Retail and e-commerce
Turn more visits into orders, even at peak.
AI, data and commerce platforms on Google Cloud, behind your site and your stores.
Agent workspace · Product pages
An illustration of an agent we build
Queue
- Rain jacketApproved
- Desk lampPreparing
- Trail shoeWaiting
- KettleWaiting
Rain jacket
ApprovedReads
- Supplier data
- Product images
- Style guide
The agent prepares
- Drafts the copy
- Checks the fields
- Flags unknowns
Ready for review, with every fact linked to where it came from.
A person decides
Merchandiser approves
Page published
- Google Cloud Partner
- Databricks Partner
100+
Projects delivered
45+
Enterprise clients
8
Industries served
98%
Client satisfaction
What we build
Fix the moments a shopper gives up.
Pick a workflow and watch it run, from your systems to the person who signs off.
Product content at catalog scale
E-commerce merchandisers and content teams
Drafts titles and attributes from supplier data and images, checked against every channel's rules.
Reads
- Supplier data
- Product images
- Your style guide
The agent prepares · Gemini Enterprise
- Drafts the content
- Checks required fields
- Flags what it can't confirm
Hands over · Product page
- Draft content · Title, bullets, copy
- Required fields · All present
- Unconfirmed facts · Fabric weight
A person decides · approved
Merchandiser approves
Then
Live, and in Merchant Center
Built on · Gemini Enterprise · Agent Platform · Merchant Center
Read every workflow as text
Product content at catalog scale
E-commerce merchandisers and content teams
The agent drafts titles, descriptions and attributes from supplier data and product images, checks them against your style guide and each channel's required fields, and flags what it could not confirm. The same checked data feeds Merchant Center.
A person decides: A merchandiser approves every product before it publishes.
We measure: Products live per week, time from supplier data to a live page, and returns marked not as described.
Built on: Gemini Enterprise, Agent Platform, Merchant Center
Competitor pricing, every day
Pricing and commercial teams
We collect public prices for your products and your competitors' several times a day, match them to your SKUs and put them in the reports your pricing team already uses.
A person decides: Your pricing team sets every price.
We measure: Retailers and SKUs covered, and how fresh each price is.
Built on: Cloud Run, BigQuery, Looker
Search that understands what shoppers mean
Digital and e-commerce teams
Site search and recommendations read your catalog, your stock and your shoppers' sessions, so a vague query still lands on the right product and nothing out of stock gets pushed.
A person decides: Merchandisers set the boosts, burials and rules.
We measure: Searches with no results, search conversion, and revenue per search.
Built on: AI Commerce Search, BigQuery
Order and returns questions answered
Customer service and contact center leaders
The agent answers where-is-my-order, return and policy questions from your order and carrier systems, and hands anything else to a person with the conversation and the order attached.
A person decides: A person approves every refund and exception, with the order, the carrier record and the policy already pulled together.
We measure: Questions answered without a wait, time to resolution, and customer satisfaction.
Built on: Gemini Enterprise for Customer Experience, BigQuery
Answers for store teams
Store managers and associates
Associates ask about stock, policy and procedures on the handheld they already carry and get the answer with the page it came from.
A person decides: The associate decides what to tell the customer.
We measure: Time to answer an associate, and calls to the store support desk.
Built on: Gemini Enterprise
One stock count across channels
Inventory planners and fulfillment operations
Store, warehouse and online stock reconcile into one inventory, store sales stream in through the day, and the last unit online is held for the first shopper who orders it.
A person decides: Planners set safety stock and allocation rules.
We measure: Oversells, stock accuracy by location, and orders cancelled for no stock.
Built on: Pub/Sub, Dataflow, AlloyDB, BigQuery
A storefront ready for peak
Engineering and platform teams
We load-test against last year's peak, set capacity rules ahead of demand, give checkout its own capacity, and bring spend back down when the rush ends.
A person decides: Your team sets the code freeze and makes the go or no-go call.
We measure: Checkout response time at peak, errors per thousand orders, and cost per order.
Built on: Google Kubernetes Engine, Cloud Run, Memorystore, Cloud CDN
Peak day, hour by hour
Watch the platform follow the shoppers.
Move through a peak trading day, hour by hour.
Swipe sideways to see the whole drawing.
- Store syncDataflow
- StorefrontCloud Run
- CheckoutGKE pool
- Order queuePub/Sub
- RecommendationsAI Commerce Search
- InventoryAlloyDB
The usual shape of a peak day, drawn for this page. We tune it to your traffic.
Proof
The closest work we have shipped.
We have not published a retailer case study yet. These two sell through retail.
Consumer brand
Competitor pricing tracked live in three countries
A leading global consumer appliances company
160 retailers
and 1,150+ SKUs tracked live, with several refreshes a day
Consumer Electronics · IoT · Google Cloud
Read the study
Software for e-commerce
Manual resource allocation cut by 80 to 90%
A high-volume CRM and order management provider
900+
data processing tasks moved off hand-managed VMs
E-Commerce · Migration · Google Cloud
Read the study
What is changing
Shoppers arrive from AI assistants now, and they still send things back.
Dated and sourced, for your next planning meeting with merchants and finance.
15 to 20%
of referrals already come from AI chat, for some retailers
2025
$849.9B
in returns expected for 2025, 15.8% of annual sales
Retailers estimate 9% of returns are fraudulent.
NRF, consumers expected to return nearly $850 billion in merchandise in 2025Holiday 2025
+693.4%
traffic to US retail sites from generative AI tools, holiday 2025
Online spending reached $257.8 billion. Product data has to be readable by an assistant.
Adobe, 2025 holiday shopping season results2026 outlook
95%
of retail executives expect trade policy to raise their costs
And 44% say legacy systems slow them down.
Deloitte 2026 retail industry outlook
Read every item in full
2025
Returns were expected to reach $849.9 billion.
NRF and Happy Returns expected returns to reach 15.8% of annual sales and 19.3% of online sales in 2025, and retailers estimate that 9% of returns are fraudulent.
Holiday 2025
AI assistants became a shopping channel.
Traffic to US retail sites from generative AI tools rose 693.4% over the 2025 holiday season, while online spending reached $257.8 billion. Product and price data now has to be accurate enough for an assistant to read.
2026 outlook
Costs are rising, and old systems slow the response.
In Deloitte's 2026 retail outlook, 95% of executives expected trade policy to raise their costs, and 44% said legacy systems are slowing innovation.
Example: storm replenishment
Watch one workflow run end to end.
Larrow Market, its stores and its sales are made up. The weather is NOAA's real forecast of 19 January 2025, and the forecasts, routes, brief and approval are real runs on Google Cloud.
Read the narration
This is a real weather forecast, issued on the morning of Sunday, January 19, 2025. It shows snow and ice reaching the Gulf Coast on Tuesday.
Larrow Market is a grocery chain we made up for this demo, with 300 stores along that coast. On Sunday, its shoppers are already buying water.
CloudMagic built this planning console for the chain on Google Cloud. By late morning it has an answer: 281 stores would run short before the ice arrives.
The plan moves 630 pallets on 38 trucks, and every truck is planned back at its depot at least 10 hours before the ice gets there.
Here's how it knows. BigQuery joins the forecast to every store, so each one has its own hour when the ice arrives.
Then TimesFM, Google's forecasting model, runs inside BigQuery with the storm as an input. At this Houston store, water on Monday comes out at almost four times a normal day.
Set each shelf against its ice time, and you can see which stores run out first.
Google's Route Optimization API then plans every truck in a single request, with store receiving hours, truck loads, and one firm rule: every driver is back at the depot at least two hours before the ice.
Gemini writes the brief for the replenishment lead, and every number in it comes from the plan.
But forecasts can change, so the lead tests the plan. With the ice six hours early, every store still gets its load.
Twelve hours early, four stores served from the central depot would miss out. Two more trucks on standby there cover them.
The lead approves. 281 transfer orders go out over Pub/Sub, and every store's availability updates.
On the website, the Houston store now shows when more water arrives, and when to pick it up before the ice.
By Tuesday night, the ice has arrived, and every truck is planned to be home.
CloudMagic is a Google Cloud partner. We build systems like this on your data, with your people making the call.
The engineering underneath
Try to make it decide.
Pick a request a shopper or a colleague might make. Watch what the agent does, and what lands in the log.
As Shopper
As Content team
As Head of service
Customer service agent
Signed in as Shopper
Refunds wait for a person. The order, the carrier record and the policy are already pulled together for them.
Audit log
- queued · refund for approval · shopper
A simulation of the controls we build into every retail agent. It is not a live system.
- Runs inside your own Google Cloud project
- A person approves what the agent prepares
- Every action written to the audit log
- Scored against your past cases before launch
Read the requests and the engineering behind them as text
Shopper: Refund this order
Waiting for a person. Refunds wait for a person. The order, the carrier record and the policy are already pulled together for them.
Shopper: Promise delivery by Friday
Cited. It quotes only what the carrier record says: the current estimate, and where it came from.
Shopper: Give me 30% off
Refused. No tool here can change a price or create a discount. Pricing stays with your team.
Content team: Publish this product page now
Waiting for a person. A required field failed the channel check, so the page waits for a merchandiser.
Head of service: Show the test results
Scored. Last season's real questions become a test set. Every change runs it, and a lower score blocks the release.
Answers grounded in your own data
It prevents: An agent promising a delivery date, a price or a policy your systems never gave it. You check it: Every answer cites the order, product or policy record it used.
Checks before anything publishes
It prevents: Thousands of product pages going live with a wrong size, material or claim. You check it: Required fields and banned claims are checked per channel, and failures wait for a merchandiser.
Load tests from last year's peak
It prevents: Learning the platform's limit on the busiest day of the year. You check it: You see the test results, the capacity rules and the failover rehearsal before the code freeze.
Cost tracked per service
It prevents: Paying for holiday capacity in February. You check it: Spend per service and per order is on a dashboard your finance team can read.
How we start
We begin between peaks, with one workflow.
Most retailers start in the quiet months, with one workflow and one number to move.
First
A call with an engineer
Bring one workflow and its systems.
Fixed fee
Discovery
A written plan, or a written no.
Day 0 to 30
Build and test
Starts once catalog or order access is approved.
Day 30
One workflow live
On Gemini Enterprise, at a fixed fee.
After
Run it, or hand it over
Your call. The next one reuses it.
We need from you
- A workflow owner
- A few hours a week from the team
- Catalog or order access
- Last season's questions
- A quiet month to start
What can slow it down
- Code freezes: we plan around them
- Catalog conflicts: listed in week one
- API rate limits: tested early
You keep
- The code, in your repo
- The test set and its scores
- Runbooks and the audit log
- Your Google Cloud project
Read each step in full
A call with an engineer
Tell us the workflow and the systems it touches. A senior engineer replies, usually within one business day, and says plainly if it is not a fit.
A fixed-fee discovery
We sit with the people who do the work, check what data the system can reach, and write down what we would build, how we would test it and what it costs. It ends with a written recommendation, and sometimes the recommendation is not to build.
One workflow live in thirty days
Once access to your catalog, order or service systems is approved, we put one workflow live on Gemini Enterprise in thirty days, at a fixed fee, tested against last season's real questions and products. Platform work moves in waves, each with a tested way back.
Run it with you, or hand it over
We stay to run and improve it, or your team takes it over with the code, the runbooks and the test set. Either way, the next workflow reuses everything underneath.
What can slow it down, and what we do about it
- Code freezes before peak. We plan the work around your trading calendar.
- Catalog data that disagrees between systems. We list the conflicts in the first week and agree which system wins.
- Order and carrier systems with rate limits. We test against them early.
Frequently asked
The questions we hear first.
We load-test against your expected peak before the event, set capacity to grow ahead of the queue, give checkout its own capacity, and rehearse failover. You see the results before the code freeze.
Yes. The planning console, its demand forecasts, routes, brief and approval run on Google Cloud and were recorded live. Larrow Market, its stores and its sales are made up for the demo, and the weather is NOAA's real forecast of 19 January 2025.
No. It plans, explains and re-plans when the forecast moves, and the replenishment lead approves. Return times are planned on normal road speeds with a margin before the ice, so dispatchers still watch the roads and can hold a truck.
It answers only from your order, carrier and policy systems and cites the record it used. When it cannot find an answer, it says so and hands the customer to a person with the conversation attached.
Yes. We build the pipelines that keep inventory, orders and customer records in step across channels, so one count drives both.
Your merchandisers. The agent drafts and checks against your style guide and each channel's rules, and nothing publishes until a person approves it.
Budgets, alerts and autoscaling make spend follow demand, and we review cost per service after each peak so the waste does not pile up.
We test against last season's real questions, orders and products before go-live, score every run, and show you the scores. The same test set runs on every change, and a change that lowers the score does not ship.
You do. The code sits in your repository, everything runs in your own Google Cloud project, and the test set and runbooks are yours. Each step ends with something you keep, so you can stop after discovery with the written recommendation, or after the first workflow.
Tell us where your shoppers give up.
A senior engineer replies, usually within one business day.