We apply science to the data problems holding your business back.
Your business decides with confidence, spends smarter, and scales what's actually working.
Most firms start with an approach
We start with the problem, not the tool.
AI shops feed everything to a language model. Traditional data science teams reach for the same statistical playbook. Either way, the tool is chosen before your problem is understood. They give you a solution shaped by their capability, not your requirement.
Our approach
Every engagement begins with a diagnostic. First we understand how your business works, what's actually known, what's uncertain, and what a quality solution genuinely requires. The approach follows from that.
We draw on statistical modelling, causal inference, data engineering, mathematical optimisation, and ML/AI engineering in whatever proportion the problem calls for. Most of the time that means the simplest, most reliable combination that works.
Example A
For one client, we replaced a demand forecast running inside a language model with a purpose-built ML model thereby improving accuracy and removing token cost completely. The end result was more interpretable, easier to monitor and fit for purpose.
Example B
For another, we built an anomaly detection agent where the language model handled natural language diagnostics and email reporting, while deterministic tools underneath it did the actual detection. Each layer doing only what it's genuinely good at.
Where we use AI agents, deterministic logic and traditional models sit beneath them as callable tools.
In regulated or lower-complexity environments, we favour self-hosted models over foundation models entirely. The environment shapes the design, not the other way around.
Diagnose before prescribing
The problem structure determines the approach. Never the reverse. We don't arrive with a preferred solution, we arrive with the discipline to understand your situation first.
Every component earns its place
If something can be solved deterministically, it should be. AI is extraordinary at some things and unreliable at others. We don't use it where it doesn't have a genuine advantage.
Rigour built in at every layer
A composed solution is only as strong as its weakest part. We hold the data, the model, and the output to the same quality standard throughout.
Tested, secure and documented
Every solution is tested against real data before it reaches production. Security, reliability, and maintainability are not afterthoughts, they are part of the specification from day one.
Customer retention: predicting and preventing churn
A scaling business wants to know which customers are at risk and intervene before they leave. Here's how we'd design the solution and why.
Data integration & pipeline
Pull together behavioural signals, transaction history, support interactions, and product usage into a single clean customer record. A churn model is only as honest as the data beneath it. This isn't where AI helps — it's where engineering discipline matters.
Survival analysis as a tool
Model time-to-churn using a proven statistical framework that respects censored data and gives interpretable, calibrated probability estimates. Survival analysis was built for this exact problem structure. Feeding raw data to a language model here would produce confident-sounding noise. We turn it into a tool that the LLM can call.
AI-powered interpretation
Translate model outputs into plain-language account summaries: why this customer is at risk, what the signals are, what the intervention options look like. Generative AI excels at synthesising structured outputs into something a commercial team can act on without a data science degree.
Communication & delivery
Set up automated scheduled jobs and email integration to share the text summaries with the team to action. Deterministically built and tested with no need to burn tokens on every call.
Inventory costs reduced through optimization
Manufacturers are faced with outdated reorder rules that ignore actual demand patterns, and excessive material waste resulting in tied-up working capital and waste.
Read moreDemand forecasting through predictive inventory modelling
Supply chain teams often react to stockouts rather than anticipating them. By the time a product shows signs of depletion, the reorder window had closed.
Read moreA/B testing for feature performance measurement
Product team ship features without a reliable way to measure their impact. Decisions are then made on intuition and incomplete metrics.
Read moreAgentic AI system for operational decision workflows
Businesses make key markdown and promotion decisions through manual review of spreadsheets. The process is slow, inconsistent, and didn't scale.
Read moreDiagnostic
A structured review of one business problem you are currently facing and what it would take to solve it using data-driven approaches or AI Automation. We choose the best approach for your specific business and document a clear roadmap detailing how to solve the problem. We will provide no-obligation pricing on implementation during the diagnostic.
- Review of existing data and measurement infrastructure
- Gap analysis, determining what's reliable, what isn't and what's missing
- Prioritised roadmap for the next step
- 45-minute feedback session with Q&A
Project
A defined implementation with a clear deliverable. The methodology depends on the problem including causal inference, mathematical optimisation, machine learning, AI engineering, Data modelling, data pipeline or a combination. You receive something working and repeatable.
- A working, tested deliverable such as a model, system, analysis, automation or integration
- A tested, documented handover so your team can use and maintain it
- Findings presented to relevant stakeholders
- Documented approach where applicable
Retainer or Maintenance
Models need attention as conditions change. Agentic systems need monitoring and improving. Ongoing retainer keeps your analytical infrastructure current and extends the relationship to new problems as they arise.
- Quarterly model refresh with updated data
- Ongoing monitoring and alerting
- New problem evaluation as the relationship develops
- Monthly check-ins
Not sure which applies? The Diagnostic is the right starting point, it defines the problem before any approach is chosen.
Built inside real scaling platforms. Production-grade and reliable.
Lerilo is a boutique data science practice specialising in data-driven solutions to your most pressing business problems.
The founder spent ten years as a Data Scientist focussed on challenging real-world business problems, optimization and automation. Not advising. The person in the data, writing the code, building production models, presenting findings to the C-Suite and VPs on decisions that mattered.
Ranging from in-depth analysis to support high-stakes business decisions to AI powered automation. Actually implementing real high-impact solutions that most consultants can only talk about.
Business first
We bridge the gap between business and deep technical work, presenting results in a way a business audience can understand and act on.
Rigorous
Every component is held to the same standard. We don't cut corners on data quality, model validation, or output integrity.
Actionable
We deliver findings that can be acted on immediately not presentations that require translation by another team.
Proven results
From warehouse slotting to marketing attribution, we implement solutions that most consultants can only talk about.
Every engagement starts with a conversation about your problem, not a pitch about our capabilities.
Start a conversationThe first conversation costs nothing and clarifies everything.
Tell us about the business problem you are trying to solve.
Message received.
I'll be in touch within one business day.
Based in Ottawa, Ontario, Canada and working with companies across Canada and the US.
Mon – Fri, 9am – 6pm Eastern