Your operations
have optimal
parameters.
We find them, and we prove they hold on data the optimizer never saw.
Regnor builds a simulator of your operation, hands it to autonomous AI agents that run thousands of logged experiments, and delivers parameter recommendations validated out-of-sample, the discipline most optimization vendors skip.
Verified
optimization.
Anyone can run an AI optimization loop and report a big number. The big number is usually wrong: it's fitted to one lucky scenario, or it exploits a flaw in the model rather than a truth about your business.
Our early pricing research produced a spectacular 77.5% improvement built on a "loss-leader" strategy. When we audited our own work, we found the result was an artifact of the simulation. When we fixed the model and re-tested on unseen scenarios, the optimal strategy inverted: the product our agent had priced at cost belonged at its price ceiling.
We published the correction, rebuilt our methodology, and made the audit discipline the product.
Strategies are optimized on training scenarios and scored on held-out scenarios. The number we quote is the held-out number.
Our optimizers see only what your business could see: prices, costs, constraints, outcomes. Structure is discovered from behavior, never read from the model's internals.
Every experiment is logged at the moment it runs, with tamper-evident IDs and strategy hashes. The winning configuration is always in the log.
Before we deliver a number, an independent review tries to break it. What survives, ships.
Every claim on this page traces to a public experiment log in our research repository.
Five domains.
One discipline.
Anywhere you have a measurable objective, historical data, and controllable parameters, the same method applies. Inventory and pricing are demonstrated domains with published research. The others are engagement-ready: same method, simulator built during the engagement.
Cost reduction · held-out scenarios
Reorder points, safety stock, and order quantities optimised against your actual demand patterns and cost structure. Full perishable goods support with FIFO batch tracking, shelf-life-aware ordering, and waste-vs-stockout trade-off analysis.
See the research · 4,463 logged evaluations →Profit improvement · held-out scenarios
Portfolio price discovery across your catalogue: cross-price substitution, customer segments, volume discounts, and bundles. The agent finds demand-cliff thresholds behaviorally, and every recommended price set is scored on scenarios the optimizer never trained on.
See the research · out-of-sample validated →Shift staffing and capacity plans matched to demand patterns, minimising labour cost while meeting service level targets. Same method as the demonstrated domains, with the simulator built and calibrated during the engagement.
Order quantities, supplier selection, and discount-tier economics optimised against your purchasing history. Same method as the demonstrated domains, with the simulator built and calibrated during the engagement.
Diminishing-returns modelling per channel with optimised budget splits. A proposed application of the method: the simulator is scoped and built as new work during the engagement, never sold as an existing result.
Done for you.
Delivered in a week.
A fixed-scope engagement. You share your data. We calibrate a simulator, run the autonomous optimisation loop, verify the result on scenarios the optimizer never saw, and deliver actionable parameter recommendations. No new software required.
You share historical data (a CSV export is enough). We build and calibrate a simulator of your operation and validate it reproduces your actuals.
Autonomous agents run thousands of budgeted, logged experiments against training scenarios, keeping what works, reverting what doesn't, and logging everything.
Recommendations are scored on held-out scenarios, adversarially reviewed, and delivered as a report with the full audit trail: what to change, expected impact with uncertainty ranges, and the structural limits where parameter tuning stops helping.
One person.
Full accountability.
Four years in global market research: structured intelligence models, data-driven analysis across technology, manufacturing, healthcare, and financial services. Then the same rigour applied to operational optimisation.
Every engagement is scoped, built, and delivered personally. No handoffs, no junior analysts, no outsourcing. You work directly with the person who built the system, and who published the corrections to his own research when the audit demanded it.
Connect on LinkedIn →Held-out numbers.
Full audit trails.
These results are from our published research simulations: progressively harder synthetic environments built to enterprise-realistic constraints. Client engagements calibrate the same simulators to your historical data. Full experiment logs and reports available on request.
A 12-product catalogue with variable lead times, quantity-discount tiers, and perishable goods under FIFO expiry. Baseline $244,072 → optimized $140,064, mean of five unseen demand scenarios, with a train-to-test gap of just 0.7%. The improvement is structural, not overfitting. Every evaluation is in the audit trail.
An 8-product, 12-month portfolio with cross-price substitution, customer segments, volume discounts, bundles, and competitor reactions. £824,963 → £953,725 on scenarios the optimizer never trained on, with fully behavioral price discovery, including finding demand-cliff thresholds to within £0.50 without ever being shown them.
Our earlier single-scenario research reported larger numbers, up to 74% on the simplest inventory model. The audit that produced today's methodology is documented publicly, including the LinkedIn corrections we posted on our own articles. We sell the discipline, not the hype.
What others provide.
What we provide.
Most approaches to operational improvement rely on guesswork, generic benchmarks, or tools that require you to do the work. This is different.
Straight answers, before you ask.
Every engagement is reviewed personally. If the method doesn't apply clearly in the scoping call, we say so and don't take the engagement.
Not yet. They are from our published research simulations, and we say so wherever they appear. Real-data calibration is exactly what a pilot engagement does. Early design partners get preferential terms in exchange for an anonymized case study.
You don't get a guarantee. You get the same protection we apply to ourselves: recommendations validated on scenarios the optimizer never saw, uncertainty ranges instead of point promises, and an audit trail you can inspect. We also tell you the structural floor, the point where further tuning cannot help and operational change is needed.
Engagements are fixed-scope and fixed-price, quoted after the scoping call. Pricing depends on the domain and the state of your data. Productized tiers arrive with the self-serve tools. Start an enquiry and we'll give you a number before you commit to anything.
Historical operational data: sales history, costs, inventory levels, pricing records, schedules, or routes depending on the domain. Minimum 3–6 months. CSV, Excel, or direct export from your existing system.
5–7 working days from the moment we receive your data. Total timeline from first contact to delivered recommendations: typically 2–3 weeks including the scoping call.
We scope before we commit. If the pattern doesn't apply clearly in the scoping call, we say so and don't take the engagement. We don't take projects we don't expect to deliver meaningful results on.
No. You receive a recommendations document with optimal parameter values, a full experiment log, and structural findings. You implement the parameters in whatever tools you already use: ERP, spreadsheet, Shopify, or otherwise.
Client data is encrypted at rest, transmitted over HTTPS only, and deleted from all systems within 90 days of delivery. We act as a data processor under a signed DPA. No data is shared with third parties.
Start with
one problem.
Share your data. Define the metric. We scope the engagement in one call and tell you exactly whether the pattern applies and by how much.