NTK MIR · Russia · Kazakhstan
How a 400-person logistics group was rebuilt into an operating model that runs on AI.
NTK MIR was operating with fragmented data, disconnected accounting systems, manual tracking, handwritten vehicle-maintenance records, weak operational visibility, and ungoverned access. Ninety days later, one operating model ran across two countries.
- Client
- NTK MIR
- Industry
- Logistics and transport
- Scale
- 400 employees, 28 warehouses
- Markets
- Russia and Kazakhstan
$21.5M
Annual operational loss identified
28% of revenue
37
AI opportunities mapped
Across eight departments
90 days
Two countries, one operating model
May – August 2026
11 / 12
Enterprise Evolution steps
Delivered or in progress
01The company at zero
A $75M business running on a spreadsheet.
The founder rated operational efficiency at three out of ten. Shipments were tracked in a Google spreadsheet. Three accounting offices could not see each other's books. A hundred vehicles were maintained from handwritten notebooks. Two security officers policed twenty-seven regions with a pen and a fuel-card statement.
3 / 10
Operational efficiency
The founder's own rating
~400
Employees
~44 in sales
28
Warehouses
South-west to eastern Siberia
~13,000
Shipments per month
At ~$490 average order value
Gut instinct. That's all. Nothing else. I've been running on that for 8 years. At the level of sensation.
Five systemic problems generated almost all the loss
Department-level losses were symptoms. Every root cause crossed a departmental boundary — which is why point solutions inside single teams had repeatedly failed at this company before.
| # | Root cause | Annual scale |
|---|---|---|
| 1 | No unified system. Google Sheets, three 1C instances, personal Excel files, messengers and paper, with no single source of truth. | ~$6.9M |
| 2 | Sales managers were administrators. 50–65% of the working day went to status checks, triple data entry, manual quoting and notifications. | ~$4.6M |
| 3 | Critical knowledge lived in individual heads. Fleet, sales and warehouse each depended on one irreplaceable person. | ~$2.8M |
| 4 | Theft and fraud with no automated control. Fuel reconciled by hand, fake invoices, a branch running a business within the business. | ~$2.4M |
| 5 | Clients left because of operational chaos. Not price — accumulated friction across a handful of shipments. | ~$2.1M |
02Phase one
Four audits ran in parallel before a single system was designed.
The engagement opened with operational, organisational, legal and technical audits running simultaneously — each using a different instrument, and each designed to catch a different kind of failure. That sequencing is the entire case.
| Track | Instrument | Headline finding |
|---|---|---|
| Operational | Eight department-head interviews plus an external logistics audit | ~$21M a year, 37 AI opportunities |
| Organisational | Anonymous 82-response employee survey, six blocks | Strong human core, broken seams between departments |
| Governance | Compliance review against four statutes | 5 statutory blockers, up to $833K exposure |
| Data and technology | Source inventory, schema design, reverse engineering | ~25 spreadsheet tabs, 60,000+ counterparty name variants |
The order matters. Had the technical track run alone — as it does in most engagements — the company would have received a well-built system that was illegal to operate, staffed by people nobody had asked whether they could use it.
The legal audit killed the original architecture
Five components of the proposed system were outright statutory blockers, with combined exposure of up to $833K on a first inspection. The voice platform originally specified would have sent Russian citizens' voice biometrics to servers in the United States. It was replaced before a single call was placed.
The organisational audit found what the interviews could not
Eighty-two employees across sales, logistics and accounting answered an anonymous questionnaire before any system was designed. Trust inside teams scored 4.5 of 5, and over 85% understood how their work affects company results — a rare asset in a distributed team, and the reason change was possible at all. But cross-department handover was a systemic failure: all three departments named the other two as the primary source of operational breakdowns.
In the regions there is information starvation. We find out that something has changed at best in a chance conversation, and more often after the fact, when the absence of information has already become the problem.
03Phase two
Eight delivery waves — and AI arrived in week eight.
Delivery ran as eight overlapping waves between May and August 2026, presented here in delivery order because the sequence itself is the argument: infrastructure, then the system of record, then governance, then integration, then measurement — and only then AI.
| Wave | What was built | Why it had to come here in the order |
|---|---|---|
| 0 | Feasibility and infrastructure | Two cloud voice platforms were tested against the incumbent telephony and failed. The bridge architecture was proven before anything depended on it. |
| 1 | The registry — a single source of truth | Nothing else could be built on a spreadsheet. Nearly 20,000 shipments and three reference books had to move first. |
| 2 | Governance and the access model | A shared system without field-level permissions would have been less safe than the spreadsheet, not more. |
| 3 | The integration layer | Triple data entry was the single largest loss driver. It ends only when the systems talk to each other. |
| 4 | Control and measurement | Once every shipment was in one place with a price on it, the pricing itself became auditable for the first time. |
| 5 | Leadership visibility | The founder could not lead differently until he could see differently. |
| 6 | The AI layer | The voice agent needed a registry to read from, a calculator to call, and a legal basis to operate. All three existed by then. |
| 7 | Replication to Kazakhstan | The real test of an operating model is whether it transplants. It did, in weeks rather than months. |
The registry: the company's process, written down
19,828
Shipments migrated
From ~25 monthly spreadsheet tabs
9,500+
Counterparties normalised
From 60,000+ raw mentions
99.6%
Recognition accuracy
Counterparties; cities 99%
11
Roles and 11 policies
Field-level permissions
The registry is not a database that replaced a spreadsheet; it is the company's process written down in a form a machine can enforce. Conditional fields turned the warehouse measurement dispute into a data point instead of an argument. Nine status values for cargo and a separate payment lifecycle separated delivery from handover, because the audit found those two events were being conflated and hiding failures. Full revision history converted the recurring question “who changed this?” from an accusation into a lookup.
Governance built from what the survey revealed
Eleven roles and eleven policies at final count. Five of those roles are combined roles — Manager-Logistician, Manager-Operator, Logistician-Operator, Universal and Universal Plus — created because the survey revealed a truth head office had missed: in the regions, one person is the whole department. Twenty-two temporary administrator accounts, a security risk created by an earlier workaround, were retired in the same pass.
Integration: ending triple entry
The registry became the hub of a four-way exchange — 1C, the customer portal, a customer's own CRM, and a two-way mirror between the Russian and Kazakh registries. Each contract was specified in writing and handed to a different external development team, coordinated without any of them having access to the registry internals. A manager who once entered an order into a spreadsheet, then into 1C, then notified a client by hand, now enters it once.
04Where it landed
The shape of the transformation.
| Dimension | May 2026 | August 2026 |
|---|---|---|
| System of record | Google spreadsheet, ~25 monthly tabs, no permissions, no history | Governed registry, nearly 20,000 shipments migrated, full revision history |
| Access control | Everyone sees everything, or nothing | 11 roles, 11 policies, field-level permissions, combined-role model |
| Cross-system flow | Manual re-entry into 1C, three times per order | Live two-way 1C exchange — 42 orders created by 1C in a single day |
| Pricing control | Discounts invisible until someone read an invoice | Every quotation scored against the tariff engine automatically on save |
| Management visibility | Office manager assembling a summary by hand each morning | Analytics application: branch plans, rankings, month-end forecasting |
| Customer contact | 200 cold calls a day for the entire sales team | Voice agent quoting live freight prices, 23-test regression suite |
| Geography | Russia only; Kazakhstan on a separate spreadsheet | Two synchronised registries, mirrored international shipments |
| Legal posture | Architecture carrying 5 statutory blockers, up to $833K exposure | Localised stack — voice, models and parsing inside the perimeter |
Eleven of the twelve steps of the Enterprise Evolution System are delivered or in active progress. The twelfth — a permanent internal AI function — was recommended and specified by MagnaQore; the founder elected instead for MagnaQore to deliver the whole programme first. Ownership of that function is now being transferred in stages as confidence in the redesigned operating model grows and AI literacy spreads through the teams.
→Result
The outcome
- 19,828 shipments migrated into a governed registry
- 11 roles and 11 policies established
- Two-way 1C integration and customer portal connectivity
- Automated pricing control and quotation auditing
- Management analytics application
- AI voice agent for freight pricing
- One operating model running across Russia and Kazakhstan
- 11 of 12 Enterprise Evolution System steps delivered or in progress
About these case studies
These case studies are based on real MagnaQore client engagements. Certain company names, figures, technical details, and operational information may be simplified, anonymized, or aggregated to respect confidentiality obligations.
They are provided to illustrate the type of AI Organization Transformation work MagnaQore delivers. They do not constitute a guarantee of results, a public offer, or a complete representation of a client's internal systems, strategy, or financial performance.
Additional technical and commercial detail may be shared for qualified due-diligence purposes under a mutual non-disclosure agreement.
→Contact
Ready to move beyond AI experimentation?
Build the leadership, operating model, governance, and internal capability required to become an AI-native organization. Start with an Enterprise AI Transformation Assessment.
We work with boards, CEOs, CIOs, transformation leaders, and operating teams that are ready to move from fragmented AI activity to a durable enterprise capability.