Custom AI software for ecommerce operations, from catalogue to customer
We build AI into the way your store runs: returns and order questions resolved end to end, catalogue data written and cleaned at scale, stock forecasts, supplier invoices matched, and shopping assistants grounded in your own catalogue. Connected to Shopify, your ERP and your warehouse.
What does AI for ecommerce actually handle?
AI for ecommerce covers three jobs. Customer facing agents resolve order status, returns, exchanges and product questions across chat, email, WhatsApp and phone. Shopping assistants help buyers find and compare products using your live catalogue and reviews. Behind the storefront, AI enriches product data, analyses reviews, routes tickets and keeps marketplace listings on Amazon and noon in step. Each part connects to your store platform, helpdesk and carriers, and acts within limits you set.
| Who it suits | DTC brands, multi brand retailers, marketplace sellers, omnichannel retailers |
|---|---|
| Channels | Site chat, email, WhatsApp, SMS, phone, helpdesk |
| Platforms | Shopify, BigCommerce, WooCommerce, Adobe Commerce, Salla |
| Marketplaces | Amazon, noon, eBay |
| Rules we design to | PCI DSS v4.0.1, FTC rules, Consumer Rights Act 2015, DMCC Act 2024, UAE Federal Law No. 15 of 2020 |
| First release | 3 to 6 weeks for an order and returns agent |
Custom AI software for ecommerce operations, explained
- 1AI answers order questions
- 2Returns are approved within your policy
- 3Demand is forecast item by item
- 4Shopify and ERP stay in sync
Read the video transcript
Custom AI software for ecommerce operations. From catalogue to customer, connected to your store and ERP. Here is the problem. Returns and support tickets pile up. Stock runs out or sits for months. And product data is cleaned by hand. Here is how it works. First, AI answers order questions. Second, returns are approved within your policy. Third, demand is forecast item by item. And finally, shopify and ERP stay in sync. What do you get? Grounded in your own catalogue. Connected to your store and ERP. No per seat fees. Most ecommerce projects start between $10,000 and $40,000. The easiest way to start is a $1,500 AI Pilot on your own data. You see a working version in ten business days, and the fee is credited to the full build. Book a free call, or message us on WhatsApp, at vrisic.com.
Why support is the best place to start with AI for ecommerce
Support tickets in an online store are repetitive, rule based and tied to data you already hold, which is exactly the work AI agents do well.
Tag a week of tickets in almost any online store and a pattern shows up. A large share asks some version of “where is my order”, known in the trade as WISMO. Next come returns and exchanges, address changes, discount code problems and sizing questions. Each has a correct answer sitting in your store platform, your carrier’s tracking data or your policy page. Your team spends its day copying that answer from one screen to another.
A chatbot with canned replies does not fix this, because shoppers want the answer for their own order. An AI agent can read the order, check tracking, apply the return policy and take the action, then write a clear reply. When it hits something it should not decide, such as a chargeback threat or a damaged high value item, it passes the ticket to a person with the facts already gathered.
The same foundations then carry other work. Once the agent can read your catalogue and orders, it can become a shopping assistant on the product page, a retail AI voice agent on your phone line, or an internal tool that drafts product copy. That is why we scope ecommerce AI as a small platform, even if the first release does one job.
Ecommerce AI development: what we build for online stores
Six systems, each useful on its own and designed to share one integration layer.
- 01
Order and returns agent
An agent in your helpdesk, site chat and WhatsApp that verifies the shopper, answers order status from live tracking, starts returns and exchanges inside your policy, updates addresses before fulfilment and issues credits up to a limit. It writes every action back to the order and the ticket.
Best for: Stores where WISMO and returns fill the support queue
- Carrier tracking lookups
- Policy checked returns and exchanges
- Refund and credit limits you control
- 02
AI shopping assistant
A conversational assistant on product and category pages that asks what the shopper needs, compares products, answers questions from specifications and reviews, and adds items to the cart. It only recommends products in stock in the shopper’s market, and it says when nothing fits rather than pushing the nearest item.
Best for: Technical, sized or highly configurable catalogues
- Grounded in your live catalogue
- Size and fit guidance from your data
- Add to cart and bundle suggestions
- 03
AI product search
Search that understands “waterproof hiking boots for wide feet under $150” and returns the right products, combining keyword and vector retrieval over your catalogue. We build it with the same retrieval methods as our enterprise AI search work, tuned for product data.
Best for: Catalogues with thousands of SKUs and poor zero result rates
- Natural language queries
- Hybrid keyword and vector ranking
- Merchandising rules preserved
- 04
Retail AI voice agent
A phone agent for retailers that still take calls: order status, store hours and stock at a branch, returns, and simple phone orders completed by a secure payment link. It speaks Arabic and English for Gulf retailers and hands off to staff with a summary.
Best for: Omnichannel retailers, furniture, electronics and grocery
- Order status by phone
- Store and stock questions
- Payment by link, never by voice
- 05
Catalogue enrichment pipeline
A generative AI pipeline that turns supplier sheets and images into consistent titles, descriptions, attributes, alt text and marketplace listings, validated against the source data. Merchandisers approve by category and exceptions go to a queue.
Best for: Retailers onboarding large supplier ranges
- Attribute extraction from images and PDFs
- Channel specific listing formats
- Fact checks against specifications
- 06
Marketplace operations agent
Automation for sellers on Amazon and noon: syncing listings and stock, answering buyer messages inside each marketplace’s rules, flagging listing suppressions and preparing weekly performance notes. Built as AI workflow automation with a person approving price and listing changes.
Best for: Brands selling across their own store and two or more marketplaces
- Amazon SP API and noon API integration
- Buyer message drafting
- Suppression and stock alerts
Behind the storefront: reviews, tickets and catalogue data
Some of the best returns come from AI that shoppers never see.
Reviews analysis
Reads every review and return reason, groups them into themes such as sizing, packaging or a faulty batch, and sends product and buying teams a weekly summary with example quotes.
Ticket triage and routing
Classifies incoming tickets by reason, language, value and sentiment, fills the order fields and routes them to the right queue, so humans start with context instead of a blank ticket.
Reply drafting for agents
For tickets that stay with people, the AI drafts a reply using the order data and your macros. Agents edit and send, which keeps tone human and saves typing.
Supplier data cleanup
Normalises units, colours and sizes across suppliers, spots duplicates and fills missing attributes, so filters and feeds such as Google Merchant Center stop rejecting products.
Returns reason insight
Links return reasons to products, sizes and suppliers so you can fix the listing or the product, not just process the refund.
Search log analysis
Reviews what shoppers searched for and did not find, and suggests synonyms, missing products and content gaps for your merchandisers.
Store platforms, marketplaces and tools we integrate
We connect through official APIs and apps you own, and confirm scopes before quoting. See our AI integration service for how we approach this.
Store platforms
BigCommerce now sits under the Commerce parent brand. Salla is widely used by merchants selling into Saudi Arabia.
Marketplaces
noon covers the UAE, Saudi Arabia and Egypt through its seller API.
Helpdesk and messaging
The agent works inside your helpdesk so your team keeps one queue.
Shipping, returns and payments
Payment platforms are used for refunds and links only; the AI never sees card data.
Models and retrieval
Product answers come from retrieval over live catalogue data, not model memory.
Voice and monitoring
Every conversation and tool call is traced for quality review.
Payments, PCI DSS scope and shopper data
The safest ecommerce AI is one your payment auditor barely needs to look at. We design for that from the start.
Out of PCI DSS scope by design
The AI never receives card numbers. Payments go through hosted checkout, a payment link or masked keypad entry, which keeps the agent outside your cardholder data environment under PCI DSS v4.0.1.
No widgets on payment pages
PCI DSS v4.0.1 pays close attention to scripts on payment pages, so we keep chat and assistant scripts off checkout and payment pages unless your assessor agrees otherwise.
Shopper verification
Before sharing order details or changing anything, the agent verifies the shopper with an order number plus email or phone, and never reveals addresses or payment details in full.
Action limits
Refunds, credits and discounts have hard limits in code, per order and per day. Anything above them goes to a person, whatever the shopper says.
PII redaction in logs
Names, addresses, emails and phone numbers are masked in logs and evaluation data, and transcripts stay in your cloud account with retention you set.
Privacy law by market
We design to CCPA/CPRA for California shoppers, UK GDPR and PECR for UK marketing, and the UAE PDPL, Federal Decree Law No. 45 of 2021, including limits on reusing consumer data for marketing.
Consumer law your AI customer service must respect
An AI agent makes promises on your behalf, so it has to know the rules of each market. These are the ones we build into policies and tests; your legal team signs off.
United States
The FTC Mail, Internet, or Telephone Order Merchandise Rule requires shipping within the time you state, or 30 days, and a delay notice with the option to cancel. The FTC’s rule on fake reviews took effect in October 2024. California’s BOT Act requires disclosing a bot in sales conversations.
- Delay notices and cancel options in WISMO flows
- No AI written reviews posing as customers
- Clear disclosure that the shopper is talking to AI
- No invented product claims
United Kingdom
The Consumer Rights Act 2015 gives shoppers a 30 day short term right to reject faulty goods, and the Consumer Contracts Regulations 2013 add 14 days to cancel most online orders. Since 6 April 2025 the DMCC Act 2024 bans fake reviews and drip pricing, and the CMA can fine up to 10% of global turnover.
- Statutory rights stated correctly in returns flows
- Cancellation window checks per order
- Total price shown up front
- Review replies drafted, never fake reviews
United Arab Emirates
Federal Law No. 15 of 2020 on consumer protection, with Cabinet Resolution No. 66 of 2023, expects information in Arabic, dated invoices, warranty handling and return or exchange options. Federal Decree Law No. 14 of 2023 governs trading by modern technological means.
- Arabic and English answers and policies
- Invoice and warranty details on request
- Complaint handling to your written process
- UAE hosting on Azure UAE North or AWS Middle East (UAE)
Ecommerce AI development cost
Typical ranges; you get one fixed price in writing after a free scoping call.
Order and returns agent
One store and one helpdesk that want WISMO and returns resolved automatically
$8,000 to $25,000£6,400 to £20,000AED 29,000 to AED 92,000
3 to 6 weeks
- Agent in your helpdesk and site chat
- Store platform and carrier tracking
- Returns and exchanges within policy
- Refund and credit limits
- Quality dashboard
Multi channel customer agent
Brands selling across web, WhatsApp, phone and marketplaces
$25,000 to $60,000£20,000 to £48,000AED 92,000 to AED 220,000
6 to 10 weeks
- Order, returns and product questions
- WhatsApp and voice channels
- Amazon or noon buyer messages
- Arabic and English
- Shopping assistant on product pages
- Staged rollout by channel
Commerce AI platform
Retailers combining service, search, catalogue and marketplace AI
$60,000 to $150,000£48,000 to £120,000AED 220,000 to AED 550,000
12 to 20 weeks
- Customer and shopping agents
- AI product search
- Catalogue enrichment pipeline
- Reviews and returns analytics
- Admin console and reporting
Typical ranges, not quotes. Running costs are separate: managed support from $1,500 per month and model, hosting and voice usage billed at cost; voice runs roughly 8 to 25 US cents per minute (our typical estimate). Market context: the Clutch AI pricing guide (2026). For agent budgets in detail read our AI agent development cost guide.
What moves the price of an ecommerce AI project
Two brands on the same platform can land in different tiers. These are the reasons.
| Cost driver | Effect on budget | Why it matters |
|---|---|---|
| Number of sales channels | High | Each marketplace, region store and messaging channel needs its own integration, rules and tests. One Shopify store is much simpler than Shopify plus Amazon plus noon. |
| Actions versus answers | High | Reading order status is simple. Issuing refunds, editing orders and creating returns need limits, verification and far more testing. |
| Voice channel | Medium | Phone adds telephony, latency tuning, payment handoff and call testing on top of the text agent. |
| Returns policy complexity | Medium | Final sale items, region specific windows and exchange rules each become logic the agent must apply correctly. |
| Languages | Medium | Arabic and English, or several European languages, add evaluation sets and review time for every flow. |
| Catalogue size and data quality | Medium | Messy supplier data needs cleaning before a shopping assistant or search can be trusted with it. |
| Ticket volume | Low | Volume barely changes build cost. It changes running cost through model usage. |
AI customer service for ecommerce: helpdesk AI, platform app or custom?
Helpdesk vendors and store platforms now ship capable AI. For a single store with standard policies they may be all you need.
| Helpdesk built in AI | Store platform AI apps | Custom build with Vrisic | |
|---|---|---|---|
| Time to launch | Days | Hours to days | 3 to 10 weeks |
| Pricing model | Per resolution or seat | Monthly app fee | Fixed build fee plus usage |
| Order status and returns | Yes | Yes | Yes |
| Several stores and marketplaces | Some connectors | No | Yes |
| Phone orders and voice | Some vendors | No | Yes |
| Arabic and English with UAE hosting | Varies | Varies | Yes |
| Your own action limits and audit log | Within product settings | Basic | Yes |
| You own the code and data flow | No | No | Yes |
If a built in tool covers your tickets, use it. Our guide to AI automation tools versus custom builds walks through the decision.
Send us a month of ticket reasons
On a free scoping call we look at your ticket mix, your platforms and your returns policy, and show you which part an AI agent would resolve first. You get a fixed price in writing, or a clear recommendation to use your helpdesk’s own AI.
Questions we hear every week
Still unsure about something? Ask us on a call. We will give you a straight answer, even if it means we are not the right partner.
What is AI customer service for ecommerce?
AI customer service for ecommerce is an agent that answers shoppers by chat, email, WhatsApp or phone and actually resolves the request: it looks up the order, reads carrier tracking, starts a return within your policy, changes an address before dispatch or issues store credit under limits you set. Anything outside those limits goes to your support team with the order and conversation attached.
How does an AI agent answer “where is my order” questions?
The agent verifies the shopper with an order number plus email or phone, reads the order from your store platform, pulls live tracking from the carrier or a tracking service such as AfterShip, and explains the status in plain words. If the parcel is late beyond your threshold, it follows your delay policy, such as reshipping, refunding or escalating, and tags the ticket.
Can an AI shopping assistant recommend products without making things up?
Yes, if it answers only from your live catalogue. We index product data, specifications, size guides, stock and reviews, and the assistant can only recommend items it retrieved, with the real price and availability. It refuses to invent features and says when a product does not suit the shopper’s need. We test it against a list of tricky product questions before launch.
Can you build an AI agent for Shopify?
Yes. We build on the Shopify Admin GraphQL API for orders, customers, returns and refunds, and the Storefront API for product search and carts, as a custom app you own. Shopify also publishes Model Context Protocol servers for AI agents, which we use where they fit. The agent can run in your helpdesk, on your site, in WhatsApp or on a phone line.
Can an AI voice agent take card payments over the phone?
It should not handle card numbers directly. Keeping a voice AI out of your PCI DSS scope is simpler and safer, so we either send a secure payment link by SMS or WhatsApp, or hand the caller to a telephony payment feature such as Twilio Pay that masks keypad entry. The AI never hears, stores or logs the card number.
Does AI for ecommerce work in Arabic for UAE and Saudi shoppers?
Yes. Current models from Anthropic, OpenAI and Google handle Modern Standard Arabic well and Gulf dialect reasonably, and we test against real customer messages. UAE consumer rules expect information to be available in Arabic, so product answers, return policies and order updates can be served in Arabic and English, with hosting in a UAE cloud region if you need it.
Can AI write product descriptions for thousands of SKUs?
Yes, in batches with checks. The system drafts titles, descriptions, attributes and alt text from supplier data, images and your style guide, then validates each draft against the real specifications so it cannot claim a material or feature the product lacks. Your merchandisers approve samples per category, and flagged items go to a review queue before publishing.
Is AI allowed to write or respond to customer reviews?
AI can summarise genuine reviews, tag themes and draft replies for staff to approve. It must never write reviews that appear to come from customers. The FTC rule on consumer reviews, effective October 2024, and the UK DMCC Act 2024, in force for consumers from April 2025, both treat fake reviews, including AI generated ones, as unlawful.
How much does AI customer service for an online store cost?
A focused agent that handles order status and returns typically costs $8,000 to $25,000 to build and takes 3 to 6 weeks. A multi channel agent with product advice and marketplace support is usually $25,000 to $60,000. Model usage is billed at cost and support plans start at $1,500 per month. You get one fixed price after a free scoping call.
Should we use our helpdesk’s built in AI or a custom agent?
Start with the built in AI if most tickets are simple and your helpdesk vendor’s actions cover your store platform. A custom agent earns its cost when you sell on several channels, run a custom returns policy, need Arabic, want voice, or need the agent to act across systems such as your warehouse, ERP and marketplaces at once.
Find the process where AI will pay off first
Message us on WhatsApp for the fastest reply, or book a free thirty minute call. You will leave with the processes most worth automating, whether to build new or upgrade what you have, a realistic timeline and a clear idea of cost, whether or not you work with us.