AI Agents for D2C & E-commerce Stores

If you run a D2C brand, most of your day disappears into repetitive questions: "Is this in stock?", "Where's my order?", "Does this fit a 32 waist?" An AI agent can handle that conversation 24/7 on WhatsApp, Telegram, email, or your own site chat — closing sales while you sleep and rescuing carts you'd otherwise lose. This guide shows exactly which jobs to hand over, in what order, and what each one is realistically worth.

The store-owner job list AI can take over

Before automating anything, list the work that eats your day. For most Indian D2C stores it looks like this:

• Answering the same pre-sale questions on WhatsApp ("price? COD available? delivery to Indore?")
• Recommending products when a buyer is undecided
• Sizing and fit guidance for apparel/footwear
• "Where is my order" (WISMO) — the single biggest support drain
• Returns, exchanges, and refund status
• Nudging happy buyers to leave a review
• Re-engaging buyers who haven't bought in 60–90 days
• Recovering abandoned carts

Not all of these are equal. A good rule: automate the high-volume, low-judgment tasks first (WISMO, FAQs, sizing) and keep the high-judgment ones (a furious customer, a damaged-on-arrival claim) routed to a human with a clean handoff. The agent's job isn't to replace you everywhere — it's to absorb the 80% that's mechanical so you handle the 20% that needs a person.

Pre-purchase: turn questions into checkouts

This is where AI agents pay for themselves fastest, because every unanswered WhatsApp message is a lost order.

Product Q&A. Feed the agent your catalogue (titles, prices, materials, care instructions, COD/UPI/shipping policy) and let it answer in the buyer's language — including Hinglish. "Bhaiya cotton hai kya, wash kaise karein?" should get an instant, accurate reply, not silence until you're free.

Recommendations. When a buyer says "gift for my wife, budget 2000", the agent should ask one or two clarifying questions (occasion, colour preference) and return 2–3 specific SKUs with images and a buy link — not a generic catalogue dump.

Sizing and fit. The highest-leverage pre-purchase fix for apparel and footwear, because bad sizing drives both lost sales and returns. Give the agent your size chart and a short fit logic ("if between sizes, size up for our oversized fits") so it can convert hesitation into a confident order.

Here's a copy-paste system prompt to start a pre-purchase agent. Replace the bracketed bits:

You are the shopping assistant for [BRAND], a D2C [category] store in India.
Goals, in order: (1) answer accurately, (2) recommend the right product,
(3) help the buyer check out. Speak in the customer's language; match Hinglish
with Hinglish. Be warm, short, and specific.

RULES:
- Use ONLY the catalogue, size chart, and policies provided. Never invent
  prices, stock, materials, or delivery dates.
- For recommendations, ask at most 2 clarifying questions, then suggest 2-3
  specific products with name, price (₹, GST-inclusive), and the buy link.
- For sizing, use the size chart. If a buyer is between sizes, apply: [FIT RULE].
- State COD, UPI, shipping, and return policy when asked — never guess.
- If a buyer is angry, has a damaged item, or asks for a refund amount you
  can't confirm, say you're connecting a human and hand off. Don't improvise.
- Never promise a discount that isn't in [ACTIVE OFFERS].

For deeper prompt patterns, see Prompt Engineering for Real Business Tasks and the 40 WhatsApp Customer-Support Prompts (Copy-Paste).

Post-purchase: tracking, returns, and reviews

Once the order ships, the conversation shifts from selling to reassuring.

Order tracking (WISMO). Connect the agent to your courier/AWB data so a buyer texting "where is my order" gets the live status instantly. This one integration typically removes the largest single chunk of your support volume.

Returns and exchanges. The agent collects the order ID, reason, and photos, checks them against your return window and policy, and either issues an RMA or routes edge cases to you. Crucially, it can deflect avoidable returns — many "I want to return" messages are really sizing or expectation problems the agent can solve in chat.

Reviews. Three to five days after delivery, the agent sends a short, friendly ask — ideally with a one-tap link. More reviews lift conversion on the very product pages the pre-purchase agent is selling from, so the loop compounds.

Win-back and loyalty automation

Acquiring a new buyer costs far more than reactivating an existing one, so this is where margin hides.

• Win-back: Identify buyers with no purchase in 60/90/120 days and trigger a personal WhatsApp message — referencing what they bought, suggesting a complement, optionally with a time-boxed offer.
• Replenishment: For consumables (coffee, skincare, supplements), nudge near the expected re-order date. A reminder timed to when they're actually running out converts far better than a random blast.
• Loyalty / VIP: Greet repeat buyers by name and history, surface early access, and quietly skip the discount for buyers who'd have bought anyway.

Keep these compliant: under WhatsApp's rules, proactive messages need approved templates and opt-in, and under India's DPDP Act you should only message buyers who consented and offer an easy opt-out. Bake the opt-out line into every template.

Connecting Shopify / WooCommerce realities

The honest part: an agent is only as good as the data it can reach.

Practical realities to plan for: catalogues change daily (stock-outs, new drops), so the agent needs a scheduled sync, not a one-time upload. Courier tracking is its own integration. And handoff matters — when the agent escalates, the human needs the full order context, not a cold "customer needs help". Start read-only (answer, track, recommend) before you grant write access (issue refunds, edit orders); it's lower risk and still removes most of the work.

For where these agents actually run and how to keep them reliable, see Deploy & Host AI Agents the Right Way and the The Production-Ready Agent Checklist.

Use-case-to-ROI table

Direction and effort, not guarantees — your numbers depend on volume and margins. Figures below are illustrative.

A two-week rollout checklist

• Pick the ONE job costing you the most hours (usually WISMO)
• Export catalogue, size chart, and policies into one clean doc
• Connect order/tracking data (Shopify/Woo API or a tracking link)
• Write opt-out + DPDP consent into every proactive template
• Define the human-handoff trigger and what context gets passed
• Soft-launch on one channel, read-only, and read 50 real transcripts
• Fix the top 5 wrong answers, then expand to the next use case

Next steps

1. Run the two-week checklist for a single use case — don't boil the ocean.
2. Map your catalogue and policies into one document; that prep is 80% of a good agent.
3. Draft your pre-purchase agent with the prompt above and test it on real past chats.
4. If you'd rather not build, browse hosted store agents on AgentDukaan, or hire a done-for-you setup via the Help Center.

You can buy a ready-made store agent, subscribe to a hosted one, or get it set up for you on AgentDukaan — start with the job that's costing you the most this week, and let the wins fund the next one.