We consolidate your data, build the models that forecast and score, and put the output where decisions actually get made. A working result in four to six weeks — reported in euros and hours.
The same reporting cycle, before and after consolidation.
Eurostat (2025) reports that 20% of EU enterprises with ten or more employees used at least one AI technology in 2025 — and the Banco de España's EBAE survey of 6,300 firms found that in the majority of those companies, use remains experimental.
The survey names poor data readiness as one of the three barriers holding adoption back, alongside skills and implementation cost. AI-powered data analytics is how Liorant closes that gap. This page describes what we build. The section below describes how you buy it.
Each one compounds the next. Consolidated data, models, and decision support convert all five into better visibility, better decisions, and better performance.
Data scattered across systems that never reconcile.
Dashboards exist, but nobody trusts the numbers.
Repeated calls made on gut feel, not evidence.
Spreadsheet forecasts that drift and quietly fail.
Teams optimising in silos, pulling against each other.
We build these when the decision they inform is repeated and consequential — not to demonstrate that we can.
CRM, ERP, marketing, operational and finance data reconciled into a single set of definitions, with KPI tracking and anomaly alerts on top.
Integrating CRM, ERP, marketing and operational sources, cleaning and standardising fields, and building structured datasets that hold up under analysis.
Demand forecasting, customer churn and behaviour prediction, lead scoring, and risk and anomaly detection — for decisions currently made on intuition.
Scenario modelling, driver analysis that explains which variables actually move the number, and alerts that fire when a threshold is crossed.
A distribution business ran monthly demand planning from a spreadsheet maintained by one person. In a five-week activation we integrated their ERP and sales data, built a forecasting model against 24 months of history, and delivered it inside the Power BI environment they already paid for. The larger gain was second-order: the planner spent the recovered time on supplier negotiation.
Not a pilot deck. A working output someone opens on a Monday morning.
Dashboards drift. Connectors break. Metric definitions change. Phase 2 monitors what we built, tunes it, adds one new build per month, and reports monthly on what it was worth. The terms sit inside the Phase 1 contract — no second sales cycle, no obligation to continue.
See how the fixed-scope activation works end to end — charter to a system in production.
See how Rapid AI Activation worksData Readiness for AI is the entry product for that situation. Two to four weeks, fixed price, ending with a measured result rather than a recommendation: we run a retrieval test using your own questions against your own sources, and score what comes back on a 0–3 rubric.
The output carries forward with zero rework. The taxonomy we build becomes the knowledge layer for whatever gets built next. You never pay twice for the same thinking.
Data Readiness for AIThe failure pattern isn't technical: nobody owns the thing, definitions drift, and by month six the numbers no longer reconcile. Inside AI Engineering as a Service, analytics work continues on a monthly cadence.
Pipeline failures, broken connectors, data-freshness checks, and model drift against the delivery baseline.
When someone redefines a metric, the change is documented and propagated — not silently breaking three reports.
Models retrained on a schedule appropriate to how fast the underlying behaviour moves.
A new report, model, or alert — drawn from the backlog scored at your last readout.
Hours recovered, decisions changed, errors avoided — in euros and hours. Not uptime percentages.
See how the ongoing operation is scoped and priced.
AI Engineering as a ServiceEvery Liorant engagement includes an option to do less.
If three departments report different revenue figures, no model fixes that. Agreeing one definition does — and that costs a workshop, not an engagement.
A forecast that improves a decision made twice a year rarely repays the effort of maintaining it. Judgement plus a good spreadsheet is often the right answer.
No modelling recovers information that was never captured. The honest recommendation is to instrument the process first and revisit in two quarters.
If Power BI's native forecasting or your CRM's built-in scoring covers the case, we'll tell you — and you won't have bought anything.
Tools you already licence carry no procurement cycle and no new vendor risk. Most engagements never leave this layer.
Scalable, secure storage for all your data.
Turn data into insights with powerful dashboards.
Build models, generate insights, and scale AI.
Orchestrate and automate data flows reliably.
Ensure quality, performance, and trust in your data.
Microsoft Copilot Studio, Google Gemini, and Claude — for natural-language access to data, automated narrative summaries, and analyst-facing assistants. Plus n8n and native connectors where data needs to move on a cadence.
Python-based modelling, custom ETL, retrieval systems over proprietary data, and MLOps monitoring — bespoke engineering, scoped separately. It usually starts with an AI Architecture Sprint so the decision is made with costed options.
Analytics that touches people carries obligations that analytics on inventory does not. We map this before building, in every engagement.
Most business analytics is minimal-risk. The threshold matters when a system evaluates people — creditworthiness, recruitment, employee evaluation, or access to essential services fall under the high-risk classification, phasing in through 2026–2027. Social scoring is prohibited outright.
Analytics on customer or employee data sits under GDPR regardless of AI Act classification. We review lawful basis, minimisation, and retention as part of the data assessment — not as an afterthought.
The 2025 AI Promotion Law establishes a national framework administered by ANIA. We align deliverables to it for Salvadoran engagements.
For US clients and partner-delivered work, we document against the NIST AI Risk Management Framework's govern, map, measure and manage functions.
If compliance is your primary constraint rather than a condition of the build, that is a separate product: AI Governance & EU AI Act.
Several clients run the retainer while hiring, then transition ownership when their team is ready. We document for that outcome from day one.
| Liorant | In-house data hire | Large consultancy | Self-managed BI tools | |
|---|---|---|---|---|
| Time to first working output | 4–6 weeks | 3–6 months to hire and onboard | 3–6 months, discovery-led | Immediate, if someone has the time and skill |
| What you get | A production analytics system your team uses, plus a scored backlog | Permanent capacity, built over time | A strategy, then a build phase quoted separately | A licence and a blank canvas |
| Who maintains it | Us, monthly, or your team after handover — your choice | The hire, if they stay | Nobody, after the engagement ends | Whoever has the time |
| Cost shape | Fixed project, then optional monthly | €30,000+ per year, fixed | Large, front-loaded | Licence only, plus internal hours |
| Compliance handled | Risk classification and documentation included | Depends entirely on the hire | Yes, at a price | No |
| When definitions change | Governed and propagated in the monthly cycle | Handled, if they notice | Out of scope | Reports break silently |
Four to six weeks from the signed charter to a system in production. If your data needs structuring first, add two to four weeks for Data Readiness — and we'll tell you that in the discovery conversation, not three weeks into a build.
No. Most engagements run against the systems and files you already have. We recommend a warehouse only when data volume or query patterns genuinely require one, and we cost that decision before you make it.
That's the common case, not the exception. Start with Data Readiness for AI. It ends with a measured retrieval score and a ranked remediation plan, and the output feeds directly into whatever gets built next.
Both, when the decision justifies it. We build forecasting, scoring and anomaly-detection models where a repeated, consequential decision is currently made on intuition. Where a well-structured report answers the question, we build the report and say so.
Operational analytics — forecasting, dashboards, performance reporting — is generally minimal risk. Systems that evaluate people for credit, employment or access to services fall under the high-risk classification, with obligations phasing in through 2026 and 2027. We classify and document every system we build.
Yes. The monthly operations phase has a three-month initial period and is then cancellable on 30 days' notice. We don't propose it where it's not warranted; if your deployment is small and stable, the honest recommendation is a clean handover.
Hours recovered, decisions changed, errors avoided, and what shipped — in euros and hours, co-authored with your internal owner. Not uptime percentages and not a dashboard screenshot.