P3How we deliver · UK manufacturers

How do you replace an old order system without stopping the factory?

Modernising manufacturers with AI and MCP

For a manufacturer running orders through an ageing in-house system and a lot of re-keying, we assess the system, rebuild it in stages without stopping production, join it to the ERP and couriers, automate order intake with AI, and add an MCP server so staff can query their own data.

Who it is for
UK manufacturers
Sector
Manufacturing and engineering
Size
Large company
Services
5 services, joined up
Line drawing of a factory conveyor feeding a dashboard of charts.

P3.1

The usual starting point

Orders arrive as emails, PDFs and spreadsheets and are typed by hand into a legacy system that only a couple of people still understand.

The order system, the ERP and the courier accounts do not talk to each other, so stock and delivery dates are checked by phone.

Managers wait on weekly reports to answer simple questions about orders, stock and lead times.

P3.2

The system at a glance

What we build, what it connects to, and who uses it.

Orders inemail, PDF, CSVAI intakematch and flagOrder clerkapprovesOrder systemmodern rebuildERPstock, pricingCourierslabels, trackingMCP serverread-onlyManagersplain questionsOrders inemail, PDF, CSVAI intakematch and flagOrder clerkapprovesOrder systemmodern rebuildERPstock, pricingCourierslabels, trackingMCP serverread-onlyManagersplain questions
From an email order to an approved order, and a plain question to an answerChoose a box to see what we do there.

What we buildAI intake

AI intake · match and flag

We build and look after this part.

The diagram in words
  • Orders in, email, PDF, CSV (existing system)
  • AI intake, match and flag (what we build)
  • Order clerk, approves (people)
  • Order system, modern rebuild (what we build)
  • ERP, stock, pricing (existing system)
  • Couriers, labels, tracking (existing system)
  • MCP server, read-only (what we build)
  • Managers, plain questions (people)

Connections:

  • Orders in → AI intake
  • AI intake → Order clerk (AI at work)
  • AI intake → Order system (AI at work)
  • Order system → ERP
  • Order system → Couriers
  • Order system → MCP server
  • MCP server → Managers (AI at work)
Orders to approve
Illustration Sketch of the approval queue: each order read by AI and matched to the customer and parts, with anything unusual flagged for a person.

P3.3

How we deliver it, stage by stage

Discovery, build, launch, grow and ongoing: each step and the service behind it.

  1. 01 · Discovery · Legacy assessment

    Code audits

    Review the legacy code, data and integrations, find the rules the business relies on that nobody has written down, and agree a staged plan that keeps production running.

    Touches: Order system

  2. 02 · Build · A staged rebuild

    Custom software development

    Rebuild the order system as modern software one area at a time, running old and new side by side and comparing results until each area is proven.

    Touches: Order system

  3. 03 · Build · ERP and courier integration

    API development and integration

    Connect the new system to the ERP for stock and pricing and to courier accounts for labels and tracking, with retries and alerts when another system is down.

    Touches: ERP, Couriers, Order system

  4. 04 · Build · AI order intake

    AI automation

    AI reads incoming orders in any format, matches customers, part numbers and prices, and queues each order for a person to approve, with anything unusual flagged.

    Touches: Orders in, AI intake, Order clerk

  5. 05 · Launch · Switch-over by area

    Custom software development

    Move each area across at a quiet point in the week with a tested way back, then retire the old system once the last area is proven.

    Touches: Order system

  6. 06 · Grow · Ask the data

    MCP server development

    Build a read-only MCP server over orders, stock and lead times, so approved staff can ask an AI assistant plain questions and get answers with links to the records.

    Touches: MCP server, Managers

  7. 07 · Ongoing · Tuning and new formats

    AI automation

    Review the orders people correct each month, use them to improve matching, and add new customer order formats as they appear.

    Touches: AI intake, Order clerk

P3.4

How the services connect

Why one partner across every stage matters.

  • Because we rebuild the order system ourselves, AI intake writes into it through a proper interface rather than a fragile screen-scraping bolt-on.
  • The same clean data model feeds the integrations and the MCP server, so an assistant's answer matches what the ERP and the courier show.
  • The assessment shapes everything after it: the undocumented rules it uncovers become tests the new system has to pass.

P3.5

How we work with your team

Governance
A monthly steering meeting with operations, finance and IT, a shared risk log, and a go or no-go check before each switch-over.
Data protection and security
Role-based access, a read-only MCP server with an audit log, customer data kept in the UK, and AI providers chosen for no training on the data.
With their team
Pair with the in-house IT team throughout, so they can run, change and extend the system themselves.
Change
Order clerks help design the approval screen and test each release before it goes live.

P3.6

What you get

  • Order clerks approve orders instead of typing them, and spend their time on the exceptions.
  • Stock and delivery dates come straight from the ERP and courier systems, without phone calls.
  • Managers answer everyday questions about orders and stock themselves, without waiting for a report.
  • The legacy system is retired without stopping production.

Built for small teams too. A smaller firm can start with AI order intake on its existing system, without a rebuild.

P3.8

What you can check

The standing promises behind this work, and where to read them.

  • Accessibility to WCAG 2.2 AA

    We design and test to WCAG 2.2 AA, by keyboard and screen reader as well as by tools, and help you keep an accessibility statement.

    How we work →
  • UK GDPR and DPIAs

    Personal data is part of the specification: what is processed, why, where and for how long. We help you run a DPIA when a project needs one.

    Security and trust →
  • Hosting region

    Hosting in UK data centres or on your own infrastructure, and AI providers set up not to train on your data.

    DevOps and hosting →
  • Handover documents

    Code in your repositories, plus a runbook and documentation written as we go, so your own team or another supplier could pick it up.

    Code ownership →
  • Support terms

    Support with agreed response times, monitoring and security updates, set out before work starts.

    Support and maintenance →
  • We run this ourselves

    DeskZap is hosted, monitored and supported by us, so the way we look after your system is how we look after our own.

    See DeskZap →

P3.9

Technology and timescale

  • TypeScript
  • PostgreSQL
  • Message queues
  • ERP and courier APIs
  • Large language models
  • Model Context Protocol
  • Role-based access

Staged modernisation over several phases, then monthly tuning.

P3.10

Services covered

T+99Next step

Talk it through with us

A free 30-minute first chat, with no obligation. Tell us what is getting in the way; we tell you honestly what we would do, roughly what it would take, and whether it is worth doing at all.