Discussion Document · Confidential
Prepared for Astic Group · AIMI
Make the ERP
investment pay off.
Four operating processes inside AIMI where manual work can be engineered out — and an intelligence layer, built above your existing ERP, that proves itself before it ever connects to it.
/01 — What we do & how we do it
Lean Artificial Intelligence.
Most operations don't have an AI problem — they have a process problem. Automating a broken process just produces broken output, faster. So we work in a fixed order: understand first, redesign second, and only then apply technology to make the fixed process repeatable and efficient. Every intervention must do one of two things: cut cost or drive revenue.
Diagnose2 weeks · embedded
See how work actually happens.
We sit with your team and ride the real workflows — how a tender is estimated, how a prequal pack is assembled, how a PO gets raised. The output is an honest map of where hours and margin leak, quantified in dirhams, plus an audit of what your ERP already covers and where the genuine gaps are.
Engineerprocess first
Redesign the process before any tool.
Standard paths, named owners, governed rules — pricing floors, approval thresholds, document checklists. This is classical process engineering: making the work consistent and measurable. Most of the early gains come from this step alone, before a line of code.
Orchestratethen technology
Apply the tech stack that makes it repeatable.
On top of each redesigned process we propose and build the specific technology — structured tools, document generation, AI-assisted parsing and matching — that eliminates the manual work: the re-keying, the from-scratch rebuilding, the hunting for current numbers. Your team keeps the judgment; the system does the labour.
Our operating rule with your ERP
Your ERP stays the system of record.
We never touch it until you've seen us work.
You've invested in a custom ERP with CRM. We don't replace it, rebuild it, or write into it. We operate as an intelligence layer above it: a read-only mirror of the data we need, on our own parallel database, with a manual bridge back into the ERP. Recommendations flow to your people; your people act inside the ERP. Only after the intelligence has proven itself — and only with your sign-off — do we graduate toward assisted write-back and, eventually, governed integration. Worst case in phase one: our output is wrong. Your ERP cannot break.
Your ERP · system of record→
Read-only mirror→
Intelligence layer→
Recommendation → human acts in ERP
Revenue · Candidate 1 of 4
/02 — Process candidate
Estimation &
Quotation
The front door of AIMI's revenue. Every tender — insulation packages for industrial clients, fabricated ductwork for MEP contractors — starts with someone working out quantities and prices. Today that work is manual, slow, and inconsistent.
Why · the leak
Takeoffs are built by hand in spreadsheets against material costs that may be weeks old. Two estimators can price the same job differently. Slow quotes lose bids to faster competitors; stale costs mean jobs are won at margins that were wrong before fabrication started.
What · the redesign + tool
- One standard estimating path with a named owner and a turnaround SLA
- A live cost library — every material, gauge and fitting with current cost behind it
- A structured estimating tool that prices consistently, with margin visible as the quote is built
- Pricing floors and discount approval rules embedded in the flow
How · safely
The cost library and estimating tool live on our parallel database — a read-only mirror of product and price data from your ERP. Finished quotes are keyed into the ERP by your team as today. Nothing in your existing quoting or order flow changes until the tool has proven itself on live tenders.
ERP untouched · Phase A · manual bridge
Benefit · the outcome
- Quotes out in hours, not days — more bids submitted, more won
- Consistent pricing regardless of who estimates
- Margin protected at the point of quotation, not discovered at job end
- Estimator hours redeployed from arithmetic to judgment
Where the intelligence works · Estimation
NLP
Reads the tender so your estimator doesn't have to. Language models parse BOQs, spec documents and RFQ emails — however each client formats them — into structured line items, and match every line to your catalogue and cost library. The hours of transcription before estimating even begins disappear. → faster bids out, more bids won
CV
Extracts quantities from drawings. Computer vision reads dimensions, duct runs and pipe schedules from technical drawings to draft the takeoff for human verification — attacking the slowest, most skilled step of estimating. Phase B — deployed once the structured flow is proven · → estimator capacity multiplied
ML
Learns what wins. Models trained on your own quote history flag when a bid's price or margin is drifting outside the range that historically wins — before it goes out the door. → higher win rate at protected margin
GEN
Drafts the quotation document itself — commercial terms, exclusions, validity — from the priced estimate, in AIMI's format, ready for review. → quotation admin near zero
Every output is a draft a human approves. The estimator's judgment stays; the labour goes.
The question the diagnostic answers: how much of AIMI's order book is catalogue-based (standard ALFA products a client can order by SKU) versus drawing-based takeoff? The split decides whether the first build is a simple catalogue-and-price tool or a full estimating workbench — and it materially changes the size of the build.
Revenue · Candidate 2 of 4
/03 — Process candidate
Tender & Compliance
Documentation
AIMI's approved-vendor status with the region's most demanding industrial and government clients is a hard-won asset. But every submission — prequalification packs, vendor-approval renewals, material certificates, GCC manufacturer documentation — is assembled by hand, from scratch, each time.
Why · the leak
Hours of skilled time go into hunting down certificates, datasheets and approvals scattered across folders and inboxes. Worse: one expired certificate or missing document in a submission can disqualify a bid you were pre-approved to win. The leak isn't just labour — it's winnable revenue lost to paperwork.
What · the redesign + tool
- A central, versioned library of every certificate, approval, datasheet and test report — with expiry tracking and renewal alerts
- Standard submission templates per client type and spec
- A document generator that assembles a complete, correct pack from the library in minutes
- A pre-submission checklist gate: nothing goes out incomplete
How · safely
The document library and generator sit entirely on our layer — they read customer and project references from the ERP mirror and produce finished packs for your team to review and submit. No ERP process is altered; this fills a gap the ERP was never designed to cover.
ERP untouched · standalone gap-fill · Phase A
Benefit · the outcome
- Submission packs in minutes instead of days
- Zero bids lost to expired or missing documents
- Approved-vendor status actively protected — renewals never missed
- A permanent, reusable compliance asset that outlives any one employee
Where the intelligence works · Compliance
NLP
Reads the client's requirements, not just yours. Language models parse each tender's submission requirements and vendor-approval criteria into a checklist — so the pack is built against what this client demands, not a generic template. → zero disqualifications on technicalities
CV
Digitizes the paper. OCR and document vision read scanned certificates, test reports and approval letters — extracting issuer, scope and expiry date automatically, so the library builds and maintains itself instead of relying on someone's memory. → expiries never missed, renewals automatic
ML
Matches documents to demands. Classification models map each requirement in a tender to the right document in the library and flag the gaps — before submission, not after rejection. → winnable bids stop dying on paperwork
GEN
Assembles the pack. Generative document assembly produces the complete, ordered, formatted submission — cover letters included — in minutes, drawn entirely from verified library documents. → days of skilled labour → minutes of review
The library is the source of truth; the AI only ever assembles verified documents — it never invents one.
The question the diagnostic answers: which client approvals and certifications does AIMI hold today, where do they physically live, and how many near-misses or lost bids trace back to documentation? That inventory defines the library — and usually surprises leadership.
Cost · Candidate 3 of 4
/04 — Process candidate
Job Costing
Plan vs Actual
The estimate says what a project should cost. The question that decides profitability is what it actually costs — and in most fabrication businesses, nobody knows until the job is closed and it's too late to act.
Why · the leak
Material overruns, rework, and scope variations accumulate silently during fabrication and site execution. Variations the client should pay for go unbilled because nobody logged them. Margin erodes job by job, invisible until year-end — and the estimating team never learns which assumptions were wrong.
What · the redesign + tool
- Every won job opens with its estimate as a live budget
- Material issues, labour and purchases captured against the job as they happen
- A per-project dashboard: budget vs actual, updated continuously, variances flagged early
- A variation log with a hard rule — no variation executed without being priced and billed
How · safely
Cost actuals are mirrored read-only from the ERP (purchases, stock issues) into our layer, where they're matched to jobs and compared against the estimate. Dashboards and alerts flow to your managers; corrective decisions are executed in the ERP as today. The feedback loop also sharpens the cost library from Candidate 1.
ERP untouched · read-only actuals · Phase A→B
Benefit · the outcome
- Margin problems visible mid-job, while they can still be fixed
- Variations captured and billed — revenue that currently evaporates
- Every closed job improves the accuracy of the next estimate
- Manual cost-tracking spreadsheets eliminated
Where the intelligence works · Job costing
CV
Captures cost documents at the source. Document vision and OCR read supplier invoices, delivery notes and site paperwork — extracting amounts, quantities and references without anyone re-typing them into a spreadsheet. → cost capture without the clerical burden that kills it
NLP
Matches money to jobs. Language models reconcile invoices to purchase orders to goods received, and attribute each cost line to the right project — even when references are inconsistent, as they always are. → a true per-job P&L, continuously, not at year-end
ML
Predicts the ending while you can still change it. Models learn each job type's cost curve and forecast final cost from progress-to-date — flagging the projects heading over budget weeks before the overrun lands. Anomaly detection catches unusual material draw or rework spikes the day they happen. → margin defended mid-job, not mourned after
GEN
Writes the variation claim. When a scope change is logged, generative drafting produces the priced variation notice for the client from the job record — so billable changes actually get billed. → leaked revenue recovered
Forecasting accuracy grows with each closed job — the system compounds; a spreadsheet only ages.
The question the diagnostic answers: what does the ERP already capture about per-job cost, and at what granularity? If actuals exist but aren't job-matched, this build is fast. If they're not captured at all, the redesign starts on the shop floor.
Cost · Candidate 4 of 4
/05 — Process candidate
Procurement &
Inventory
AIMI's raw materials — cellular glass, PIR/PUR, phenolic foam, sheet metal — are imported, expensive, and slow to replace. Today, purchasing is reactive and re-keyed by hand, and cash sits in stock nobody is watching.
Why · the leak
When a job is won, its material requirements are re-typed into purchase orders line by line. Reordering happens when someone notices a shelf is empty — which means expedited freight on fast-movers and dead capital in slow-movers. Working capital is tied up precisely where it earns nothing.
What · the redesign + tool
- Won estimate flows straight into draft purchase orders — no re-keying
- Reorder points per material, set from real usage and supplier lead times
- Automatic alerts: below reorder point, aging stock, dead stock review
- A consolidated purchasing calendar per supplier instead of piecemeal POs
How · safely
Stock levels and purchase history are mirrored read-only from the ERP. Our layer generates draft POs and reorder recommendations; your procurement team reviews and raises them in the ERP exactly as today. The ERP remains the sole record of stock and purchasing throughout.
ERP untouched · draft-and-review bridge · Phase A→B
Benefit · the outcome
- PO re-keying eliminated; procurement hours redeployed
- Stockouts and expedited-freight premiums cut
- Working capital released from slow and dead stock
- Stronger supplier terms through consolidated, planned ordering
Where the intelligence works · Procurement
ML
Forecasts demand instead of reacting to it. Models learn consumption patterns from your own usage history — project pipeline, seasonality, job mix — and set reorder points that adapt, recommending what to buy, how much, and when. → stockouts and expedited freight down; dead stock stops accumulating
NLP
Reads the supplier's paperwork. Language models parse order confirmations, price lists and shipping notices from BASF, Dow, Pittsburgh Corning and sheet suppliers — updating expected costs and arrival dates automatically, and flagging price changes the moment they appear in a document. → cost library always current; margin never silently eroded
CV
Verifies what actually arrived. Vision and OCR on delivery notes and packing lists match received goods against the PO at the gate — catching short shipments and wrong items on day one, not at stock-take. → paying only for what you received
ML
Learns each supplier's real lead time — not the promised one — and builds it into reorder timing and job planning, so import delays stop surprising the shop floor. Phase B — needs order history on the mirror · → planning on reality, not optimism
Recommendations, not autopilot: every PO is drafted by the layer, reviewed by your buyer, raised in the ERP.
The question the diagnostic answers: how much cash is currently sitting in stock older than six months, and what did expedited orders cost in the last year? Two numbers the ERP likely holds — that nobody has put side by side.
/06 — How we engage
Prove it first.
Then subscribe.
We don't ask you to commit to a transformation on a document. The engagement is staged so that each step is earned by the one before it — and priced only when we both know exactly what's being built.
Step 1Diagnostic
2 weeks · fixed fee
Operating Diagnostic — the only decision on the table today.
Two weeks embedded across AIMI. We map the four candidate processes against reality, audit the ERP's coverage and data, resolve the catalogue-vs-takeoff question, and put a dirham figure on each leak. You receive a prioritized roadmap and a precisely scoped, precisely priced build plan — which you can execute with us or without us.
AED 10,000fixed · payable on kickoff · data access starts the clock
Step 2Build
milestone-based
Iterative build, delivered in working increments.
The highest-value candidate from the diagnostic is redesigned and built first — on the parallel intelligence layer, ERP untouched. Delivery is milestone-based: you see working software at each stage and pay against accepted milestones, not promises. Each subsequent candidate follows only when the previous one is proving itself in daily use.
Scoped & priced from the diagnostic — setup fee per milestone, agreed before any build starts
Step 3Run
subscription
Handover, then a subscription that keeps it alive.
Once a build is handed over — documented, trained, owned by your team — it moves onto a monthly subscription covering hosting, upkeep, cost-library and compliance-library maintenance, and continuous improvement. No lock-in: the process designs and playbooks are yours regardless.
Subscription quoted at handover · scales with what's live, not with headcount
ContraSix Management Consultancy – FZCO
w contra6.com o IFZA · Dubai, UAE
