Our work

Analytics, optimization and AI solutions we built for enterprise clients.

Çimsa

Çimsa S&OP: from global network optimization to the fuel procurement decision

We brought Çimsa's production, logistics and sales network across 15 locations into a single mixed-integer model, then built a second model that turns the resulting clinker production plan into a fuel recipe and procurement decision.

10% ↓ Inventory cost
3% ↓ Manufacturing, warehousing, distribution cost

Derimod

Derimod: from data warehouse to Optail — three layers, seven projects in three years

We brought data scattered across AX ERP, Excel and CRM into a single data warehouse, built more than 20 Power BI dashboards on top, and placed Optail — initial allocation, replenishment and inter-store transfer decisions — at the top. Then came stock analytics — lost sales, size-run breaks, idle stock — and a CRM layer. This page is the umbrella; each of the seven projects has its own page.

20+ Dashboards
7,000+ Optail scenarios

Turkish Basketball Federation

TBF: a decision support system that builds referee and evaluator assignment on rules, fairness and proposals

A system that turns the Turkish Basketball Federation's Central Referee Board's weekly assignment work into mixed-integer optimization respecting federation rules and fairness between referees. The system proposes; the board reviews, edits by hand and commits. Referee assignment is in production pilot, evaluator assignment was added as a second problem, and the multi-user web interface is under way.

Proposal Decision
9 Hard rules

Defacto

Defacto DSO: four years from the founding of a data science office

We started in 2021 as mentors at the hiring Datathon; it continued to the end of 2024 with the team's training, the DSO's database and server infrastructure, and more than thirteen projects written and put into production. Today the DSO is a directorate serving Defacto's retail analytics, AI and business intelligence needs.

4 years Duration
13+ Projects

Defacto

Defacto Norm Staffing: calculating the staff each store should have, from data

A model that takes the store staffing decision away from actual payroll and subjective requests and grounds it in measured workload and an efficient peer in the same segment. It produces a 2025 norm, a 2026 target and an increase / reduce / keep decision for every store × role — end to end on BigQuery, fully parametric.

Store × role Decision unit
Frontier Reference

MMK

MMK: line balancing and order–raw material matching in flat steel production

An optimization system that matches open orders to the right raw material while respecting production line capacity and every stage of the bill of materials. Built across three phases.

4,000+ Open sales orders
~9 Raw material candidates per order

Gürmen Group

Gürmen MATS: moving inter-store transfers from Excel to a system that decides at SKU level

The Inter-Store Transfer System that moves products not selling in one Ramsey or Kip store to a store that sells them, before markdown. It selects receivers and senders by cover at option level, derives need from sales velocity, matches at SKU level, and runs block, single-unit and split-and-spread transfer types on one infrastructure. Delivered as a SQL project on Gürmen's server; results go to the ERP.

SKU Decision unit
Cover Receiver / sender

Kuveyt Türk

Kuveyt Türk: master–apprentice mentoring for two analytics teams — from digital marketing to retail customer analytics

It began in 2023 with training for Digital Marketing's analytics team: SQL, Power BI, Python, machine learning. The team then built its own projects under Lumtify's mentoring — digital score, target management, churn cause analysis, UI usage analysis, product recommendation and churn prediction models. From 2025, spending analytics, a luxury spend score and next-best-action prioritisation with the Retail Customer Analytics team. The team writes the code; Lumtify sets direction, brings method and checks the output.

Master – apprentice Model
8+ Digital projects

Kalekim

Kalekim: from MVP to a forecast management system — three phases in three years

We started with a machine-learning MVP for two product groups fed from Excel; in year two it became a forecasting process retrained every month for every product, and in year three the Kalekimforecast application that planners run themselves. The forecast is now an input to Kalekim's supply, production and logistics planning.

2 groups → all Scope
3 Encoding methods

Motor Aşin

Motor Aşin: from two ERPs to one data warehouse, from dashboards to decision support

We brought data split across Logo Tiger and J-Platform into a single SQL Server data warehouse through SSIS, and built sales, stock, product, warehouse fill rate, budget, salesperson scorecard and purchasing dashboards on top — some of them built by Motor Aşin's own team under mentoring. Data has flowed without interruption since October 2024; phase three in 2026: basket analysis, stock-outs and lost sales, idle stock, supplier performance.

2 ERPs Sources
Uninterrupted Data flow

Hayat Kimya

Hayat Kimya: weekly e-commerce order forecasting on Microsoft Fabric

A system that forecasts marketplace and e-commerce orders weekly, at the granularity the warehouse shift plan needs, and runs end to end as a single Fabric pipeline from data warehouse to report.

Weekly Forecast horizon
One pipeline Automation

Defacto

Defacto HR Analytics: from dashboard to alert, from alert to decision model

A second period with Defacto after the DSO: consulting for Human Resources' own analytics team. Turnover, demographics, academy and HRBP dashboards; time-and-attendance and bonus automation; an alert setup connected to the company's internal assistant; and the Norm Staffing model at the backbone. Ongoing since December 2025.

2nd period Period
4 Layers

Türkiye Finans Participation Bank

Türkiye Finans: training and mentoring for the HR analytics team — from recruitment assessments to employee attrition

One day a week in 2024 with the participation bank's HR Analytics team: Python and machine learning training, then mentoring on the team's own projects. The relationship between the recruitment assessment inventory and performance and retention was analysed by job family; a literature review and model design were done for employee attrition prediction. The method learned was applied to the institution's own data at once.

1 day a week Format
Python · ML Training

Kale Seramik

Kale Seramik: a demand forecasting PoC — and why we did not continue

An MVP that started on the same roadmap as Kalekim: two product functions, 34 methods, three encoding approaches. The result fell short of the target and the second phase was not started. This page explains what did not work, and why.

141 DFUs
200+ Product attributes

Mayadem

Mayadem: from game events to a ClickHouse data warehouse

Two pieces of work in two periods for Mayadem, a studio building children's games and content apps: event design and marketing dashboards for TRT Çocuk Oyun Dünyası in 2022; and in 2025, for the MagicPages reading app, a ClickHouse analytics warehouse that brings PostgreSQL and S3 together — join-free array-based dimensions and fact tables that answer in seconds. Continues as data-organisation consulting.

7 → 1 Book dimension
130+ Event definitions

Turkish Basketball Federation

TBF: the Basketball Management System's database — architecture and migration from the old system to the new

While the Turkish Basketball Federation was writing a new Basketball Management System with its own development team, we designed the database the system would run on and moved the old system's 454 tables and five million records into the new structure module by module. A naming standard, 1,685 column mappings, 226 foreign keys, document export, cut-over migration and documentation. The BYS the federation uses today runs on this database.

454 → 121 Tables
1,685 Column mappings

Upily

Upily: discovery and audit of the Wonjo Kids data infrastructure

The Firebase, Adjust and RevenueCat data of Wonjo Kids, a games app for children, had been gathered in BigQuery but had grown organically into complexity. Across seven areas — table inventory, pipelines, cross-platform matching, platform parity, taxonomy, event tracking, dashboards — we examined more than 1,500 BigQuery objects and 330 events, and delivered strengths, risks and a four-phase transformation roadmap in a single report.

1,500+ Objects examined
330+ Events

Karaca

Karaca: data analytics consulting — from the data warehouse to CustoMation

Consulting alongside the data analytics team in 2022–2023 for the store and e-commerce data of the Karaca group — Karaca, Karaca Home, Emsan, Homend, Kaşmir: the warehouse's analytics layer and cube, a product and idle-stock dashboard, store and e-commerce lost sales, a capacity system design, a transfer proposal, and CustoMation, which selects campaign audiences by RFM segment. This page is the umbrella; CustoMation has its own page.

6 Projects
5 Brands

Banks Association of Türkiye (TBB)

TBB: an end-to-end data scientist programme for bankers

Within the Banks Association of Türkiye's training programme, a 62-hour, 21-day 'Data Analytics Journey' for bank employees who use data in their work: introduction to data science, SQL, Python, statistics and EDA, visualisation, machine learning, industry cases. Evening sessions, homework for every module, hands-on work in Colab; every module was scored out of 10 in TBB's participant survey — averaging above 9.

62 hours Duration
9+ / 10 Evaluation

Defacto

Defacto IofTheCustomer: an NLP system turning hundreds of thousands of customer reviews into sentiment, category and action

Every product review from defacto.com and the marketplaces passes through sentiment analysis within seconds; if negative, it is assigned one of nine categories and routed to the responsible team. Built in 2021, running without interruption for three years, Stevie award winner in 2023.

9 Categories
700K+ Reviews processed

Defacto

Defacto Sorting: which product sits where on the category page — the brain behind smart sorting

A dynamic scoring algorithm that decides the order in which products appear when a category page opens on defacto.com. It weighs sales, clicks, broken size runs, stock and shelf life together, with a cold-start solution for new products and scenario-based weighting.

5 Signals
Cold start New products

Defacto

Defacto ELOSS: measuring the sales lost to stock-outs in e-commerce

A weekly calculation answering 'what would have sold had the product been there' for every SKU that runs out of stock on Defacto's own site and the marketplaces. It computes lost sales from pre-stock-out velocity with special-day and trend effects, then splits them by day using the daily sales distribution. An incremental design cut a calculation that took hours down to minutes.

Weekly Run
~30 min Runtime

Defacto

Defacto NoSales: measuring stock that has not sold for 28 days, every week, by the same rule

A weekly system that checks, for every option × store pair with stock on Sunday, whether there was a sale in the last 28 days, and reports idle stock by store, product and region. Scheduled queries in BigQuery, a five-page Power BI dashboard, a trend since 2021.

28 days Rule
Monday 05:30 Run

Defacto

Defacto PAT: meeting analytics that survived the move from Teams to Meet

A dashboard producing attendance, talk-time and meeting-efficiency metrics from online meeting data. Built on Microsoft Teams' relational data; when the company moved to Google Meet, Meet's nested event data was mapped to the Teams schema and the dashboard continued unchanged. Row-level security means everyone sees only their own hierarchy.

2 Platforms
Daily Refresh

Defacto

Defacto DISCO: if I give this product that price, how many units sell in 14 days?

A system that grounds the markdown decision in a forecast: an XGBoost model learning, for every product whose price changes, the units sold in the next 14 days from stock, broken sizes, sales velocity, price history and calendar effects; a pipeline that runs every day, an error measured against actuals 14 days later, and a demo interface where the planner tries five psychological prices side by side. The MVP was completed; by management decision it was not taken into the pricing process.

Price → units Question
Two-way Direction

Derimod

Derimod data warehouse: from AX, Excel and CRM to a single analytics database

The layer everything at Derimod sits on. We brought AX ERP's transaction-level data, the teams' Excel files and the CRM together in one analytics database on SQL Server: schema and naming standard, procedure registry and log mechanism, daily flow with SSIS, uninterrupted loading with partition switch, delta detection with RowHash. Both the dashboards and Optail are fed from this warehouse.

3 Source systems
7 Schemas

Derimod

Derimod business intelligence: 20+ Power BI dashboards fed from one warehouse

We turned the clean data in the warehouse into dashboards every Derimod team can reach and read on its own: sales, product, customer, budget, store and warehouse capacity, logistics, e-commerce operations, Optail success reports. Built along a four-step catalogue — reporting, deep analysis, trading, CEO — all fed from the same warehouse; no report has a data source of its own.

20+ Dashboards
4 Catalogue steps

Derimod

Optail Initial Allocation: first distribution of new product with a master plan, store grouping and capacity

When a new product leaves the warehouse there is no sales data yet; the decision rests on similar products' history, store groups and capacity. For Derimod we built the Master Plan screen where planning is done before and throughout the season, store clusters, fill and dispatch priority management; footwear and bags went live first, then apparel. In 2026 Flow Through, open-quantity allocation and simulation are being added.

3 Categories
Master plan Decision input

Derimod

Optail Replenishment: automatic warehouse-to-store feed for what sells

Replenishment, live since October 2024, manages the warehouse-to-store flow by rate of sale: how many weeks of cover, minimum lot, which warehouse, which store group. In the first two months two thirds of replenishment came from Optail; in 2025 it approached 95% in footwear. Whether every replenishment turns into sales is measured — Optail's sell-through rate is higher than manual and than last year.

~95% Automation
~1.5× Sell-through

Derimod

Optail Inter-Store Transfer: size-run consolidation, out-of-collection and season-end transfers

Optail's first module. It started with block transfer in February 2024; on top came the consolidation transfer that gathers broken size runs, the out-of-collection transfer that moves product to stores that do not carry it, closing-store clear-out, and the class-based transfer that concentrates product in selling stores at season end. Transferred goods sell through within 21 days at five times the manual rate.

~25% Sell-through
5 Transfer types

Derimod

Derimod stock analytics: lost sales, size-run breaks, idle stock and the success of every shipment

The layer that measures whether Optail's decisions work. We defined lost sales, size-run breaks and non-selling stock and tied them to weekly dashboards; made stock in warehouses and stores and every product movement — in transit, on dispatch order, incoming — visible; computed the sell-through of initial allocation, replenishment and transfer with a first-in–first-out link. In one season footwear lost sales fell from 14% to 10% and size-run breaks from 10.6% to 8.4%.

14% → 10% Lost sales
10.6% → 8.4% Size-run breaks

Derimod

Derimod CRM analytics: customer 360, a campaign success datamart and dashboard

Once the stock side had settled, the customer's turn came. We connected customer, membership, points and campaign data from Dynamics CRM, Shopify and the ERP to the same data warehouse; in place of campaign results summarised in a single line, we built a star schema at receipt-line grain — MASS and loyalty campaigns, points earned and spent, by channel, segment and product. Daily segment and points updates are live; the customer 360 view is a prototype; churn and next best offer are on the roadmap.

Receipt line Grain
2 Campaign types

Karaca

Karaca CustoMation: campaign audience selection by RFM and behavioural segments

We took the question of who a campaign goes to out of Excel filters and into a single tool. CustoMation selects customers for the Karaca group's five brands and eleven channels by RFM segment, brand-level behavioural segment, product group purchased, brand and recency window; include and exclude rules are defined together, every selection is saved as a scenario, past runs are compared, and the audience goes to the campaign tool as Excel.

10 + 19 Segments
8 Filter dimensions