Share of run cost

50
  • Inference50%
  • Integration22%
  • Data prep10%
  • Change mgmt9%

Cohort retention

Share of users still using the product weekly

W1W2W3W4W6W8
Cohort 11059990796462
Cohort 21079486756451
Cohort 31079684726050
Cohort 41029489777255
Cohort 51059383786558

Performance by geography

Indexed 0–100 across five signals

VolumeAccuracyDeflectionCSATRevenue
USA9695656395
EU8169529069
Bangladesh3789677299
Vietnam7063376298

Model quality

Live vs target

MetricCurrentTarget30-dayStatus
Precision9290+3.1On target
Recall7485+4.4Closing gap
Grounded-answer rate8595+1.8Closing gap
Escalation rate56-2.2Closing gap
P95 latency (s)23-0.4Closing gap
Hallucination flags / 1k-21-0.9Closing gap

Service levels

85%

Impact score

50%

Delivery ease

96%

Uptime (30 days)

Unit economics

₹ per year, indexed to pilot scale

  • Revenue influenced₹3.9Cr
  • Cost avoided₹36L
  • Run cost₹17L
  • Net value₹1.4Cr

Cumulative ROI by quarter

Q1136% peakQ4

Channel mix

  • Voice31%
  • WhatsApp33%
  • Web-10%
  • Field app0%
  • Email46%

Language mix

3
  • English-16%
  • Hindi-4%
  • Gujarati2%

Segment performance

Volume, automation and win rate per market

MarketInteractionsAutomationSampling rounds per orderValueTrend
USA214
42%
-5%₹-44L
t0t1t2t3t4t5t6t7
EU307
48%
6%₹-12L
t0t1t2t3t4t5t6t7
Bangladesh353
31%
-2%₹4L
t0t1t2t3t4t5t6t7
Vietnam376
23%
-6%₹-28L
t0t1t2t3t4t5t6t7
India domestic388
39%
6%₹-4L
t0t1t2t3t4t5t6t7

Experiment log

What we tested on this product

ExperimentStatusResultOwner
Prompt v3 vs v2Won+11% sampling rounds per orderAiera Labs
Voice-first vs chat-firstRunning+0% so farGrowth Pod
Retrieval window 8k → 32kWon-1% 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

  • Fabric archive / PLM

    Construction, finish and test data per quality

    Live

  • Lab systems (Datacolor)

    Colour measurement and shade approval

    In build

  • Sample dispatch / courier APIs

    Sample status visible to the buyer

    Live

Product brief

Who uses it

Buyers, product development, merchants

Metric it moves

Sampling rounds per order

Why Arvind wins

Decades of developed qualities that competitors would have to re-develop

Spectral + image matchingFabric archive embeddingsSample tracking

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 sampling rounds per order 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

Spectral + image matching indexes the 3 Arvind product sets, pricing rules and policy language so nothing is answered from memory alone.

03

Reason and decide

Fabric archive embeddings scores the situation and picks the next best action — answer, quote, schedule, escalate — with a confidence threshold behind it.

04

Act in the workflow

Sample tracking 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 sampling rounds per order 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

    Buyers, product development, merchants hits the moment this product exists for — sampling rounds per enquiry down from three to one

  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 sampling rounds per order is attributed to this product.

Metric tree

What we steer and what we protect

North star

Sampling rounds per order

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

    0.5 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 market licence + usage
Break-even
6 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: Sampling rounds per order 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 sampling rounds per order get compared against?
← Back to portfolio