By industry
AI agents and automation for e-commerce and retail operations
Most retail businesses do not have an AI problem. They have a gap between the storefront and everything behind it — the ERP, the 3PL, the marketplace, the inbox — and people paid to stand in that gap and re-key. We build the layer that closes it, with a review queue on anything that touches money or stock.
Two weeks to a written build plan · no obligation to build with us
- The hop that is usually still manual
- Order → ERPThe hop that is usually still manual
- Storefront we built and still run
- US east coastStorefront we built and still run
- Nothing posts to the ERP unseen
- Reviewed writesNothing posts to the ERP unseen
- To a written build plan, via the audit
- 2 weeksTo a written build plan, via the audit
Where the time goes
If two of these are true, there is something here
Written the way the people doing the work describe it, not the way a process document does.
Two people re-key orders all day
The storefront knows about the order. The ERP does not, until somebody types it in. Marketplace orders are worse, because they arrive in a different shape and someone has to translate.
Stock is right in one place and wrong in three
The website, the warehouse and the marketplace listing each hold a number. They agree on the morning of the stock count and drift apart by lunchtime. Oversells are the visible cost; the invisible one is the buffer you carry to avoid them.
Support answers the same forty questions
"Where is my order", "can I change the address", "is this back in stock". The answer is sitting in a system the agent has to open a second tab to read.
Nobody can enrich the catalogue fast enough
A supplier sends 400 new lines in a spreadsheet with three usable columns. Titles, descriptions, attributes and categories are then written by hand, which is why half the catalogue is thin and unfindable.
What we connect to
The systems you are already running
We plug into what you already run rather than asking you to move. These are the systems we have connected most often in retail — if yours is not here, it almost certainly has an API or a flat-file drop we can work with.
- Storefront and marketplaces
- Shopify / Shopify Plus
- WooCommerce
- Magento / Adobe Commerce
- BigCommerce
- Headless (Next.js + commerce API)
- Amazon Seller Central
- eBay
- Daraz
- Back office and fulfilment
- NetSuite
- SAP Business One
- Odoo
- Microsoft Dynamics 365 BC
- QuickBooks
- Xero
- 3PL / WMS APIs
- ShipStation
- Customer, marketing and data
- Zendesk
- Gorgias
- Freshdesk
- HubSpot
- Klaviyo
- Mailchimp
- Google Analytics 4
- Meta / Google Ads APIs
What we build here
Five things worth building for e-commerce & retail
Each one names the systems it touches, because an integration that is not named is an integration nobody has thought about yet.
- 01
Order intake into the ERP
Orders from every channel are read, normalised to your SKU and customer records, and posted to the ERP. The awkward ones — a new customer, a SKU that does not map, a discount outside policy — go to a queue instead of guessing.
- Touches
- Shopify · marketplace APIs · NetSuite / SAP B1 / Odoo
- What changes
- The re-keying job disappears; the exceptions become a short daily list.
- 02
One stock number across channels
A single availability service reads the warehouse, applies your channel buffers and pushes to storefront and marketplace listings. Adjustments are logged with a reason, so a discrepancy can be traced instead of argued about.
- Touches
- WMS / 3PL · Shopify · Amazon · eBay
- What changes
- Fewer oversells, and a smaller safety buffer holding cash.
- 03
A support agent that can actually look things up
Order status, tracking, returns eligibility and stock answered from the live systems, with the customer identified before anything is disclosed. Refunds, address changes and cancellations are drafted for a human to approve, not executed.
- Touches
- Zendesk / Gorgias · Shopify · carrier tracking APIs
- What changes
- The routine tickets close themselves; the agents get the ones worth a person.
- 04
Catalogue enrichment at supplier scale
A supplier feed becomes titles, descriptions, attributes and category assignments in your voice and your taxonomy, with the source values kept beside the generated ones. Everything lands in a draft state a merchandiser approves in bulk.
- Touches
- Supplier feeds · PIM / product tables · storefront
- What changes
- New lines go live in days rather than sitting in a spreadsheet.
- 05
Returns and reconciliation
Returns matched to their original order, condition noted, restock or write-off proposed, credit memo drafted. Marketplace settlement files reconciled against your own order records so the gap is found before month end.
- Touches
- Marketplace settlement files · ERP · accounting
- What changes
- Month end stops being a hunt for a number that does not match.
How it is put together
How an order actually gets into your ERP
This is the flow we are asked for most often in retail, and it is the one that shows how the whole access layer is put together. Every other retail use case is a variation on it.
- 01
Read
Webhooks from the storefront where they exist, polling with a cursor where they do not, and a mailbox reader for suppliers who still send PDFs. Every message is stored raw before anything interprets it.
- 02
Resolve
Channel identifiers become your identifiers: SKU mapping, customer matching, currency and tax treatment. Anything that cannot be resolved with confidence is not guessed — it is queued.
- 03
Decide
Your rules, written down and versioned: margin floors, stock thresholds, discount policy, fraud signals. The model drafts; the rules decide what may pass unattended.
- 04
Write
A scoped ERP credential that can create an order and nothing else. Writes are idempotent on the channel order id, so a retry cannot post the same order twice.
- 05
Account for it
Every decision is logged with its inputs, its rule version and whoever approved it. When finance asks why an order posted at that price in March, the answer is a record, not a memory.
Your data
What we will and will not do with it
Stated plainly, and without claiming a certification we do not hold. If a requirement here is harder than what is written, say so on the call and we will tell you whether it is buildable.
We do not touch card data
Nothing we build stores, transmits or processes a card number. We work on order metadata — totals, line items, statuses — and leave the payment data inside your gateway, which keeps your PCI scope where it already is.
Customer data stays in your systems
The access layer reads from your ERP and storefront and writes back to them. If you need model calls to avoid a jurisdiction or a vendor entirely, we run an open-weights model on your own infrastructure instead.
Marketplace rules are a design constraint
Amazon and eBay rate-limit, throttle and change schemas without much warning. We build against those limits deliberately — backoff, cursors, replay from the raw store — because a sync that silently falls behind is worse than one that stops loudly.
Proof
We build and run retail commerce, not just talk about it
Gardyn is a US east-coast hydroponics brand. We built the storefront, catalogue and fulfilment flows, and they are built to hold a seasonal traffic curve rather than an average day. Retail is one of the sectors where we have shipped the boring part — the fulfilment path — as well as the interesting one.
- Live storefront
- US east coastLive storefront
- Built to hold the curve
- Seasonal peaksBuilt to hold the curve
Technology
What this gets built with
We pick the boring option unless there is a reason not to — the framework is the part most likely to be abandoned before your system is.
- Commerce
- Shopify APIs
- Storefront API
- WooCommerce REST
- Next.js
- React
- Integration
- Python
- FastAPI
- Celery
- Redis
- Webhooks
- Idempotency keys
- Agents and models
- LangGraph
- OpenAI
- Anthropic
- Open-weights (self-hosted)
- Structured output
- Data and operations
- PostgreSQL
- Raw event store
- OpenTelemetry
- Docker
- Per-run cost logging
How the engagement runs
Five phases, each with something you can hold
Every phase ends in a named deliverable. You can stop after any of them and keep what has been built.
- 01Week 1–2
Sit with the work
We watch the process as it is actually done, not as the process document says. The output is a map of every step, which system it touches and how long it takes.
DeliverableProcess map and a shortlist of what is worth automating - 02Week 2–3
Design the access layer
Which systems get read, which get written to, what scope each credential carries, and which decisions stop for a person. Agreed in writing before anything is built.
DeliverableArchitecture and permission model - 03Week 3–8
Build against your real data
Built in your sandbox against real records, not a demo dataset. The write path goes through a review queue from the first day, so nothing lands in a live system unseen.
DeliverableRunning system in staging, with the review queue - 04Week 8–10
Run it beside the humans
It runs in parallel with the people doing the job. We compare the two outputs, fix what differs, and only then start letting the routine path through unattended.
DeliverableAgreement rate report and a go-live threshold - 05Ongoing
Hand over and watch
Source code in your repository, deployment in your accounts, cost and failure dashboards you can read without us. Support if you want it, not because you are locked in.
DeliverableRepository, runbook and monitoring
Why Devs Core
Three reasons that are checkable
We run two of our own products
Cadenza and Budge are ours. What an agent costs per run, where it fails and which decisions need a person came from operating them, not from a vendor deck.
The write path is reviewed by default
Anything that changes a record in a system of record goes through a queue a person can see. You choose what graduates to unattended, and when.
You own the code
Source in your repository, infrastructure in your accounts, model keys in your name. There is no runtime of ours you have to keep paying for.
Questions
E-commerce & Retail — the questions we get asked
Do you replace our ERP or our storefront?
No. Replacing either is a year of work and most of the pain is not in them — it is in the space between them. We build the layer that connects what you run, and leave your systems of record where they are.
How do you stop an agent from posting a wrong order to the ERP?
It cannot post a wrong order unattended, because unattended is something an order path earns. Writes run through a review queue while we measure agreement against the people currently doing the job. Only the categories that match consistently get released, and your margin, stock and discount rules sit outside the model as code.
Will this work with our marketplace listings as well as our own site?
Yes, and the marketplace side is usually where the savings are, because marketplace orders arrive in the least convenient shape. We have worked with the Amazon, eBay and Daraz patterns; the constraint is their rate limits, which we design around rather than fight.
What about our supplier who still emails spreadsheets?
That is a normal input, not an edge case. A mailbox reader picks up the attachment, the values are extracted and mapped to your catalogue, and anything ambiguous goes to a person rather than being guessed. The original file is kept, so you can always see what the supplier actually sent.
Can the support agent issue a refund?
It can draft one with the order, the reason and the amount filled in, and a person approves it. We do not hand an agent an unattended refund path — the downside is uncapped and the upside is a few seconds.
How long before it is doing real work?
A single flow — order intake, or stock sync, or ticket deflection — is typically running in staging against your real data in three to six weeks, then runs beside your team before anything goes unattended. The two-week AI Readiness Audit is the cheaper way to find out which flow to start with.
We are on a platform you have not listed. Is that a problem?
Usually not. What matters is whether your systems expose an API, a webhook or a scheduled file drop — almost all of them do at least one. If yours genuinely does not, we will tell you that in the audit rather than after you have paid for a build.
Related
What e-commerce & retail buys most often
- Read more
AI Agent Development
Agents that take real actions in your systems — reading live data, drafting work, and writing back only what a person has approved.
From $6,500 - Read more
AI Integration Services
The access layer between the software you already run and any AI model — so an agent can reach your ERP, CRM or helpdesk without being handed the keys.
From $8,000 - Read more
Process Automation
Replace the manual steps between two systems — the re-keying, the copy-paste, the spreadsheet that reconciles what neither system knows.
From $4,500 - Read more
Web Development
Fast, accessible, search-ready sites and web applications — the kind that load in under a second and rank because of it.
From $3,500
Tell us which process is costing you.
Thirty minutes, no preparation, no deck. You describe what keeps going wrong and we tell you whether AI is the answer — including when it is not.
Prefer email? contact@devs-core.com