The AI growth team, at a glance
Six teams, each with the live consoles its head works in. Everything below is clickable and nothing here is a slide. Figures are modelled from published Titan Company results and the 528 store network.
The argument in one line
Jewellery is bought against occasions, and occasions have dates. An instalment falls due, a scheme matures, a wedding is set, a repair is ready, an exchange window opens, points expire. Tanishq already holds the date on roughly 1.18 million customer events a month. Around 915,000 of them receive no outbound contact from anyone.
Event volumes are modelled from disclosed store count, scheme structure and published segment growth. Every input is listed under Method and limits.
Start here
Four stops, in order.
Teams
The platform organised the way Tanishq is, one view per head.
Platform map
Every module, the engine behind it, and the working console it opens.
Value model
What the pilot is worth, from disclosed figures, with every assumption shown.
Commercial terms
Flat retainer, usage as pass-through, and what is carried free until the month six review.
The three consoles to open first
Each is a working screen you can operate, not a picture of one.
Aiera Live
Pick an agent, place a call, watch the conversation run turn by turn in the customer's language with the outcome and quality record written as it goes.
The Golden Harvest book
The instalment lifecycle with its three leaks, a five step call ladder, and a recovery model you can move the inputs on.
Dormancy and reactivation
Five dormant segments across 1.9 million identities, with a segment builder that produces a dated call list.
What this build is
- The DeployOne platform configured for how Tanishq operates, running on the published store network and scheme structure.
- Operational volumes, contact rates and average bill values are modelled from comparable retail networks of similar store count, and are editable inputs on every console.
- Reported Titan and Tanishq figures are drawn from published results and are marked as such wherever they appear.
- No individual customer is scored, profiled or ranked anywhere in this system. Sentiment sits at store and issue level only.
Platform map
The whole system on one screen. Six lines of business, twenty modules, five engines, and the working console behind each module. Filter it, then open any screen directly.
Engines
Every module runs on one or more of these. The engine is what DeployOne operates; the module is what the Tanishq head sees.
The six teams
The platform organised the way the business is, rather than the way the software is. Each head opens one view and sees the modules they decide on, the modules they work with someone else, and the consoles behind both.
Aiera Live
Pick an agent, place the call, and watch the conversation run in the customer's own language. The agent captures structured fields as it talks, writes an outcome, and leaves a quality record behind. Nothing here is a recording; the transcript plays the way the live agent runs.
The conversation runs turn by turn.
Place a real call now
This is the live agent, not the simulation. Choose the desk, enter a name and an Indian mobile number, and the agent calls in a few seconds. It speaks English and Hinglish, and ends the call politely on its own.
Two desks, one build
Both agents run the production configuration: English with Hinglish switching, a 300 second call ceiling, voicemail detection, interruption handling, structured extraction and a summary written at the end. Only the opening line and the brief change between the sales desk and the support desk.
Calls are placed on a rented Indian number. Nothing is stored on this page; the record sits in the call log with the transcript, outcome and quality score.
What the call desk is doing underneath
The date decides the call
Every call is placed against a dated event that already exists in the Tanishq record: an instalment due, a maturity, an appointment, a repair ready, an expiry. No list is dialled without a reason attached to it.
Structured capture, not notes
The agent fills named fields as the conversation moves. Commitment dates, occasions, store preference, need-by dates and objections land as data, not as free text a supervisor has to read.
Outcome, record, handover
An outcome is written, the recording and transcript are stored and searchable, and anything requiring a person is passed to a named store advisor with a clock on it.
Governance held on every call
- Recording disclosure is given on every call, in the language of the call.
- No price is quoted on a specific piece. Pricing is a store advisor conversation and the agent says so.
- No purchase pressure, no repeated calling after a stated refusal, no calls outside permitted hours.
- Consent state is checked before dialling and honoured immediately when withdrawn on the call.
- No caste, religion or community field exists anywhere in the record.
The Golden Harvest book
The savings scheme is the one part of the business where Tanishq already knows the date, the amount and the customer. It is also where value leaks quietly, because a missed instalment and an unredeemed maturity both look like nothing happening.
The call ladder
Five contacts across the life of one account. Each has a date, a purpose and a permitted outcome.
Seven days before the instalment
A short reminder on WhatsApp with the amount and date. No call yet. Cost is close to nothing and it clears most of the volume.
Two days after a missed instalment
A voice call in the customer's language. The only job is to separate cash timing from lost intent, and to set a commitment date if it is timing.
Thirty days before maturity
A call explaining what maturity unlocks and offering an appointment. This is where redemption value is set, well before the customer walks in.
Day 120 after maturity
A reminder that the account has a closing date, with the remaining days stated plainly and an appointment offered again.
Day 300 after maturity
The last call, escalated to a named store advisor if unanswered twice. Beyond this the account is at risk of closing unused.
Recovery model
Move any input. Everything recalculates. Every figure below is an editable input, seeded from published Titan results and comparable retail benchmarks.
- A recovered lapse is counted once, at instalment value, not at the full scheme value.
- Redemption uplift is applied only to accounts that take a guided appointment, not to all maturities.
- No value is claimed for enrolments the scheme would have won anyway.
- Calls are costed separately as usage and sit outside this figure.
The appointment desk
Store advisors currently answer the phone between customers who are standing in front of them. The desk takes that load, books against real store capacity, confirms twice, and hands the advisor a prepared appointment rather than an interruption.
Inbound demand shown by shape and mix, as it arrives from the store network.
At booking
Slot, store, advisor and directions sent on WhatsApp in the language of the call.
Two days before
A single confirmation message with a one tap reschedule rather than a cancellation.
Morning of
A short call if unconfirmed, which is where most of the no-show reduction is found.
After a no-show
One rebooking attempt within forty eight hours, then the record closes. No repeated chasing.
Dormancy and reactivation
A dormant customer is not a lost one. Most have simply passed out of an occasion window and nobody has given them a reason to come back that is not a discount. The desk builds a segment, attaches a dated reason to each identity, and produces a call list a store can actually work.
| Identity | Home store | Dormant | Reason attached | Language | Window |
|---|
Customer names are masked. Stores, ticket sizes and ageing follow the live network mix.
The reason is not a discount
A purity check entitlement, an exchange window, an expiring point balance or an anniversary all give the customer a reason to walk in that costs Tanishq nothing in margin.
One attempt, then stop
A dormant identity is called once per reason and once only. A refusal closes the record for that reason permanently, and consent state is checked before every dial.
Conversation intelligence
This is the part that is difficult to buy anywhere else. Every conversation the platform has produces structured evidence about demand: what people asked for, what they hesitated over, which occasion they were buying against, and which store they nearly chose instead. That evidence goes back into targeting, into assortment and into what the next call says.
What last month's conversations said
Themes extracted from conversation volume, at issue level only. No individual is scored.
Movement is measured against the prior four week window on the same store set.
This is why the fee is a platform fee
Per minute pricing values the call. What is actually valuable is the record the calls leave behind, which compounds every month and which no media agency or call centre can hand over.
Three destinations
Performance media targeting, so spend follows stated intent rather than inferred lookalikes. Regional assortment, so stores stock against asked-for designs. And the call scripts themselves, which are rewritten monthly against what worked.
Gold rate trigger desk
Gold rate movement is the single most reliable demand trigger in this business and it is currently worked by broadcast. The desk turns a rate move into a set of dated, named, reachable cohorts and puts a specific message against each one.
| Cohort | Identities | Channel | What the message says |
|---|
- The agent states a rate movement as fact and never forecasts a direction.
- No urgency language is used beyond the stated closing date of an entitlement.
- A customer who has declined rate-triggered contact is removed from all four cohorts permanently.
Account basis
What this pilot covers, what it does not, and the figures it is built on. Everything downstream of this page depends on the scope stated here.
Reported figures from published Titan Company results. Operational volumes, contact rates and conversion assumptions elsewhere are modelled from comparable retail networks and shown as editable inputs.
Scope of the pilot
Tanishq, India
The 528 store Tanishq network, the Golden Harvest scheme book, the Encircle identity as it applies to Tanishq customers, and inbound and outbound contact across voice and WhatsApp in fourteen languages.
The other Titan businesses
Mia, Zoya, CaratLane, beYon, Watches, EyeCare, Taneira, Fragrances, Damas and TEAL each have a scoped module set already drafted. They sit outside this pilot and outside this fee until Tanishq is proven.
Why Tanishq first
Golden Harvest gives a clean number
A lapsed instalment either resumes or it does not. It is the rare case where the effect of a call can be measured without an attribution argument.
The largest event base in the group
More dated customer events sit inside Tanishq than in all the other businesses together, which means the platform proves itself fastest here.
What is built here carries
The identity spine, the language coverage and the governance model are the same across every other Titan business. Tanishq pays for the build the others inherit.
Value model
Seven value lines, each measured at jewellery segment EBIT margin rather than at billed value, and each carrying an attribution haircut for the share that would have happened anyway. The point of this page is that the assumptions are visible and can be argued down.
| Value line | Module | Gross billed value, ₹ Cr | At EBIT margin | After attribution |
|---|---|---|---|---|
| Recognised annual value | ₹0 Cr | |||
- No value is taken at billed revenue. Every line is reduced to segment EBIT before it is counted.
- No value is claimed for footfall, brand awareness, or anything that cannot be traced to a dated event and a call record.
- The advisor hours line is counted at forty per cent realisation, because time returned to the floor is not automatically time converted.
- Usage cost for calls and messages sits outside this model entirely and is charged as pass-through.
- Nothing here is a guarantee. It is a model whose inputs are visible so that Tanishq can set them.
Required inputs
What the platform needs from Tanishq to move to a live pilot, and what can wait. Nothing on this list requires a data warehouse project.
| Input | Needed by | Form it can take | Blocks what if absent |
|---|---|---|---|
| Golden Harvest account bookAccount, instalment schedule, status, maturity date, home store | Week one | Flat file export is sufficient. No integration needed to start. | Modules 05 to 07 entirely |
| Store masterStore, city, state, opening hours, advisor count, appointment capacity | Week one | Spreadsheet | Appointment booking against real capacity |
| Consent and contactability statePer identity, with the date and source of consent | Week one | Flat file, refreshed weekly | All outbound contact. This is a hard gate. |
| Encircle identity and tierRegistered number, tier, point balance, expiry date | Week three | Flat file to start, API later | Modules 08 to 11 |
| Service and repair job statusJob, store, stage, ready date | Week three | Flat file or direct read | Module 16, collection calling |
| Inbound number routingAbility to point a store or central number at the platform | Week three | Telecom change, no system change | Inbound answering and the appointment desk |
| Purchase historyIdentity, date, category, value, store | Week six | Flat file | Dormancy segmentation quality, not the module itself |
| WhatsApp business senderVerified sender on the Tanishq brand | Week six | Existing sender can be used | Message ladder. Voice runs without it. |
What DeployOne provides
- Agent configuration, voice build and language tuning across the fourteen store languages.
- Script authorship for every call type, reviewed with Tanishq before a single live call is placed.
- The full console estate, hosted, with named access per head.
- A named delivery contact and a weekly working session through the pilot.
- Recordings, transcripts, outcomes and quality scores stored and searchable from day one.
Commercial terms
One fee, flat across the network, for a six month initial term, billed monthly in advance. Usage sits outside it as pure pass-through. Nothing is priced on transaction value, on leads, or on jewellery sold.
Store contact centre
- Inbound answering across voice, WhatsApp and website chat for the store network
- Appointment booking against real store capacity, with the confirmation ladder
- Recordings, transcripts, summaries and outcomes stored and searchable
- Escalation paths, service level clocks and live scorecards
- One named point of contact and continuous tuning
End-to-end AI transformation
- Everything in the store contact centre
- The full twenty module estate across all six teams
- Outbound campaign design and script ownership
- Conversation intelligence feeding media, assortment and scripts
- Continuous transformation, optimisation and delivery support
The store contact centre retainer is ₹6,00,000 a month. End-to-end AI transformation adds ₹24,00,000, taking the total to ₹30,00,000 a month. A one-time setup fee of ₹15,00,000 is payable at kick-off and covers console build, agent configuration across fourteen languages, knowledge base, integration, testing and store training.
A flat fee is approved once
A share of revenue or of recovered value makes a fixed cost platform more expensive because the business grew, which reopens the conversation every quarter and is difficult for a listed company to budget. A flat number is approved once and defended easily. If the measured value holds, the number can move at the month six review against something proven rather than something forecast.
Pass-through at cost
Voice minutes, WhatsApp conversation charges and telecom costs are billed at cost with the underlying statement attached. DeployOne takes no margin on usage, which removes any incentive to place calls that do not need placing.
What sits inside the retainer
Three lines, each with a measure that moves inside the first month
- Retail and stores, modules 01 to 04
- Schemes and savings, modules 05 to 07
- Method, modules 19 and 20
Each has a baseline that can be locked in week one.
Switched on from week six, charged for nothing until measured
- Customer and loyalty, modules 08 to 11
- Marketing and campaigns, modules 12 to 15
- Service and exchange, modules 16 to 18
Everything is switched on. The distinction is what is charged for now and what is proven first.
The first six months
| Month | What happens | What is measured |
|---|---|---|
| One | Baseline locked, scheme book loaded, scripts written and reviewed, two languages live | Baseline lapse rate, no-show rate, contact rate |
| Two | Instalment and maturity calling live across three states, appointment desk live on twenty stores | Lapse recovery rate, appointment confirmation rate |
| Three | Fourteen languages live, network wide inbound answering | Advisor hours returned, first response time |
| Four | Loyalty, marketing and service modules switched on, carried free | Dormancy segment response, service collection rate |
| Five | Conversation intelligence feeding media targeting and regional assortment | Cost per qualified walk-in against the baseline channel |
| Six | Review against the measured lines, and a decision on the other ten Titan businesses | Recognised value against fee, on Tanishq's own numbers |
The failure condition, stated up front
- If the lapse recovery rate does not move against the locked baseline by month three, the scheme modules have not worked and should be said to have not worked.
- If advisor hours returned to the floor do not appear in store rosters by month four, the contact centre case has not been made.
- Either outcome is grounds to end the engagement at the month six review with no tail and no penalty.
Method and limits
Where every figure in this build comes from, which assumptions remain assumptions, and what this system will not do.
Where the figures come from
| Figure | Status | Source or basis |
|---|---|---|
| Jewellery segment revenue, store count, segment growth, EBIT margin | Reported | Published Titan Company results |
| Golden Harvest structure, instalment minimum, 400 day closing rule, maturity benefit | Published | Scheme terms as publicly stated |
| Encircle tiers and cross-brand identity on one registered number | Published | Programme description as publicly stated |
| Account volumes, lapse rate, maturity volumes, average redemption value | Modelled | Modelled from store count and scheme structure. To be replaced with Tanishq data. |
| Inbound call volume, call mix, advisor hours, no-show rate | Modelled | Modelled from comparable retail networks of similar store count |
| Dormant identity counts, conversion and response rates | Modelled | Modelled. Every one is an editable input on its own console. |
| Language distribution across the store network | Derived | Derived from disclosed state level store distribution |
Language coverage, stated accurately
The conversational stack runs approximately thirty two languages end to end, which covers every language required by the Tanishq store network. A higher figure is sometimes quoted for text to speech alone; it does not describe full conversational capability and is not used anywhere in this build.
What this system will not do
- It will not quote a price on a specific piece of jewellery. That is a store advisor conversation and the agent says so plainly.
- It will not score, rank or profile an individual customer. Sentiment and theme analysis sit at store and issue level only.
- It will not hold caste, religion or community data. No such field exists in the schema.
- It will not call outside permitted hours, will not call an identity that has withdrawn consent, and will not call again on a reason once refused.
- It will not replace a store advisor. Every outcome requiring judgement is handed to a named person with a clock on it.
- It will not claim value it cannot trace to a dated event and a call record.
- Replacement of modelled volumes with Tanishq's own figures, which takes one session with the scheme and store teams.
- Confirmation of whether an existing in-group reference may be named in writing.
- The route in: Tanishq sits under the Jewellery Division, and the group level argument sits with the Managing Director. These are two different opening conversations.
- Whether the pilot runs on a state subset or on the full network from month three.
Concierge desk and private viewing rooms
Built on the member concierge pattern from the WeWork Connect Hub: one desk that holds every request a customer makes of a store, matches it to real capacity, and closes it with a named person rather than a queue. In a jewellery store the equivalent of a meeting room is a private viewing room and a senior advisor's diary.
What arrives on the desk
Every request type, where it comes in, who owns it, and whether the platform can finish it without the shop floor.
| Request type | Channel in | Volume a month | Owner after routing | Clock | Closed without an advisor |
|---|---|---|---|---|---|
| Private viewing appointmentBridal, high value, festive collection | Voice, WhatsApp | 1,240 | Store manager diary | 4 hrs | Yes |
| Bridal consultation with a stylistTwo hour slot, two advisors held | Voice, web | 410 | Named bridal advisor | 6 hrs | Part |
| Piece held for inspectionShip From Store or inter-store transfer | Voice, chat | 760 | Inventory desk | 1 day | Yes |
| Repair or resize statusReady, collected, delayed | Voice, WhatsApp | 905 | Service book | 2 hrs | Yes |
| Exchange and purity check bookingSlot plus documents required | Voice | 640 | Exchange desk | 4 hrs | Part |
| Scheme instalment or maturity questionAnswered from the scheme ledger | Voice, WhatsApp | 580 | Scheme book | 1 hr | Yes |
| Complaint or escalationNever auto-closed | All | 285 | Named escalation owner | 24 hrs | No |
Request mix modelled on inbound call reason coding from comparable retail deployments.
The booking ladder
Request captured in the customer's language
Occasion, budget band, brands of interest, preferred store and two acceptable time windows, taken in conversation rather than on a form.
Matched against real capacity
Advisor roster, viewing room availability and the piece being physically present in that store, checked before a time is offered.
Briefing note written to the advisor's existing view
What the customer said, what they looked at online, what they hold with the group, and the occasion date. No new login.
Confirmation ladder
Confirmation at booking, a reminder the evening before, and a live check two hours ahead that can rebook rather than cancel.
Outcome written back
Attended, rescheduled or lost, with the reason coded, the transcript attached and the follow-up dated.
Capacity is the product
A booking desk is only trusted once it never offers a slot that does not exist. Room, roster and stock are checked in the same call, exactly as space and desk inventory are in the member hub.
The occasion sets the clock
Coworking bookings are same-week. Bridal buying runs against a date months out, so the desk holds a dated follow-up series rather than a single confirmation.
People close, not the system
Complaints and escalations are never auto-closed. Both stay open until a named owner closes them and the customer confirms it.
- Advisor roster and viewing room inventory per store, refreshed daily.
- Read access to the appointment book already in the store application.
- An agreed escalation owner per cluster, with working hours.
Collection lookbook and the stylist brief
Taken from the brand hub work for Arvind and Trent: one place where a collection launch, its lookbook and its store level performance sit together, so the sales floor is briefed with the same asset the campaign runs on. For Tanishq the unit is a collection against an occasion, not a season against a category.
Collections on the floor now
What is live, how the floor is engaging with it, and what it is doing to attach and to average bill value.
| Collection | Occasion anchor | Stores live | Brief opened | Enquiries | Attach | Bill value effect | State |
|---|---|---|---|---|---|---|---|
| Rivaah bridalRegional bridal sets | Wedding date | 486 | 81% | 4,120 | 2.4 pieces | +14% | Peak |
| Festive gold everydayLight weight daily wear | Festival calendar | 528 | 74% | 6,880 | 1.6 pieces | +6% | Peak |
| Diamond solitaire editAnniversary and milestone | Purchase anniversary | 402 | 63% | 2,310 | 1.3 pieces | +22% | Building |
| Gifting under a price bandCoin, pendant, small sets | Festival, birthday | 528 | 69% | 5,140 | 1.2 pieces | -3% | Steady |
| Temple and heritageRegional, south first | Regional festival | 214 | 58% | 1,640 | 1.9 pieces | +11% | Regional |
| Men's editChains, rings, cufflinks | Gifting, self purchase | 361 | 44% | 930 | 1.1 pieces | +4% | Trial |
Collection names describe collection types, not confirmed Tanishq campaign plans. Engagement and attach figures are modelled on comparable launches.
Where the brief is landing
One master, many floors
The apparel brand hub proved the pattern: a single approved master, translated and cut per region, beats a mail thread of attachments that half the network never opens.
Performance next to the asset
The lookbook and the store numbers sit on the same screen, so a collection that is not moving is visible to the person who can change the brief.
Occasion, not season
Every collection is anchored to a dated occasion, so the brief is issued against a wedding or festival window rather than a retail calendar quarter.
What this does not do
- It does not generate a price, a purity claim or a certification statement. Those come from the ledger only.
- It does not publish anything without a named merchandising approval recorded against it.
- It does not create imagery of a piece that does not exist in the catalogue.
Store audit and display compliance
The retail audit pattern, applied to 528 stores. A short structured audit per store per week, read automatically, scored against the display standard, and turned into a dated action with a named owner. What today takes a regional manager a fortnight of travel to see becomes a Monday morning list.
Compliance by cluster
| Cluster | Stores | Audits returned | Display score | Collection present | Open actions | Ageing over 7 days | State |
|---|---|---|---|---|---|---|---|
| Tamil Nadu | 65 | 98% | 94 | 96% | 11 | 1 | Good |
| Karnataka | 48 | 96% | 91 | 93% | 9 | 2 | Good |
| Maharashtra | 52 | 93% | 88 | 90% | 16 | 4 | Watch |
| Andhra and Telangana | 58 | 90% | 86 | 88% | 19 | 6 | Watch |
| Kerala | 34 | 95% | 90 | 92% | 7 | 1 | Good |
| Delhi and NCR | 41 | 84% | 79 | 81% | 24 | 9 | Intervene |
| Gujarat | 36 | 81% | 77 | 79% | 21 | 8 | Intervene |
| West Bengal | 29 | 88% | 83 | 85% | 13 | 3 | Watch |
| Rest of network | 165 | 91% | 87 | 89% | 27 | 7 | Watch |
Store counts follow the disclosed 528 store base. Scores, actions and ageing are modelled on audit cycles of comparable networks.
The eleven checks
Window and focal counter
Live collection present, priced correctly, lit, and matching the brief issued for that week.
Bridal room readiness
Room clean, sample trays complete, stylist brief printed or open on the advisor view.
Gaps against the plan
Missing sizes and price bands in the collections the store is briefed to sell this week.
Repairs awaiting collection
Pieces ready but uncollected for more than seven days, cross-checked against the service book.
Enrolment material
Golden Harvest collateral present, current, and in the store's own language.
Purity and pricing notices
Rate board current, purity and buy-back notices displayed as required.
An audit finding becomes a call
Repairs sitting uncollected in an audit are handed straight to the service book as a dated call list, so the finding produces an outcome rather than a report.
No individual scoring
Findings sit at store and issue level. No advisor is ranked, scored or named in a compliance report, in line with the position taken everywhere else in this build.
- The current display standard document, so the checks are scored against Tanishq's own definition.
- Who owns an action per cluster, and the hours in which the clock runs.
- Whether audits are completed by store staff, by cluster managers, or both.