Redacted sample · AI Profit Roadmap

Northwood Components Ltd

Specialist contract manufacturer based in the West Midlands. £18m turnover, 84 staff, dual UK and EU customer base across aerospace tier-two and precision automotive. Currently mid-ERP-selection, running Sage 200, a paper-and-spreadsheet shop floor, and a HubSpot CRM that no one in operations opens.

Prepared for: Operations Director, Northwood Components Ltd
Prepared by: The founder, The Margin Labs
Date: 27 May 2026

This is a fictional, anonymised composite based on the kind of business we typically work with. Numbers, names and details have been edited. The structure, format and depth are the same as the document you receive after your discovery call.

Executive summary

The on-record problem you brought to the call was quoting velocity (RFQs taking five to nine working days, with win rate slipping when you push past the front of the queue) and planning trust (your weekly production plan no longer survives Monday lunchtime). Both of those are real. Neither of them is, primarily, an AI problem.

The reframe is this: your operating model has outgrown the tooling underneath it. Quoting, planning and quality are running on tribal knowledge, three spreadsheets and a daily standup. Bolt a language model onto that and you get faster wrong answers. Build the spine first (one source of truth for the work, one place buyers acknowledge a job, one daily read on WIP and on-time performance), then put AI where it pays. That order matters more than the AI itself.

We recommend three quick wins your team can run inside the next two weeks with no consultant involvement. We then recommend six ranked opportunities sequenced over twelve months, in the order that pays back fastest, with a clear stop after the third while you complete ERP selection. We also list three things AI is not the answer for yet at your size and data maturity, with a date to revisit each.

The combined annualised value across the six ranked opportunities lands at approximately £189,000 against a phased investment of approximately £41,000, for a year-one return north of 4x. Costs assume SME-grade tooling (Katana-class MES at a small-team tier, low-code build on your existing CRM and Sage 200, Metabase rather than a paid BI stack) and a delivery model that uses your team for change with a thin layer of outside help, not a full-service consultancy. Values are anchored to published benchmarks for UK specialist manufacturers in your turnover band (Make UK insights, BCG operations work, Gartner MES reference programmes); most opportunities ramp toward higher steady-state values in year two. The first quick win pays for itself inside a fortnight.

Three quick wins

Week-one actionable. No consultant required.

The first three things any team can run inside the next two weeks. Pulled out of the roadmap so they pay you back before the bigger work begins.

Daily WIP digest from MES extract + Sage 200

Replace the 07:30 production manager's manual whiteboard snapshot with a one-page automated PDF, emailed at 07:15 to the ops triangle. Pulls last-shift job movements from your existing shop-floor printouts (scanned overnight) and matches them against the open works orders in Sage 200. Two days of joinery in a low-code tool, one day of layout. No new licences.

Hours saved: 6/wk
To implement: 1 wk
Est. cost: $2,000

Quote-on-receipt acknowledgement template

A standard reply, sent within four working hours of every RFQ landing in sales@, that confirms receipt, restates the buyer's spec back to them in your words, asks the two clarifying questions you always end up asking on day three, and gives a date for the priced quote. Stops the buyer chasing, surfaces missing info early, and signals seriousness. Owned by sales support, not estimating.

Hours saved: 4/wk
To implement: 1 wk
Est. cost: $600

Engineering drawing tag library for buyers

Build a shared tag library (material, tolerance class, finish, certification, takt band) that buyers apply when uploading drawings to the quote folder. Two afternoons with your senior estimator to define the taxonomy, then a one-page guide for the team. Cuts the back-and-forth on missing context, and it is the foundation every later automation will need.

Hours saved: 5/wk
To implement: 2 wks
Est. cost: $1,000

Costed roadmap

Six opportunities, ranked by payback.

Sequenced in the order that pays back fastest. Each opportunity carries an estimated cost, a payback window and an annualised value. Risk is a directional read, not a score.

Estimated annualised value

Six opportunities, ranked by payback. Values in USD.

Total annualised: $240,000

Unified quoting workspace

Low risk

One workspace where every live RFQ sits, with the drawing, the buyer thread, the costed BOM, the routing draft and the approval state all on one screen. Replaces the four-folder, three-inbox model. Cuts quoting cycle from five-to-nine days to two-to-three, and gives sales a real-time view of the quote book. Cost reflects a low-code build on your existing HubSpot and Sage 200 (around six to eight days of consulting effort plus a couple of months of part-time tuning); we are not selling you a platform. Value combines three to four additional won quotes per year at a £40k average order (around £30k to £40k of incremental contribution at a 25 per cent gross margin) and roughly three hours per week reclaimed across two estimators (around £10k of capacity).

Payback: 2 months
Est. cost: $10,000
Annual value: $70,000

Production planning and WIP visibility

Medium risk

Lightweight MES (Katana class on top of your existing Sage 200) with sequenced jobs visible on the shop floor, finite-capacity load by cell and real-time WIP. Restores trust in the weekly plan and shortens reaction time when a job slips. Sequenced second so the quoting spine is already feeding it clean job data. Cost is the realistic year-one delivered price for a small contract manufacturer (Katana annual licences for ten to fifteen shop-floor users plus around three weeks of implementation effort), deliberately not the enterprise Tulip configuration. Value reflects a two to three per cent recovery of direct-labour capacity on roughly £6m of direct cost (around £35k to £50k captured at gross margin) plus a measurable reduction in expediting and overtime as the plan stops slipping.

Payback: 3 months
Est. cost: $14,000
Annual value: $70,000

Quality and first-article inspection digitisation

Low risk

Move FAI records and in-process inspection from paper to a tablet workflow tied to the works order. Pulls drawing tolerances in automatically, flags out-of-spec on entry, archives a signed PDF against the job. Cuts FAI cycle by roughly half and makes customer-audit prep a one-click pull rather than a two-day scramble. Cost is a few shop-floor tablets, low-code build, and integration with the works order. Value is around 0.4 FTE of quality-admin time (roughly £15k at UK rates), a six to eight per cent reduction in non-conformance and rework cost (around £12k to £15k on a business your size), and a step-change in customer-audit readiness that protects retainer revenue from your aerospace and automotive tier-twos.

Payback: 2 months
Est. cost: $7,000
Annual value: $38,000

AI-assisted RFQ triage

Medium risk

A small model that reads each new RFQ, extracts spec fields against your tag library, scores fit against your sweet spot (material, volume, tolerance, lead time), and routes to the right estimator with a draft cost band. Sits behind the quoting workspace, not in front of the customer. Worth doing only after opportunities 1 to 3 are in flight, because it depends on the tag library and the unified workspace existing. Cost is mostly build effort; the LLM API itself is in the low hundreds per month. Value is around two to three hours per estimator per week reclaimed (about £10k across two estimators) plus one additional won quote per year from better routing (around £8k).

Payback: 5 months
Est. cost: $10,000
Annual value: $28,000

Supplier scorecards and early-warning

Medium risk

Auto-generated monthly scorecards (on-time, on-quality, on-spec, RMA rate) for your top forty suppliers, plus an early-warning rule that flags drift before it shows up as a missed delivery to your customer. Pulls from goods-in records and the quality module on the data spine assembled in opportunities 1 to 3. Built in Metabase, which is open-source, so there is no licence in this number. Year-one value is conservative because most of the benefit accrues over twelve to eighteen months as supplier behaviour changes; the year-one figure reflects one to two avoided late deliveries to end customers and modest negotiation leverage on the bottom-quartile supplier set.

Payback: 4 months
Est. cost: $5,000
Annual value: $19,000

Connected dashboard: P and L, WIP and on-time

Low risk

One board-level dashboard that ties contribution by job-family to WIP value and on-time delivery, refreshed nightly. Built in Metabase on top of the spine assembled by opportunities 1, 2 and 3, so the marginal build is a few days of dashboarding rather than a new platform. Gives the leadership team one number to look at on Monday morning rather than three reports that disagree. Value here is deliberately conservative because the benefit is faster, better leadership decisions, which is real but soft. We measure it as ten to fifteen hours per month of senior time reclaimed and roughly one avoided commercial misfire per year.

Payback: 5 months
Est. cost: $6,000
Annual value: $15,000

No-go list

Where AI is not the answer yet.

Three things we recommend you do not buy or build right now, with the reason and a date to revisit. The honest no-go list is part of why this document is useful.

Customer-facing chatbot

Your buyers are senior engineers placing low-volume, high-spec work. They want a named person who knows their parts, not a chat widget. A chatbot here would lower perceived quality and would not deflect work your team should not already be doing. The right answer is the quote-on-receipt template plus the unified quoting workspace.

Revisit in12 months

AI-generated CAD for safety-critical assemblies

Generative-CAD tooling is improving fast for low-criticality consumer parts. It is nowhere near the accuracy or auditability required for the tier-two aerospace and precision automotive work that is your margin engine. The liability picture is also unresolved. Worth tracking, not worth piloting.

Revisit in18 months

Demand-forecasting AI on the last twelve months of data

You do not have enough clean disaggregated history to train a forecasting model that beats your estimator's gut. You also have a structural change (post-tariff EU mix shift) inside that history, which would teach a model the wrong pattern. Revisit once the data spine from opportunities 1 to 3 has produced eighteen months of clean per-customer demand.

Revisit in9 months once the data spine is in

Architecture fit note

The spine, in five layers.

The recommended spine is deliberately small and deliberately yours. Sage 200 stays as the financial system of record. A lightweight MES (Tulip or Katana class) sits over the shop floor to capture job state in real time. HubSpot remains the CRM, but is finally wired to the quoting workspace so sales and ops see the same job. An integration layer (n8n or make.com) moves records between them. A single reporting surface (Metabase or Power BI on top of a small warehouse) gives leadership one read of the business. No vendor lock-in at any layer, no rip-and-replace. This pattern is intentionally compatible with whichever ERP you land on at the end of your current selection process: the MES, the integration layer, and the reporting surface all survive an ERP change with minor reconfiguration rather than a rebuild.

CRMHubSpot
FinanceSage 200
Shop floor (MES)Tulip / Katana class
Integration busn8n / make.com
Reporting surfaceMetabase / Power BI

AI Governance Starter

A one-page acceptable-use policy, four sections.

Before any AI tool ships to anyone outside the leadership group, you want a one-page acceptable-use policy that the whole company has read and signed. Not because it eliminates risk, but because it makes the trade-offs explicit, gives your team a clear answer when a customer or auditor asks how you use AI, and protects the people on the shop floor from carrying a decision that should have been made at the top. Four policies below cover the floor of what you need.

Acceptable use

Which tools are approved (named), which are prohibited (named), and what categories of work each approved tool may be used on. Personal use on personal devices is out of scope; work use on any device is in scope. One page, plain English, signed annually as part of the staff handbook update.

Data handling and confidentiality

Customer drawings, pricing, and personal data may not be pasted into any tool that retains prompts for training. Default to UK or EU data residency. Customer-data segregation rules from your existing GDPR policy apply, with one addition: an AI tool is treated as a sub-processor and must be on the approved list before any customer data touches it.

Model selection and audit log

The approved-tools list is reviewed quarterly by the operations director. Each approved tool has a named internal owner, a stated purpose, and a log of significant decisions (chose, dropped, changed config). Six months from now this log is the answer to the inevitable customer-audit question of how you govern AI use, and you will be glad you started it now.

Human-in-the-loop for safety-critical decisions

Any decision that touches a part on a flight-critical or vehicle-critical assembly is reviewed and signed by a named human before it leaves the building. AI output is allowed as input to that review and must be labelled as such. This is the rule that lets you defend a yes-to-AI position to your aerospace and automotive customers without losing them.

Want your own?

Your roadmap, your numbers, in a week.

We deliver your own AI Profit Roadmap within seven days of a thirty to forty-five minute discovery call. Yours to keep, whether or not you ever work with us.

Get my free AI Profit Roadmap