sector
software for energy brokers, suppliers and the people who reconcile their invoices
Broker ERPs, supplier tendering and invoice extraction pipelines built one supplier format at a time.
- 11
- supplier formats extracted
- 100+
- suppliers in one tender
- ERP
- not a spreadsheet
- live
- under active support
Clutch
the problem
what this sector actually deals with.
Energy broking runs on documents nobody designed to be read by a machine. Every supplier issues invoices in its own format, contract data lives across a CRM, a spreadsheet and an inbox, and the margin on a deal depends on arithmetic that has to be right every single time.
sound familiar?
- Supplier invoices arrive as PDFs, each in a different layout
- Tender comparison happens in spreadsheets that only one person understands
- Management fee and uplift calculations are done by hand
- Contract renewal dates are the business, and they live in someone's calendar
- A misread figure is not a bug, it is an invoice you cannot defend
what we have built here
systems delivered in this sector.
A full brokerage ERP
Supplier tendering, contract lifecycle, brokerage accounting and client relationships in one system rather than four — built for a UK commercial energy broker and still under active support.
Eleven invoice extraction pipelines
One per supplier, because each supplier's layout is genuinely different. Scottish Power, SSE, Shell Energy, npower, Utilita, Yorkshire Gas & Power, United Gas & Power, Yu Energy, Regent Gas, TEM and Total — each with its own quirks.
Reconciliation, not trust
Extracted values are checked against the system's own records and updated on mismatch. That is the difference between a demo-grade parser and something you can bill from.
Tender and quotation documents
Management fee calculation on both the tender edit screen and the client summary, with generated analysis reports and tender quotes carrying per-document headers.
the hard part
eleven suppliers, eleven layouts, one number that has to be right.
Document extraction demos beautifully and fails commercially. A generic parser gets 90% of a clean invoice and silently mangles the tenth — which is the one with the credit adjustment on it.
- 01
The approach that works is unglamorous: build extraction supplier by supplier, tuned to that supplier's actual layout, rather than one clever generic parser that is wrong in eleven different ways.
- 02
Every extracted value is then reconciled against the system's own bill records rather than trusted. Where they disagree, the system updates and flags rather than quietly accepting the document.
- 03
The API surface is segmented by supplier and client entity, so adding a twelfth supplier is a contained piece of work rather than a regression risk across the other eleven.
- 04
This is the pattern any business drowning in unstructured supplier documents needs — utilities, insurance, logistics and finance all have the same problem wearing different clothes.
what it runs on
the stack we reach for here.
Chosen for the sector requirements and for maintainability over the life of the system.
- Application
- Laravel — tendering, contracts, brokerage accounting
- Extraction
- Per-supplier PDF pipelines with reconciliation against system bills
- Documents
- Generated analysis reports and tender quotes, per-document headers
- Operations
- Dedicated development and error-log channels; production under active support
how an engagement runs
what the first weeks look like.
Start with the documents
We take your real invoices — the messy ones, not the samples — and prove extraction on them before anything else is scoped.
Model the money
Uplift, management fee, commission and reconciliation rules agreed on paper. This is where a brokerage system is won or lost.
Build supplier by supplier
Each format is its own piece of work with its own acceptance test, so progress is visible and additions are safe.
Run it with you
Production error logs go to a channel we watch. An ERP is not a delivery, it is a system somebody has to keep alive.
selected work
written up in full
The engagements with the detail attached — and the numbers.
01
B2B commerceThe engineering partnership behind a cross-border wholesale marketplace.
Next.js storefront, Python ingestion pipelines, DynamoDB data layer and AWS infrastructure.
02A commission-free ticketing platform built for ownership and scale.
Multi-organiser commerce, Stripe instalments and two native apps in one connected platform.
03
Global logisticsEmbedded product engineering for a global shipping platform.
Rate, tax and duty calculators, server-rendered courier pages and a custom MongoDB CMS.
04A custom furniture marketplace engineered from storefront to infrastructure.
Server-rendered Next.js commerce over a Laravel API, bespoke operations tooling and re-architected AWS infrastructure.
05An AI-native operating system that runs a consultancy lead to invoice.
Discovery briefs, e-signed contracts, Stripe deposits, delivery milestones and time tracking in one operational system.
06A conversion platform for a growing UK boiler-cover provider.
Custom plan configuration, postcode-qualified lead journeys, CRM synchronisation and campaign landing pages.
07A custom Shopify store for a French beauty brand.
Custom theme, customer accounts, loyalty rewards, referrals and gift-with-purchase offers.
08Made-to-measure shutters with live pricing.
A seven-step product builder with live previews, sample orders and supplier tools.
09
AdvertisingA self-service advertising platform for venues.
Campaign creation, audience targeting, in-app ads and reporting linked to venue check-ins.
10Social eventsAn events app with the tools to run it.
Mobile app, backend, advertising tools, a digital marketplace and website.
11Two mobile apps for real-time companionship.
Customer and buddy apps with per-minute billing, wallets, moderation and admin tools.
12Player profiles and discovery for youth football.
Verified profiles, video highlights, coach discovery and safeguarding on web and mobile.
13Websites and a talent platform for DeepSpatial.
Corporate and investor pages, the Xploor talent platform and ongoing releases on AWS Amplify.
14
TravelTour bookings with a custom mobile-first checkout.
A multilingual website connected to the booking API, with deposits, coupons and affiliate tracking.
15One platform for tenders, contracts and accounts.
Supplier tenders, contract management, brokerage accounting and client records.
Teams that trusted us with the thing that matters
UK · Europe · Worldwide
the practices behind it
what this work is made of.
product engineering
Multi-tenant SaaS, marketplaces, ticketing and booking engines — the systems your business actually runs on, built to survive their own success.
AI systems
Retrieval pipelines, support deflection and internal copilots — costed per token before a line is written, so the unit economics work at volume.
rescue & security
AI-generated prototypes and stalled builds taken to production, and compromised live systems recovered — audited first, stabilised second, then finished and hardened properly.
reviews · verified on clutch
the people who signed off the work
Clutch★★★★★5.0 / 5.0Across 18 independently published client reviews
“They have a deeper technical knowledge than any web designer I've met to date.”
faq
questions from this sector
Can you handle a supplier format we have not seen you do?
Yes — that is the normal case. Each new format is scoped as its own contained piece of work, which is exactly why the architecture separates them.
Is this AI document extraction?
Where a model genuinely helps, yes. But the reliability comes from reconciling against your own records, not from the extraction method. We will tell you which parts need a model and which do not.
We already have a CRM. Do we throw it away?
Usually not. We extend what works and rebuild the parts holding you back — decided during architecture, not assumed at the pitch.
How long does a brokerage ERP take?
A focused first phase in 12 to 16 weeks. The extraction pipelines then extend supplier by supplier as you need them.
keep reading
the other sectors we work in.
got invoices nobody can face reading?
A thirty-minute call with the engineer who would lead the work.