Prepared byIncredible
AI Discovery · RBPN AI Advisory Pilot · Scoping

The Mātanga Brain

Your knowledge - past tenders, NZ Standards, district plans, supplier specs, and what lives in the director's head - pulled into one LLM-agnostic Brain, with a Tendering Agent as the first tool it powers. This is a working view of what we heard and what we'd build, ranked together - nothing locked.

Capture the knowledge Win more tenders Extend it yourselves
The shape of it

Capture the knowledge, then win the work

You've chosen the Foundation Path: build the Brain first, then the Tendering Agent that sits on top of it, then extend the agent set yourselves. Delivery follows pain and impact, not strictly this stack order.

The AI Brain

Your knowledge, in one place - LLM-agnostic

We start from the HoDs workload tracker you already live in, then layer in standards (OneDrive), project knowledge (12D Synergy) and meeting insights. Agents read from the Brain rather than trawling raw PDFs.

Foundation

The Tendering Agent

The first tool the Brain powers

Drafts 80-90% of an RFP response from past wins, evaluator keyword patterns, council-specific tone and Mātanga's non-price attributes - with an Opportunity Scanner watching GETS and council panels daily.

Flagship tool

The agents you build next

Standards, suppliers, QA - self-built with coaching

Once the Brain is in place, your team extends it in Phases 3 and 4 - Design Standards & District Plan lookup, Supplier Availability, Design QA - each roughly 30-40% faster to build than starting cold.

Extensions

Delivery follows pain and impact, not this stack order - the Brain and Tendering Agent land together in Phase 2; the rest you extend yourselves with light coaching.

What we'd tackle

Priorities, in order

Ranked by impact against effort - tap to expand. The Brain and Tendering Agent are built together in Phase 2; the Standards agent is a fast follow your team builds. Nothing locked - we rank this together.

1

The Tendering Agent

Draft 80-90% of every RFP, and never miss one worth chasing
Highest painMedium build

The pain

  • Responses are the non-price attributes - methodology, technical capability, track record, weighted criteria - not pricing (Brad keeps AI away from pricing).
  • Historically 3-4 days locked away per tender; Copilot has cut that to ~2 hours, but it's manual, one question at a time.
  • Paste the whole RFP into Copilot and it loses the plot - the context window fills with non-question text and answers degrade.
  • Tone needs 2-3 back-and-forth rounds to stop it sounding generic.

What we'd build

  • A tendering skill that reads past tenders and your real differentiators from the Brain, controls the tone, and drafts a strong first-round response in one pass.
  • Scoped to the non-price written response - pricing stays with Brad.
  • Fed by a set of won / lost / want-to-do-more tenders, so it learns which work to chase and how you win it.
  • An Opportunity Scanner watching GETS and council panels daily.
"I used to block out my room for 3-4 days on a tender. With Copilot it's about 2 hours - but I go one question at a time." Live example: a Gisborne District Council RFP for 3.6km of new potable water main.Brad, on the call
2

The AI Brain

The foundation the Tendering Agent - and everything after - plugs into
FoundationBuilt in Phase 2

The pain

  • Knowledge lives in PDFs, spreadsheets, supplier sites - and the director's head.
  • Newer PMs are slow to reach senior-level accuracy.
  • Copilot is the default only because it's in Microsoft 365 - with accuracy concerns and a lock-in worry.

What we'd build

  • A central, LLM-agnostic knowledge layer, seeded from the HoDs Excel tracker you already run, then fed by OneDrive standards, 12D Synergy project folders and Fireflies insights.
  • Every agent queries the Brain, not raw documents.
  • Swap the underlying model (Copilot → Claude / Gemini) without rebuilding the agents.
The team wants to stay platform-agnostic - free to move off Copilot if a better model comes along. The Brain is what makes that possible, and it's why future agents build ~30-40% faster.Foundation Path rationale - Part 1 roadmap
3

Design Standards & District Plan Agent

Ask the standards a question, get the clause back
Daily frictionQuick to build

The pain

  • Standards are large PDFs bought from Standards New Zealand - NZS 4404 alone is ~265 pages, some run to 500.
  • Ctrl-F is keyword-fragile: searching "minimum pipe size" returns no match; you search "sewer" and scroll.
  • Varies by TLA too - no single national spec for sewer or stormwater. The highest-frequency pain; design is keenest.

What we'd build

  • Conversational lookup over the TLA standards from the Brain: "what's the minimum grade of a 150mm sewer pipe?" answered from the loaded standard.
  • Staff lift to senior-level accuracy faster; less knowledge trapped in one head.
  • Data is ready - standards are all PDF in OneDrive and Brad can share them now. Team-built with coaching in Phase 3.
The team is scratching the door down for this one - especially design. "We can only retain so much information, and there's so many standards."Brad, on the call
+

On the radar, not yet

The Foundation Path trades broad Phase-2 coverage for long-term leverage
Phases 3-4 / parked

Extend on the Brain later

  • Design QA agent - Brad asked for this by name: a first pass over a drawing set (~28 pages, ~2 hours by hand today) for spelling, punctuation and title blocks - and flagging overlapping or illegible numbers and leaders.
  • Supplier Availability agent - watches Humes, Hynds, Corey's for discontinued items.
  • Project Insights / knowledge-capture agent.
  • Email triage & drafting (Copilot-native).

Worth a live decision

  • The Cash & Margin dashboard - which speaks directly to your winter-cash pain (pain #2) - sits here rather than in Phase 2 under the Foundation Path.
  • Is indirect coverage OK for now, or do we carve a lightweight winter-cash view into scope?
Option A deliberately addresses one top pain head-on in Phase 2 (blind tendering); the others are covered as your team self-builds. The main honest trade-off is winter cash - let's decide together how to handle it.Foundation Path trade-off - to confirm with Brad
How we prioritise

Pain against ease of build

We target the top-right first - most pain, most doable. The Tendering Agent is the highest pain and a solid build; the Brain sits under it as the foundation; the Standards agent is a fast follow your team builds.

Pain / impact →
1Tendering Agent
2AI Brain
3Standards agent
+Future / parked
Ease of build →
Tendering Agent AI Brain (foundation) Standards agent Future / parked
Who connects to whom

The team, and how knowledge flows

A ten-person civil consultancy in Taupō - subdivisions, roading and water - with a Tuesday in-person HoDs hui that plans the week in two-hour blocks. Today the deep knowledge runs through Brad and the senior team; the Brain changes that. Toggle to see the shift.

Asks / waits for an answer Holds knowledge to capture

From tender arriving to submitted - and where the Agent plugs in

01

Download from GETS

Brad

Grab the RFP docs and drop them in OneDrive.

02

Summarise the RFP

Copilot

"What does the council want?" - a quick first read.

03

Answer the non-price questions

Brad + Agent

Methodology, technical capability, track record - each weighted question, drawing on past tenders. Where the agent plugs in.

04

Tone & review

Brad + Amy

2-3 tone rounds; Amy (business partner) reviews and flags "too much AI"; Brad finalises by hand.

05

Submit

Brad

Pricing stays separate - Brad prices it himself.

The people

Brad McKenzie

Director / owner

Runs tendering solo, carries the standards and past-win knowledge, and repairs client trust when details slip. The Brain's first job is to capture what's in his head so the team can draw on it. Knowledge feeder.

Civil designers (×3, names TBC)

Civil design · 12D

Own the design output; deep on TLA standards and detailing. The keenest for the standards agent - key feeders to the Brain.

Surveyors (×2, names TBC)

Surveying · Trimble

Trimble-based surveying; consume standards and TLA rules. Ask the Brain.

Project managers (×3, names TBC)

Delivery + PM check

Own client relationships and delivery, and the final PM check. A mix of feeder and consumer.

Amy (business partner)

Technical lead & co-owner

Brad's business partner and technical lead; reviews every tender and pulls him up if there's "too much AI" in it. A key feeder to the Brain.

Gemma (office manager)

Operations

Sets up every project in 12D Synergy from the folder template and owns the file structure. A process feeder.

Andrew (name to confirm)

Tender review

Reviews and comments on tender responses alongside Brad. A feeder for how a winning bid reads.

Where the knowledge lives

The systems, and how they fit

The Brain becomes the new home; your existing tools feed into it or flow back out. Some are core integrations, some are standalone for now.

The integration prize: 12D Synergy, OneDrive and the HoDs tracker

The knowledge sits in three places: project folders in 12D Synergy (a tidy, standardised structure - Procurement, Engineering & design, Delivery, Wash-up), standards as PDFs in OneDrive (ready to share now), and the HoDs Excel tracker you already live in. Seeding the Brain from the tracker and wiring in 12D Synergy is the biggest unlock - 12D Synergy access is the one to verify.

The new home

The AI Brain

The knowledge layer every agent reads from and writes back to.

The hub others feed into
Comms & AI

Microsoft 365 + Copilot

Teams, email and Copilot-in-Teams. Some staff also run their own Claude / Grok accounts.

Copilot stays; the Brain sits above whichever model
Project files

12D + 12D Synergy

All project folders - docs, Excel, drawings - built from one template, with document check-in/out and full version history. Structure: Procurement → Engineering & design → Delivery → Wash-up.

API via 12D Synergy Web - Flo to sort access
Accounting

Xero

Timesheets and budgets - projects set up with task-level budgets and % tracking bars for cost-vs-actual. Holds every past project.

Budget-vs-actual signal and the archive of past work - has an API
Tender pipeline

GETS + Lead Manager

Tender opportunity tracking.

The Opportunity Scanner watches these daily
Reference & IP · OneDrive

NZ Standards & TLA rules

Bought from Standards New Zealand (e.g. NZS 4404, ~265pp) plus TLA-specific rules - large PDFs in OneDrive, a few Word contract docs. Brad can share now.

Ingested into the Brain - powers the Standards agent
Capture

Fireflies

Meeting capture across the business.

Meeting insights feed the Brain
Supplier & panels

CoLab / Council panels

Preferred-supplier panel access.

Feeds the Supplier Availability agent later
Brain seed · Excel

HoDs workload tracker

A manual Excel sheet: current projects, ones ending, and priced jobs awaiting client approval - plus each person's week in two-hour blocks. Whole team visible; Brad delegates from it at the Tuesday hui.

The seed the Brain is built from
Survey

Trimble Business Center

Surveying.

Standalone for now
Website

Rocket Spark

Public website.

Standalone
Fleet

eRoads

Vehicle and fuel tracking.

Out of scope
What we still need to know

Open questions

A few answers sharpen the roadmap. We work through these together - some we'll knock off on this very call.

The Brain & your data

?
Brad drafts RFPs in his own OneDrive - can we get access to that tender working area?that's where the past responses the agent learns from actually live
?
Is 12D Synergy Web configured? The API needs it - Flo to advise.it's the gate to reading project files programmatically
?
Do the PDF standards retrieve accurately, or should we get them as CSV / Excel / a database?500-page PDFs may need reformatting for reliable answers
?
When Standards New Zealand issues an update, how do we keep the Brain current?standards change and are re-purchased; the Brain must not go stale

The Tendering Agent

?
Can you pull together a set of won / lost / want-to-do-more tenders?Matt asked for this - it teaches the agent which work to chase and how you win it
?
Beyond GETS, which councils and panels matter most for the scanner?focuses the daily watch on the work you'd actually chase
?
What are your non-price differentiators - the angle behind methodology, technical capability and track record?the agent needs your voice, not boilerplate

Platform & LLM

?
Claude vs staying on Copilot - Brad leans Claude if the benefit is clear; we'll A/B test and show the cost.the Brain is model-agnostic, but this sets the starting platform
?
Any data-residency or privacy constraints on tender or client data?shapes where the Brain and its data can sit

Adoption & the team

?
We have Gemma (office manager) and Amy (technical) - can we get the other designers, surveyors and PMs by name?the design team is the day-one champion; we want a feeder / consumer map with real names
?
What does the team's winter (build-season) load actually look like?enablement fits the quiet season - we want to land it there

The winter-cash question

?
The Foundation Path doesn't build the Cash & Margin dashboard in Phase 2 - is indirect coverage OK, or do we carve a lightweight winter-cash view into scope?it's your #2 pain and the main honest trade-off of Option A
What we're banking on

Working assumptions

These shape the estimates; the first is the big lever. We flag confidence honestly and verify before committing.

Can the core systems be read programmatically? (API / MCP)

Standards are PDFs in OneDrive, reachable via the Microsoft Graph API. Xero has a mature API. 12D Synergy has an API too, gated behind its Web module - Brad is connecting us with Flo at 12D Synergy to sort access. Fallback if the Web module is not enabled: a scheduled export / import from the project folders.

Option A - Foundation Path is the chosen scope

Phase 2 = the AI Brain + Tendering Agent (with Opportunity Scanner), inside the 200-hour / $25,000 +GST envelope, RBPN 50/50 co-funded.

Confirmed

Standards access is sorted

In OneDrive and Brad can share now - but they are large PDFs (up to ~500pp), so we will test whether PDF retrieval is accurate or a better format is needed.

Confirmed

The team is on board

No adoption resistance expected - the team is keen, "scratching the door down", the design team especially.

Confirmed

The tendering time-saving is real and shown

Copilot already took Brad from 3-4 days to ~2 hours per tender, one question at a time. A skill reading the Brain should beat that - a strong first draft in one pass.

Confirmed

Winter is the build and enablement window

Phase 2 to complete by end of August 2026, ahead of the summer build ramp when the team has no slack.

High confidence

Past tenders are recoverable and can be labelled won / lost

Draft quality on day one depends on a usable back-catalogue of bids with outcomes attached.

To confirm

A default model - leaning Claude

Brad: "if we're investing in this, we probably want to go to Claude." Copilot Studio caps the model at what Microsoft predefines; Claude lets us switch model per task and manage cost. We'll A/B test and show the cost before locking it in.

To confirm

12D Synergy API via the Web module

12D Synergy has an API, but it needs the Web module configured - Brad is connecting us with Flo at 12D Synergy to confirm and sort access.

To confirm
Where we go from here

Next steps, and who owns what

The next move is a deep-dive on access - getting us to the standards, tenders and project files so we can start building. Most of the homework sits with your side, because it's your data.

Incredible will

  • Set up Copilot Studio and run early A/B tests - Claude vs Copilot - then show you the difference and the cost.
  • Get in touch with Flo at 12D Synergy about API access (needs the Web module).
  • Work out the SharePoint / OneDrive access route for your tenders and standards, with Amy.
  • Test whether the PDF standards retrieve accurately or need a better format.
  • Start building the Brain from the HoDs tracker, and the Tendering Agent on top.
  • Line up the access deep-dive call in about two weeks.

Mātanga will

  • Share the standards set (the PDFs from OneDrive).
  • Pull together past tenders - won, lost, and ones you'd like more of - so the agent learns the commonality.
  • Give us access to Brad's OneDrive tender working area.
  • Introduce us to Flo, your 12D Synergy contact.
  • Sort out SharePoint access.
  • Send the org chart / onboarding flowchart if it's handy.
  • Decide on Claude vs Copilot once we've shown cost and benefit.

The follow-up

  • Next call: an access & assumptions deep-dive in about two weeks - we'll show what we're pulling from the data and check it against what you see.
  • Between now and then: Mitch and Brad / Amy to liaise async on access.
  • Team homework: each time you think "would the Brain have saved me here?", jot it down.