Catalog, inbox, returns, and the ads desk behind the storefront

If you came here looking for a new Shopify theme, a headless storefront, or a Nidrosoft-branded cart, stop. We do not ship that product. Cyriac Zeh founded Nidrosoft in San Diego to wire production systems into tools a company already runs. On e-commerce that means the catalog, the shared inbox, the returns queue, and the ads account. The storefront you already have can stay. We will not pretend Octana or NanoBrief is a store. We will tell you what those products actually do, and we will tell you when a theme shop or a Shopify Plus partner is the better first call.

A mid-market catalog operation looks like this on a Tuesday. A supplier sends a spreadsheet with new SKUs, three of which already exist under a different UPC. Someone merges rows by eye. Titles still say “new” from last season. The size grid is missing EU. The first image is a lifestyle crop that marketplaces will reject. The feed to Google and to a retailer portal goes out anyway. Support spends the afternoon answering “is this in stock” for a variant the site never created. That Tuesday is the work. We map it before anyone talks about a generative product page.

The inbox is a second storefront that nobody designed. “Where is my order.” “This arrived damaged.” “Can you price match.” “Cancel, I used the wrong address.” Those threads already have policies. They also have exceptions that live in one lead’s head. A chatbot that cannot see the order, the warehouse scan, and the return window becomes a third queue. We open the mailbox in week one. We count which threads are tracking, which are returns intake, which are catalog lies, and which are ads replies that should have stayed in the ads tool. That count decides the first build.

We are the wrong seat when the only approved project is a homepage redesign, a new cart, or a marketplace of your own that needs a full company. We are the wrong seat when you want unattended refunds and unattended ad spend with no named owner. If you want ChatGPT to write product titles in a vacuum, buy seats and a style guide. If you want titles that cannot publish without a required image and a live variant, keep reading and then use /services/audits.

Catalog work is identity work. A SKU, a UPC, a parent, a variant, a warehouse bin, and a marketplace ID can all disagree and still look fine on the website. The person who knows the truth is usually a merchandiser who has been there five years and a feed file they are afraid to touch. We sit with that person. We write the keys they already trust. A model that invents a “new” SKU because a supplier used a new color name will create a ghost that support cannot cancel. The first rule we put in software is: do not create a product without a key the PIM already understands.

Copy is a constrained write. Title length, forbidden claims, material names, and size strings that a filter can parse. Marketplace A wants a different title shape than your own site. A banned phrase in one country is required in another. Teams already have a style sheet that nobody opens. We put the constraints on the draft: character counts, banned tokens, required attributes. A merchandiser still publishes. A model that dumps a supplier paragraph into the store is how you get a medical claim on a cream you are not allowed to make. We have no interest in that demo.

Images have a gate. Primary image on white, or the marketplace rejects the item. Lifestyle can be secondary. A phone photo from the warehouse floor will fail. Coordinators already know this and still miss it when a drop lands at 4 p.m. A build here is a required shot list and a hold on the feed until the list is complete. We use the same habit on listing photos in real estate and on lot photos in automotive. The object is a SKU, not a house or a truck. The refusal to syndicate a bad first frame is the same.

Feeds fail in the delta. A price changes in the PIM and not in the retailer file. Inventory goes to zero in one warehouse and the site still sells. A variant is disabled and the parent stays live. Someone notices when ads spend hits a 404. We treat the feed as a queue with a last-success time and a diff, not as a nightly prayer. We do not replace your integration platform if you already have one. We make the exception list visible to a human before the file goes out. If your only problem is a bad iPaaS contract, call that vendor.

Tracking questions are the volume. The carrier has a scan. The customer has a screenshot that is six hours older. The warehouse has a wave that has not left. A good reply names the last scan and the next expected event, and it does not invent a delivery day the carrier did not promise. Most teams paste a template that says “your order is on the way.” That sounds kind and creates a second ticket. We draft from the scan. A human still sends when the scan is stale or the address is in dispute. The gain is that the draft can see the order.

Returns are a clock plus a condition. Window open or closed. Item worn or unused. Tag on or off. Wrong size versus damaged. Restocking rule versus courtesy. The lead who has been there longest already runs a private policy that marketing would not put on the site. We write the public rule and the exception path as two different routes. An agent can draft a return label when the rule is clean. A named human still approves a courtesy that costs money. We will not ship an unattended refund to win a demo. That is the same studio rule we write on the FinTech page.

Damage claims need photos and a warehouse note. A customer sends a blurry corner. The warehouse says the box left clean. Chargebacks start here. We attach the photo to the order, we ask for the two angles the desk already knows it needs, and we leave the yes-or-no with the person who already talks to the payment provider. A model that apologizes in the voice of the brand and then issues store credit is how you teach the internet a new exploit. We would rather lose that slice of work than staff your returns queue with a generous ghost.

Marketplace messages have their own SLA. Amazon, eBay, a retailer portal: the clock is public and the tone is watched. A late reply is a score hit. A wrong promise is a score hit. The safe product is a draft that can see the order and the catalog row, sitting inside the tool the team already uses, with a person who hits send. We do not scrape a marketplace you do not have the right to automate. You own the seller accounts. If your plan is a hidden farm of identities, we will walk, the same way we walk a dealer who wants a hidden Facebook farm.

Ads operations is a desk with a feed, a budget, and a graveyard of campaigns that still spend. A SKU goes out of stock. The ad does not. A title changes. The creative still says the old price. A merchandiser launches a drop and tells nobody on the ads account. The person who notices is the one who watches ROAS on a Friday. We sit with that person. We map the objects they already use: campaign, ad group, SKU, budget cap, and the feed that is supposed to stay true. A model that writes a clever headline for a dead SKU is a cost, not a win.

Budgets need an owner the same way API keys do. A cap that lives in a slide does not pause spend at midnight. TokenTra taught us that a meter without a hand on the switch is a report you ignore. Ads are the same shape with different vendors. We can draft a pause when the PIM says zero, and we can alert a named person when a campaign crosses a line they set. We will not let an agent raise budget without that person. If your agency already owns the account and will not share login, the first finding of the audit is ownership, not copy.

Creative production is where Octana can sit next to the desk without pretending to be the store. Octana is a live Nidrosoft system: one message in, a finished video out. A drop that needs a short clip for paid social can start there. A merchandiser still checks the SKU, the claim, and the end card. We will not tell you Octana replaces a studio that shoots the product on a table. We will tell you it is a way to get a cut into the ads account on a Tuesday when the shoot is next month. If you need the shoot, hire the shoot.

Briefs for agencies and in-house studios are another ads-ops object. Who is the customer, what is the offer, what must never be claimed, which SKUs are in. NanoBrief is a live Nidrosoft system that asks a short questionnaire and drafts a brief in under two minutes against the three-plus hours that packet often takes. Nine industry templates exist, including marketing and video. A person still sends the brief. We mention NanoBrief here because the ads desk already writes that document badly, not because NanoBrief is an e-commerce platform.

Nidrosoft is the San Diego studio of Cyriac Zeh. He ships end-to-end AI product work under the name The AI Architect. Before the studio he worked at Anthropic, Microsoft on Bing and Edge, Intuit, and Gap Inc. Gap is a retail past, not a claim that we run Gap.com. We have shipped 125 products and we keep 16 public systems in the archive. We will not list merchant logos we do not have. If a store is not on /work, we will not hint that they are. That rule is how you should read every paragraph on this page.

Omnira UI is the open-source design system under the Omnira products. It is useful if we build you an internal catalog console. It is not a storefront theme. Omnira as a platform is for traditional industries with a record (a chart, a unit, a notice). A SKU can be a record. We have not shipped a commerce vertical and we will not announce one here to make the page feel finished. Custom work is the honest product.

TokenTra matters to a retailer that has already let every desk call a model. Ten to 30 percent of wasted spend is the published range when features have names and keys have owners. Protectron matters if you sell into Europe and a buyer just sent an AI questionnaire. We are not your bank and we will not move settlement funds. We will put a human on any write that looks like a refund or a courtesy credit. Those two sentences belong on an e-commerce page because that is where the incidents happen.

Embed means we sit with merchandising, support, and the ads operator on a live week. We watch a spreadsheet become a product. We watch a return get a courtesy it should not have gotten. We watch a campaign spend on a 404. We take the PIM, helpdesk, and ads logins you already own, under your policies. We do not start with a workshop about generative storefronts. The first week is access and shadowing. The longer path is on /how-we-work.

Diagnose means we cost the repeating work. Hours on catalog cleanup. Hours on tracking tickets. Hours on returns that are actually catalog lies. Spend that continued after inventory hit zero. We score whether we can reach the PIM, the mailbox, and the ads account. A team that will not show a redacted week of tickets cannot buy an inbox project in good faith. We will say that rather than invent a deflection rate.

Scope means you buy a result. For catalog, the result might be a draft that cannot publish without a key, a title shape, and a primary image. For inbox, a tracking draft that can see the scan, and a returns path that splits rule from courtesy. For ads, a pause when the PIM says zero, with a human on any budget raise. We refuse “AI for the whole store.” If you cannot name the object, you are not ready.

Build means we ship into your stack. Your PIM, your helpdesk, your ads account, your warehouse notes if we are allowed to see them. Your people sit with us while the first live draft goes to a small slice of SKUs or a small slice of tickets, then the slice grows. Someone in-house has to be able to edit a banned-claim list and pause a feed on a Friday. That person is part of the build. Twelve-plus years of product work has made us impatient with a vendor-only admin.

Own means the credentials, the rules, and the approval policy stay yours. We maintain what you ask us to maintain. We stay out of the rest. If a marketplace changes a title limit, the owner of the catalog is you, and we update the constraint with you. Continuous work is fewer ghost SKUs, fewer invented delivery dates, and fewer ads on zeros, measured in the queues you already run. If those queues do not move, we say so and we stop expanding the slice.

Do not hire Nidrosoft to rebuild your storefront. Theme, checkout, payments, a new marketplace: those are other firms. We will say so in the first call and we will point you at them if we can name one we trust. We will not take a cart project so we can “be full stack.” The studio is full stack on the systems we actually ship. A cart we have not shipped is not one of them.

Do not hire us for unattended refunds, store credit, or ad budget raises. A named human approves writes that cost money. If your growth lead wants a bot that always says yes, they can hire a different shop and they can own the chargebacks. The same no applies to inventing delivery dates and to medical or performance claims a model pulled from a supplier PDF. We will put those on a banned list or we will leave.

Do not hire us to scrape a marketplace you do not have the right to automate, or to stand up a farm of seller identities. You own the seller accounts and the relationship with Amazon, eBay, or the retailer portal. We build drafts and queues on access you provide. A ban that takes down the store Amazon login is a company problem we will not create for a demo. That is the same line we write for dealers who want a hidden Facebook farm. If your plan is a shrug when the listing site blocks you, we will walk in the first week.

We are a poor fit for a founder who only wants a keynote, and for a store that has no catalog owner and wants “AI merchandising” as theater. Nidrosoft’s main seat is an operator with a repeating queue and a decision maker in the room, often 30 to 250 people, sometimes smaller when one feed or one mailbox pays for itself. A two-person brand can still write. We will tell you if Octana or NanoBrief is enough, or if you should stay in the admin you already have.

An AI audit is two weeks in the work. We take a slice of the catalog, a week of tickets, and a look at campaigns that still spend. We sit with merchandising, support, and whoever owns the ads login. We write down every handoff: who creates the SKU, who hits publish, who answers the marketplace SLA, who can issue a courtesy. The deliverable is a ranked map and a list of things we will not touch. Details live on /services/audits.

Access is the gate. PIM, helpdesk, ads, and a redacted order sample if returns are in scope. We use your accounts and your policies. We do not ask you to paste customer orders into a consumer chat so we can “try a prompt.” If counsel needs a DPA, we do that first. If the ads account lives at an agency that will not share, we write that as an ownership finding in week one. That finding is often worth the audit by itself.

We score failure the way the P&L scores it. A ghost SKU is a cancel and a support hour. An invented delivery date is a second ticket and a chargeback risk. An ad on a zero is spend you cannot get back. A courtesy refund with no record is a hole. Those outrank a slightly nicer product title. Sometimes the first recommendation is a required image and a banned-claim list, with no model in the loop. That is a successful audit. You still own the map.

At the end you get a build order you can take to another shop. If the first line is “hold the feed until the primary image exists,” we write that. If the first line is “hire a theme shop, not us,” we write that too. The audit is the product. The build is a separate buy. Start at /contact or send cyriac@nidrosoft.com a redacted feed error and a week of ticket tags.

Automotive transfers the overnight queue and the photo hold. Nexuvo dealers pull private-party cars from Facebook, Craigslist, and Cars.com while the lot is closed, and they acquire three times the inventory with eighty percent less manual outreach. A catalog desk is not a lot. The transfer is a night job that drafts and a morning person who commits, plus a refusal to publish a bad first frame. We will not tell a merchandiser that a car buyer is their customer.

Real estate transfers listings and inbox routing. Eviction Wizard holds a file with a clock. Coordinators fight syndication lag the way you fight a stale feed. Professional services transfer intake and briefs: NanoBrief lives more naturally there, and we still mention it here because ads ops writes briefs. FinTech transfers the meter and the human on money. Healthcare transfers the record habit from Omnira Dental. Manufacturing transfers the floor board, which in commerce is often the warehouse exception list.

Music and media transfer Octana and Clear Audio AI when the brand actually makes sound and picture. We will not force a furniture shop into a mixing assistant. If your catalog is tracks or stems, say that and we will talk about MixGuidera and Clear Audio AI as the products they are. If your catalog is physical goods, stay on catalog, inbox, returns, and ads. The sibling pages exist so we do not have to stretch one story over every SKU type.

Write cyriac@nidrosoft.com with a redacted feed error, a week of ticket tags, or a campaign that spent after inventory hit zero. Name the PIM, the helpdesk, and the person who can approve a courtesy refund. Cyriac Zeh reads that mail from San Diego. We will tell you whether the next step is a two-week audit, Octana or NanoBrief as they already run, or a no because you need a theme shop or a warehouse project we do not do. We do not need a mood board.

If you already want the ranked map, book the audit on /services/audits. Two weeks, your accounts, your people, a build order you can take elsewhere. If you already know the catalog is the fire, say so in the first line. If the inbox is the fire, say that. If ads spend is the fire and the agency owns the login, say that so we can talk about ownership before we talk about headlines.

A workshop is available when merchandising, support, and ads do not share a story. Corporate AI training at Nidrosoft uses your SKU, your ticket, and your live campaign, not a toy catalog. Leadership sees what a model can draft and what a person still has to publish. The playbook stays. A workshop is enough when alignment is the problem. It is not enough when the feed is already lying and the courtesy refunds are already unattended. We will say which side you are on.

Nidrosoft is twelve-plus years of product work in a studio that still answers mail. 125 products. 16 public systems. The e-commerce page has no flagship store because we have not shipped one. The adjacent proof is Octana, NanoBrief, All AI Tools, and the queues we already run in other industries. If that mix matches the desk you run, write us. If it does not, use the sibling pages or go back to the admin. Both are honest endings.