GLO25 · Method · Algorithm Audit

From hidden decision rules to EBITDA drivers

Every business runs on algorithms nobody formally designed: how a price is set, when stock is bought, who follows up a lead. The Algorithm Audit makes those rules visible and gives each one a verdict. The EBITDA X-Ray then shows what each fix is worth in the margin bridge.

Start a Decision Intelligence Scan Algorithm Audit™ product page

By · Published 27 September 2026 · Updated 4 October 2026

The map

Rule → verdict → lever → driver → EBITDA

Six domains of decision rules pass through one verdict each. Every Fix or Kill becomes a lever with a FIND-code, lands on one EBITDA driver, and only counts as upside on top of a normalised, adjusted EBITDA.

Six business-algorithm domains pass through a Keep, Fix or Kill verdict, map to FIND-code levers, feed four EBITDA drivers, and roll up from reported to adjusted EBITDA plus identified upside. 1 · BUSINESS ALGORITHMthe rule as it runs today 2 · VERDICTper rule 3 · FIND-CODE LEVERwhich module fixes it 4 · EBITDA DRIVERwhere it lands in the bridge 5 · EBITDAnormalised, then uplifted Pricing & tariff settingdiscounts, margin floors, quotes PRICPricing & revenue optimisation Sales & lead follow-upwho is chased, which customers SALE · ICPGo-to-market · customer targeting Procurementwhen, how much, from whom PROCProcurement / supply chain Planning & operationsscheduling, routing, depot use OPEXOperations efficiency Capacity & structurestaffing, roles, approvals HRCA · ORGCapacity · organisational design Data & systemsERP rules, reports, handoffs TECHDigital efficiency, data quality KEEPworks: protect FIXleaks: repair KILLharmful: remove Fix + Kill becomeEBITDA levers Pricerealised rate per unit Volume & mixhow much, to whom Direct costsmaterials, labour, logistics Overheadstructure, systems, allocation ReportedEBITDAas in the accounts normalise AdjustedEBITDA one-offs, owner andrelated-party itemsremoved; eachadd-back sourced + fixes Identifiedupside each € tied to onerule, one lever,one owner verified quarterlybefore it counts Rows 1→4 show the typical path. A single rule can touch more than one lever (e.g. PRIC+SALE, OPEX+PROC); the audit records the exact path per client.
How to read it

Four steps, no assumptions

  1. Make the rules visible. We write down each decision rule as it runs today, per domain, from interviews and the real data. Nothing is assumed.
  2. Decide per rule. Keep, Fix or Kill. No vague improvement points: one decision per rule.
  3. Price the fix. Each Fix or Kill gets a FIND-code, a cause, an annual € impact and a source label. It lands on one EBITDA driver in the margin bridge.
  4. Start from real EBITDA. Reported EBITDA is first normalised to a defensible adjusted EBITDA. Upside is only added on top of that base.
Where this sits in the journey

Diagnose, then playbook, then monitor

Act 1

Diagnose

Algorithm Audit + EBITDA X-Ray. Rule inventory, Keep/Fix/Kill, normalised baseline, priorities.

Act 2

Value Playbook

Quantified EBITDA bridge, initiatives ranked by € impact × feasibility, owners, milestones.

Act 3

Monitor & Attribute

Algorithm Watch / dashboard: expected vs actual per driver, quarterly validation.

Evidence discipline

Every number carries a label

MeasuredEstablished from reconciled evidence
IdentifiedUpside found and sized, not yet delivered
ScenarioModelled under stated assumptions
VerifiedDelivered, measured and signed off
FAQ

Algorithm Audit and EBITDA — common questions

What is a business algorithm?

A decision rule a business runs on that nobody formally designed: how a price is set, when stock is bought, who follows up a lead. GLO25 groups them in six domains: pricing, sales, procurement, planning and operations, capacity and structure, data and systems.

What does the Algorithm Audit deliver?

An inventory of the decision rules as they run today, with one verdict per rule: Keep (works, protect), Fix (leaks, repair) or Kill (harmful, remove). Each Fix or Kill gets a FIND-code, a cause, an annual € impact and a source label.

How is the EBITDA impact calculated?

Reported EBITDA is first normalised to a defensible adjusted EBITDA: one-offs, owner and related-party items are removed and each add-back is sourced. Identified upside is only added on top of that base, with each euro tied to one rule, one lever and one owner, and verified quarterly before it counts.

How are numbers labelled?

Every number carries one of four labels: Measured (established from reconciled evidence), Identified (upside found and sized, not yet delivered), Scenario (modelled under stated assumptions) or Verified (delivered, measured and signed off).

Further reading

Auditing AI and machine-learning models

This page covers business decision rules. For AI and machine-learning models specifically, Sandu, Wiersma and Manichand propose an audit framework built on model risk management, covering the full algorithm life cycle in the language internal auditors already use.

Sandu, I., Wiersma, M. & Manichand, D. (2022). Time to audit your AI algorithms. Maandblad voor Accountancy en Bedrijfseconomie, 96(7/8), 253–265.

Related

Go deeper

Find the rules that move your margin

Start with the decision you are facing. We reply by email with the next step.

Start a Decision Intelligence Scan