More done in Magento, with your team signing off every change.
We connect an AI assistant to your store through a secure, bespoke MCP server. Your team asks in plain English, the AI drafts the work, a person approves it, and every change is checked on the live site.
The repetitive admin work, at catalogue scale.
The jobs that sit on a to-do list for months because they mean hundreds of small edits in the Magento admin. Every change is drafted by AI, approved by a person, then checked on the live site.
Catalogue
Bulk catalogue work
Missing specs, cross-sells, alt text and unique product copy across every product.
Content
Pages and locations
Buying guides, category content, landing pages and location pages built with Hyvä CMS.
Search
SEO and Shopping feeds
Meta titles, redirects and Merchant Center fields fixed in batches, not one by one.
From a request to a checked, live change.
Nothing reaches your live site until someone on your team approves it. Each item is checked once it is live, so one bad change can never take a whole batch down with it.
1. You ask
A plain-English request, such as fixing missing meta titles.
2. AI drafts
It plans the batch and drafts every change for review.
3. You sign off
Approve or reject each item on a shared sign-off page.
4. MCP publishes
Through the Magento API, one item at a time, with an audit log.
5. Live check
Each item is checked on the live site before the next one goes.
Pass or roll back
If the check passes, the next item publishes. If it fails, the item is rolled back automatically and your team is alerted.
A bespoke MCP server, built on the Magento API.
Your AI assistant only ever reaches your store through tools we build for you, with the access each phase needs and nothing more.
Every change is signed off on one shared page.
Your team sees exactly what will change before it changes, and a record of everything that has gone live.
Checked on the live site after every single item.
Nothing is assumed to have worked. Each change runs the checks that apply to it. If a check fails, that item is rolled back automatically and your team is alerted before anything else publishes.
Page loads
Returns 200, with no errors or 404s.
Matches approval
What’s live is exactly what was signed off.
Schema valid
Product and LocalBusiness markup still validates.
Images load
No broken or missing product images.
Mobile layout
Hyvä templates still render cleanly on phones.
Feed intact
The Shopping feed still carries the field.
The riskier the change, the more sign-off it needs.
A starting point we agree with you. Anything can move up a level.
What your team could ask for.
Real requests, written the way people ask for them. Filter by area of the site or by how much sign-off each one needs.
A bespoke visual merchandising algorithm for every category page.
Instead of sorting by newest or a single best-seller rule, we build a scoring algorithm around your trading priorities. It sets the order of products on every category page, and your team adjusts it by asking in plain English.
Weighted signals
Sales, conversion rate, margin, stock depth and newness each carry a weight we agree with you.
Visual balance
No rows of near-identical products. Colours, styles and price points are mixed across the first screens.
Stock-aware
Lines with low stock or broken size runs move down, so customers see products they can buy.
Pins and boosts
Pin hero products, boost a brand or range for a campaign, and move lines you are phasing out further down.
Scheduled re-sorts
Categories re-sort automatically from the latest data, for example every Monday morning.
Signed off and checked
Each re-sort is shown before and after for approval, then checked on the live category page.
Built so a bad instruction can’t damage your store.
AI assistants can be misled by instructions hidden in what they read, such as a supplier spreadsheet, a product review or a web page. We design your MCP server so this kind of prompt injection can’t change your store on its own.
Content is treated as data
Text the AI reads from product data, files and pages is handled as information, never as instructions.
Only the tools we build
The AI can only call specific tools, and each one checks what it is allowed to change.
Limits on every tool
Caps on batch sizes and price changes, so one request can’t rewrite thousands of products or slash prices.
A person approves every change
Nothing changes on your store until someone on your team approves it on the sign-off page.
Keys and customer data kept out
API keys never pass through the AI, and customer records and payments stay out of reach.
Everything logged
Every prompt, tool call and approval is recorded, so anything unusual can be traced and rolled back.
What we agree before anything goes live.
Nothing goes live without a person saying yes. And everything that does go live is checked and logged.
Customer data stays out
Customer records and payments are kept out of scope (UK GDPR).
Read-only first
We start read-only, then switch on write access one area at a time.
Tested on staging
Every new type of change is tested on staging first.
Named approvers
A named senior person approves prices and promotions.
Full audit log
Every request, approval and publish is logged.
Automatic rollback
If a live check fails, that item is reverted straight away.
Phased, lowest risk first. Each phase starts once your team is happy with the one before.
Phase 1 · Read-only
Read-only insight
Monday health checks, stock balancing and a weekly trading summary.
Phase 2 · Auto-publish after sign-off
Content and SEO
Meta titles and alt text, location and category pages, and Shopping feed fields.
Phase 3 · Named approver
Trading changes
Seasonal edits, promotions and price modelling.
Three decisions, then a read-only pilot
We agree who signs off and who the named approvers are, the Magento API permissions for your MCP server, and a first read-only batch to pilot. We usually start with a Monday health check on your own catalogue, so you see real results with no risk to the live site.
Results, year after year.
Our ecommerce clients generate over £100m in online sales every year.
Average client ecommerce revenue growth
Premier Welding
Average year-on-year ecommerce growth since launch, with INDEZ managing paid media on a pay-for-performance basis.
Work with a team that truly understands ecommerce growth.
Our team has taught ecommerce at the University of Strathclyde and presented at the Scottish Parliament, and runs the Ecommerce Club and Ecommerce Scotland.
Ready to grow your ecommerce business?
Tell us where you are and where you want to be. We'll come back with a plan, not a sales pitch.
Book a Growth Call →Scotland · England · Wales