The second facet of the Enterprise Parasite model, the debt below the waterline where the cost of neglect stops being theoretical, and where your AI ambitions live or die.
Every enterprise believes it has a data problem. Most have the opposite. They are not short of data. They are drowning in it, and starving for anything they can trust.
Data debt is the second facet in the Enterprise Parasite model, it sits below the waterline, and it is where the cost of neglect stops being theoretical. A single question exposes it. Ask five systems who your largest customer is, and you will often get five different answers. Those five answers are the surface of a debt that has compounded for years, taxing every decision and integration, and now every AI initiative built on top of it.
The tip of the iceberg, the applications you can see and budget for, is the cheapest part. Gartner projects that by 2027 roughly 80 percent of technical debt will sit below the waterline. Data debt is a large share of that hidden mass, and unlike a failing server you often don’t know it’s there. It shows up in the gap between what your reports claim and what is actually true.
Start from the other end, though, and the picture inverts. When data is organized and trusted, it stops being a liability and becomes the most productive asset the enterprise owns. Decisions get faster because no one is relitigating whose numbers are right. Teams stop rebuilding the same reconciliations and start building on each other’s work. New capabilities like analytics and AI can be layered on with confidence, because there is a solid foundation beneath them. Trust compounds: the more an organization relies on data it can believe, the more value every new use of that data returns.
Poor governance breaks that trust, and the damage is easy to miss because the data still looks fine. Untrusted data behaves like counterfeit currency. It circulates alongside the real thing, gets spent in reports and decisions, and corrupts every transaction it touches. Eventually people stop accepting it. They keep private spreadsheets they trust more than the system of record, and they call meetings just to reconcile numbers that should have agreed in the first place. The result is worse than a slowdown. The enterprise keeps moving at full speed on information it has stopped believing.
The Four Faces of Data Debt
Data debt has four faces, and they compound on one another:
1. Redundant and orphaned data
Across structured, semi-structured, and unstructured stores, the same records get copied again and again until no one owns or remembers the originals. It is an attic filled with unlabeled boxes. You keep paying to store all of it precisely because you can no longer tell what is inside. By the time anyone asks what is in the boxes, the honest answer is that nobody knows.
2. No master or common data: the keystone
Without master data management, every application keeps its own version of customer, product, or patient. Reconciling them becomes a permanent tax. Analytics cannot be trusted, and AI has nothing reliable to learn from. When five systems disagree about a basic fact, the organization does not have a single source of truth. It has five competing rumors. This is the keystone of the entire facet: its absence is what forces every other kind of data debt to persist, and, as the architecture deep-dive shows, it is what pushes integration into ungovernable point-to-point sprawl.
3. ROT and dark data
A large share of stored enterprise data is redundant, obsolete, or trivial, much of it never classified and never monitored. Veritas Databerg research, a vendor-sponsored estimate, has put that figure near 85 percent. Whatever the exact number, the pattern is there. You are paying to store and secure a vast volume of content that delivers no value, while quietly widening your compliance exposure. Point AI at it and the noise is much worse. Garbage in, garbage out, now at machine speed.
4. Decaying, un-curated data
Data does not sit still. Contact data decays 25 to 30 percent a year, and only about 30 percent of enterprise data is reliably high-quality. The rest erodes the decisions made on top of it. Chasing and reconciling that decay consumes close to 30 percent of the knowledge worker’s week. That is the reconciliation tax, and almost no one budgets for it.
The Unstructured and Semi-Structured Blind Spot
Most data governance stops at the relational database. The debt does not.
Beyond the relational database sit two more kinds of data, and they are not the same thing, which is where most audits go wrong.
The largest and least governed layer of the modern enterprise is unstructured data, which has no predefined model at all. It is raw, native-format content, and it makes up the large majority of what an enterprise generates, roughly 80 to 90 percent. The contracts, emails, policy documents, transcripts, knowledge-base articles, and slide decks that hold the working knowledge of the business carry no schema to catalog, and the fastest-growing part of it is multimedia: the recorded presentations and video where facts are asserted out loud rather than written down. A renewal date lives in a paragraph. A commitment lives in an email. None of it can be queried with SQL, and reading it takes natural language processing, computer vision, or vector search, which is why it stays ungoverned in most organizations.
Semi-structured data does not live in rigid rows and columns, but it is not shapeless either. It carries its own organization: tags, keys, and metadata that name the fields and set a hierarchy a machine can parse without a fixed schema. JSON moving through APIs, XML and HTML, YAML configuration files, and the records in NoSQL stores like MongoDB and Cassandra are all semi-structured. That built-in flexibility is what makes it drift. Fields get added or nested differently from one service to the next, and with no relational constraint holding the line, the same customer or product ends up described several ways.
The result across both layers is a photocopy of a photocopy. A number is stated in one deck, lifted into the next, and repeated in a proposal, a briefing, and a board slide, and by then no one holds the master copy up to the light. Capabilities get overstated. Two teams cite the same figure three different ways. The errors do not surface when the document is written. They surface later, in front of a customer or an auditor, when the distance between what was claimed and what is true finally becomes visible.
It also happens to be where generative AI reaches first. Ask a model to summarize what the company does, and it will ingest the decks and the documents, not the database. Ungoverned content is how yesterday’s unvalidated claim becomes tomorrow’s confident, automated answer.
This is also where the fix has to become intelligent. Traditional, rules-based cataloging cannot read a slide or interpret a contract, which is why this layer has stayed ungoverned for so long. McLean Forrester uses UnicornIQ as the rationalizing engine for this content. For semi-structured data, it resolves the self-describing fields across services to one representation. For unstructured content, it reads the material the way a person would, pulls out the facts each document asserts, and tests them against the ratified master data, and reconciles whatever does not match. The layer that was ungovernable because nothing could parse it becomes governed because something finally can. The Source of Truth now reaches past the database to the decks and the drives.
The Keystone: Why One Fix Resolves the Rest
Pull the four faces together and a single dependency emerges. Redundant copies exist because there is no authoritative original to defer to. Reconciliation is a permanent tax because there is no ratified answer to check against. The claims buried in documents drift for the same reason. Each of these traces back to one missing thing: a governed source of truth.
Master data is the keystone. In an arch, the keystone is the single wedge that holds every other stone in place. Remove it and the structure collapses into rubble. Remove master data and integrations collapse into point-to-point spaghetti, analytics into contradiction, and AI into confident error.
The fix is an active Source of Truth: a rationalization platform that takes in master data from across the enterprise, ratifies it into one authoritative version, and hands that trusted data back to the systems that need it. Master data flows in. Ratified data flows back out. That is what turns master data from an aspiration into something the enterprise can rely on.
Master data is also the hinge to the layer above it. Its absence is precisely what forces systems into ungovernable point-to-point integration, so data debt and architectural debt rise and fall together. How that connective layer is structured, and how a governed hub replaces point-to-point sprawl, is the subject of the companion deep-dive on Enterprise Data and Systems Architecture.
What Data Debt Costs
Untrustworthy data does not stay inside the data team. It reaches customer experience, boardroom decisions, and the AI roadmap alike.
The AI Reckoning
For a decade, data debt was tolerable. It made analytics slower and dashboards less trustworthy, but the business could route around it with enough human judgment. AI removes that slack.
AI is only as good as the foundation it stands on. It needs clean, governed data, and it treats whatever it is given as ground truth. Fragmented, contradictory data does more than slow an AI initiative. It teaches the model to be confidently wrong, then scales that error to every user who trusts the output. The debt that was survivable before AI becomes the thing that decides whether AI works at all.
There is an upside hidden in that constraint, and it compounds in your favor. Data that is governed and consolidated into a trusted master-data hub is precisely the substrate that grounds vertical AI in facts the business actually believes. The same foundation that reins in redundant storage and reconciliation cost is the foundation that makes AI reliable. Remediation stops being pure cost. It becomes the platform for the capability everyone is racing toward.
Data Rationalization: Two Engines, One Source of Truth
Naming the problem is the easy part and lots of people and companies tell you that it needs to be fixed, but few tell you how. Rationalization is the work of fixing it: reducing a sprawling, redundant estate to what is trustworthy and needed, then grounding what remains in a governed Source of Truth. Tooling can find the problem. Deciding what is actually true takes judgment. The work runs on two engines, one for structured data and one for everything else.
Structured data: find the connections, then choose the source
The structured estate is the databases and the systems that read from and write to them, and you cannot rationalize it safely until you can see it. What hides it is not the data so much as the web of connections around it. As a strategic partner of CAST Software, named a Leader in the first-ever Gartner Magic Quadrant for Technical Debt Management Tools, McLean Forrester uses CAST Imaging to find every data connection across the estate: where each store lives, which systems depend on it, and how data moves between them. That map is the important first step, and only the first step.
The map does two jobs. It exposes the redundant stores, orphaned copies, and pipelines no one owns, which can then be consolidated or retired without breaking what is downstream. And it shows where the same fact lives in several systems at once, which is where the real work starts. Finding the copies does not tell you which one is right. For each contested element, the customer record in the CRM weighed against the one in billing and the one in the service system, a subject matter expert has to judge which source is the most accurate and complete. That decision cannot be fully automated, and made deliberately it is what turns a pile of copies into a single trusted source.
Once the authoritative source for a data element is chosen, it is established as the master for that data. In a centralized model, its ratified values populate a central Master Data hub that every system then draws from. In a hybrid model, the winning system is named the master for its data type in place, and the hub references it instead of keeping its own copy. Either way the result is one agreed answer per fact, with a clear owner, promoted into the governed Source of Truth.
Semi- and unstructured data: establish the truth in UnicornIQ
The larger, messier half of the estate has no schema to map. The decks, documents, contracts, transcripts, recorded presentations, and drives cannot be cataloged the way a database can, which is why rules-based tools leave them ungoverned. UnicornIQ handles this half through a console built for the people who know the answers. A subject matter expert works in that console to rationalize what the content asserts, confirming which statements are true and marking them as the source of truth for that kind of information. The same pattern as the structured side holds: the system surfaces the candidates, and the expert makes the call.
Establishing that truth once changes everything after it. Every later ingestion is compared against it, so a new deck, contract, or recording is checked against a known-good reference instead of being taken at face value. UnicornIQ becomes the source of truth for that kind of data. It flags where new content diverges, scores what teams produce before it reaches a customer, and can generate fresh material straight from validated facts rather than from the last deck someone copied. The layer that used to manufacture drift becomes the layer that catches it.
That content source of truth connects to the Master Data hub as a bi-directional rationalization loop. Ratified facts from the hub flow in to validate content, and the truths established in the console flow back to enrich the hub, kept synchronized in real time so both stay current. The two engines then meet. CAST and the structured adjudication produce a clean master for the data that has a schema, and UnicornIQ produces one for the content that does not. Both answer to one governed Source of Truth, so the whole estate finally tells one story.
The Path Forward: From Data Estate to Source of Truth
Both engines run inside a single engagement, sequenced so that value compounds at each step. The first move is always visibility, because you cannot govern, rationalize, or retire what you cannot see. From there the engagement rests on four areas:
- DISCOVER – Full inventory of the data estate. CAST maps the structured connections and UnicornIQ reads the unstructured content, so dark data, ROT, and business-critical assets are identified on evidence, not assumption.
- GOVERN – Ownership, lineage, and retention policies embedded into existing platforms and workflows, with master data anointed, ratified, and made authoritative.
- RATIONALIZE – A structured reduction program that consolidates or retires redundant stores and pipelines without disrupting operations, and reconciles unstructured content against the Source of Truth.
- SUSTAIN – Continuous governance so the Source of Truth stays clean as the business grows, powered by UnicornIQ from day one so rationalization becomes a practice, not a project.
What This Means for Your Enterprise
The debt is already on your books. It is accruing interest in dark data you are backing up, in reconciliation hours no one bills, and in AI initiatives that will stall the moment they reach for data they cannot trust. The only real decision is whether you manage it deliberately or keep paying the interest in the dark.
For most organizations, the AI ambition is clear. What is harder to see is how tightly that ambition is bound to the maturity of the data beneath it. The relationship is symbiotic: an AI or automation program is only as capable as the data environment it draws from, and that environment is more than data alone. It is the applications that generate the data, the architecture that connects them, the knowledge that explains them, and the governance that decides who owns them. Data debt is one facet of that composite, and it does not resolve in isolation from the rest. Reading the whole picture, how each form of technical debt feeds the others and where an AI initiative will hold or quietly stall, is the work McLean Forrester was built for. We define what an AI project requires, map it to the real architecture and sources, and establish the clean, governed sources of truth those ambitions have to stand on.
The organizations that will lead the next decade are not waiting for a perfect moment to govern their data. They are establishing a Source of Truth now, and grounding every decision and model in data they can actually trust. The cost of waiting compounds every quarter.
If establishing a Source of Truth before your next AI initiative is on your agenda, let’s talk.
Sources: Veritas Global Databerg Report; Gartner; Cisco AI Readiness Index; Precisely Data Integrity Trends; IDC Global DataSphere; IBM Cost of a Data Breach.
