Shopify, the CRM and the Excel sheet, in one question.
A ₹30–150 crore D2C brand has its orders in Shopify, its customers in a CRM, its marketing spend in three ad dashboards and its margins in an Excel sheet one person maintains. Every real question ("which SKUs are selling but losing money after returns and ad cost?") needs all four. This blueprint shows how a founder asks that question on a Tuesday and gets the answer on Tuesday, and how the support inbox stops being a full-time job.
Who this is for
Modelled on representative organisation profiles. Test these parameters against your internal numbers.
Size band
Teams
Systems
Regulator
Chapter
The three documents and two processes
The 3 documents and 2 processes you already have. No new content has to be created to begin.
The 3 Core Document Sets
Orders, returns and SKU data (Shopify, 3PL)
Exported to Excel for any question that spans more than one report
Return rates per SKU are known a month late
Customer and conversation records (CRM, helpdesk, WhatsApp)
Searched by order number; the history of a customer across channels is a reconstruction
Repeat customers are treated as new; complaints repeat
Margin sheet and vendor price lists
One Excel file, one owner, updated monthly
The only place true contribution margin exists
The 2 Mission-Critical Processes
Support replies on order status, returns and product questions
60–70% of tickets are the same five questions; answered by hand, often by pasting into a public AI
Customer data leaves the company with every paste
The weekly growth review
A growth lead spends Friday building a sheet from Shopify, ads and the margin file
Decisions on Monday are made on Thursday's numbers
Days 1–90 in four stages
A phased, non-disruptive roadmap. Governed rules signed first, production live by week three, full autonomy by day 90.
- PHASE 01Days 1–15
Connect and write the rules
We map which systems hold what, which support staff are already pasting customer data into public tools, and which numbers the founder actually decides on. The rules (next section) are written in plain language and the founder signs them.
Milestone: Signed rules & zero software footprint - PHASE 02Days 16–36
Open the door
Vectalk connects to Shopify, the CRM and the margin sheet, read-only, in your cloud account or on NeoPhoenix's India infrastructure as your own tenant. Three weeks to production. From day 36 the founder asks "Which SKUs had a return rate above 15% last quarter, and what was their contribution margin after ad cost?" and gets a table with the source row for each number.
Milestone: Vectalk deployed in private boundary - PHASE 03Days 37–60
The first worker: support
A Neo Agent reads each incoming ticket, pulls the order status, return policy or product detail from your own records, and drafts the reply. Routine replies go out under a support lead's approval; anything about a refund above a threshold, a complaint or an unknown question is routed to a named person. Nothing is drafted from outside your records.
Milestone: First autonomous worker live with approvals - PHASE 04Days 61–90
The second worker: the Friday review
A second agent assembles the weekly growth sheet every Friday morning from Shopify, the ad platforms and the margin file. The growth lead checks and sends it. The audit log of what was asked, what was drafted and who approved is the DPDP evidence that customer data stopped leaving the company.
Milestone: Complete audit evidence & board report
The rules written first
The governance engine runs before the models. Rules are codified in software and signed off by responsible executives.
Access
ENFORCED RULESupport sees order and conversation history, not margins; growth sees sales and ad data, not individual customer phone numbers; the founder sees everything
Masking
ENFORCED RULEPhone numbers, addresses and payment details are masked for every role except fulfilment and the support lead
Approvals
ENFORCED RULENo reply with a refund, replacement or discount goes out without the named approver; routine order-status replies are approved in batches by the support lead
Escalation
ENFORCED RULEA complaint, a legal word, a health claim or a question the records do not answer goes to a named person, not to the model's guess
Where it runs
ENFORCED RULEYour cloud account or your own tenant on NeoPhoenix's India infrastructure; no customer record is sent to a public AI service
What the board sees at day 90
Modelled, with the inputs beside each number. Replace with your own.
- BOARD DELIVERABLE 01
A cross-system question answered the same day, with the source.
InputShopify, CRM and margin sheet connected; the answer shows the row each number came from.
- BOARD DELIVERABLE 02
Support drafts for the five repeat questions, approved in batches.
Input1,200 tickets a month, 65% routine; the support lead approves rather than types.
- BOARD DELIVERABLE 03
Zero customer records pasted into a public AI.
Inputthe agent drafts from your own records; the audit log shows it.
- BOARD DELIVERABLE 04
The Friday growth sheet ready at 9 am.
Input4 hours of assembly removed; the growth lead reviews, not builds.
- BOARD DELIVERABLE 05
Return-rate and margin by SKU known weekly, not monthly.
Input800 SKUs; the same question asked every Monday.
Price band and what recurs
Transparent pricing with zero hidden seat taxes or foreign inference surcharges. Board-ready clarity.
Vectalk on Shopify, CRM and the margin sheet
Licence and usage
First Neo Agent (support drafts)
Per quarter
Second Neo Agent (Friday review)
Per quarter
One-time setup and connector configuration
Does not recur
FAQ
Straightforward answers to the compliance, security, and IT architecture questions boardrooms ask.
No. It is your team asking questions across your own systems, and an agent drafting support replies from your own records for a person to approve. A customer-facing bot can be added later under the same rules.
Shopify, WooCommerce, most CRMs and helpdesks, Google Sheets and Excel connect through standard connectors; anything unusual is scoped on the briefing call.
Only the replies you mark as routine, and only after the support lead approves the batch. Refunds, complaints and anything unknown go to a person every time.
No. The models run inside your boundary; there is no API key to a public AI provider in the deployment.
The pilot is sized for a brand with one finance person and a support team of three. It is the same platform the larger deployments run on, at a usage-based price.
Similar shape, different documents: see the textile manufacturer blueprint
Compare how this governed 90-day pilot architecture applies across adjacent operational models.
AI your board can say yes to.
Bring the question you had to build a spreadsheet to answer.