The 1-10-100 Rule: What Bad Data Really Costs A Caribbean Board

Every board meeting runs on numbers someone else prepared. The revenue slide, the aged-receivables report, the inventory position, the loss ratio: directors approve decisions on all of them without ever seeing the raw records underneath. That is normal, and mostly fine. It stops being fine when the records underneath are wrong, because a wrong number does not look any different on a slide than a right one. It carries the same font, the same confidence, the same decimal places. The board approves it just the same.

Poor data quality is expensive precisely because it is invisible at the level where decisions get made. Nobody signs off on "reorder 3,000 units of the wrong item." They sign off on a demand forecast built from a sales history that quietly double-counted a month. The cost is real, it is recurring, and most Caribbean firms have never put a figure on it. This piece is about how to put that figure on it, in maths a director can check, and what to do once you have.

Where The Money Actually Bleeds

The most useful frame for a board is the oldest one. In 1992, George Labovitz and Yu Sang Chang set out what became known as the 1-10-100 rule. It says the cost of a data error grows by an order of magnitude at each stage it survives. Preventing the error at the point of entry costs about one unit. Correcting it once it is already sitting in your systems, after someone notices, costs about ten. Carrying the cost of failure, when a real decision is made on the bad record and nobody catches it until the damage is done, costs about a hundred.

The exact multipliers are a heuristic, not a law of physics. The ratio is the point. The same duplicate customer costs pennies to stop at the counter, a few dollars to clean up in a data project next quarter, and a genuine amount of money if it drives a marketing spend, a credit limit or a stock decision before anyone finds it. Read against a board's actual behaviour, the rule explains an uncomfortable pattern: firms spend almost nothing on prevention, a little on periodic clean-ups, and absorb the failure cost silently as "just how it goes."

Put A Real Number On It: A Worked Example

Averages do not move boards. Your own numbers do. So here is the calculation on a company that looks like a lot of our clients: a mid-sized Jamaican distributor with 40,000 active customer records driving its credit, delivery and marketing decisions. The three inputs you need are things a firm can estimate about itself, not figures borrowed from a report.

The point of the exercise is not the J$8 million. It is that the number exists, it is defensible, and it is almost always larger than the cost of the governance that would prevent it. When a director asks "what would fixing this actually save us," this is the arithmetic that answers. Change the three inputs to your own and the method holds. A bank with two million accounts and a lower per-error cost lands somewhere very different from a distributor with 40,000 and an expensive delivery failure, and both are worth knowing.

Photo via Unsplash.

Why This Belongs On The Board Agenda, Not The IT Backlog

Data quality gets filed under IT because it involves systems, and that is exactly why it never gets fixed. IT can build a deduplication routine. It cannot decide that the finance director owns the customer master file, that a credit decision may not be made on a record older than 90 days, or that a duplicate rate above two percent triggers a review. Those are policy calls about accountability and acceptable risk, and they sit with leadership by definition.

The risk framing is what makes this urgent in our part of the world. A Jamaican business is already pricing under real volatility: hurricane exposure, a still-recovering economy, tighter household budgets. Every one of those decisions leans harder on the data than it would in a calm year. Reprice a policy, restock ahead of a storm, extend or pull credit in a downturn, and a wrong record no longer costs you a wasted mailing. It costs you a mispriced risk you are now carrying. The error rate did not change. The stakes riding on each error did.

The Four Numbers A Board Should See Every Month

You cannot govern what you do not measure, so start with four dimensions on your most important tables and put each on a monthly dashboard as a plain percentage. Completeness: the share of records with no missing required field. Accuracy: the share that match a trusted reference. Uniqueness: the share with no duplicate. Timeliness: how current the record is against a defined limit. None of this needs a large platform to begin. It needs a threshold for each number and the discipline to treat a breach the way finance treats a covenant breach.

The cheapest move on that list is the first, and most firms have not made it. Assign a named senior person to each critical data set and the vague, everybody-and-nobody complaint about "our data is a mess" becomes one person's measured responsibility. That single change does more for data quality in a Caribbean company than most software purchases, because it converts a shared shrug into an accountable line on someone's objectives. It is also the change a board can mandate in a single meeting.

Governance First, Then The Automation

None of this is an argument against AI. Machine learning genuinely lowers the cost of correction: it finds duplicates, flags outliers and standardises messy addresses far faster than a person working through a spreadsheet. What it cannot do is decide what a record is supposed to mean, who is accountable for it, or what an error is allowed to cost. Those stay human. Point a model at an ungoverned data set and it will give you faster, more confident wrong answers, which is worse than slow ones because they arrive with the authority of a machine behind them. Fix the ownership and the definitions first. Then let the tools run fast on a foundation you can trust.

There is a limit worth admitting here. Not every error is worth preventing. Chasing a data set to 100 percent accuracy can cost more than the failures it stops, and a good governance programme decides where "good enough" actually is rather than pretending perfection is the target. The skill a board needs is not zealotry about clean data. It is the judgement to know which data sets carry real decisions and to spend the governance effort there.

Frequently Asked Questions

What is the 1-10-100 rule in data quality?

The 1-10-100 rule, set out by George Labovitz and Yu Sang Chang in 1992, says it costs about 1 unit to prevent a data error at entry, about 10 to correct it once it is in your systems, and about 100 to carry the cost of failure when a decision is made on the bad record and never caught. The ratio matters more than the exact numbers: the cheapest place to fix data is where it is created, and the most expensive place is a decision that has already gone out the door.

How much does poor data quality cost a business?

Gartner estimated in 2021 that poor data quality costs the average organisation US$12.9 million a year, mostly in wasted effort, wrong decisions and lost revenue rather than one visible line item. For a specific business the figure is calculable: records that drive decisions, times the error rate, times the average cost of acting on a wrong record. That produces a defensible figure from your own files rather than an average.

Is data governance an IT problem or a board problem?

It is a board problem with an IT component. Data quality decides whether the numbers a board approves budgets, pricing and risk on are trustworthy. IT can build the pipelines, but only leadership can decide who owns each critical data set, what a data error is allowed to cost, and which single source is the version of the truth. In a risk-exposed economy, a board that cannot trust its own numbers is deciding blind.

What is a data owner and why does a company need one?

A data owner is a named senior person accountable for the accuracy and definition of a specific data set, for example the customer master file or the price list. Without a named owner, every department assumes another one is checking, and no one is. Assigning owners is the single cheapest data governance move most Caribbean firms have not yet made, and it turns data quality from a vague complaint into someone's measured responsibility.

How do you measure data quality?

Track a few measurable dimensions on your most decision-critical tables: completeness, accuracy, uniqueness and timeliness. Report each as a monthly percentage, set a threshold, and treat a breach the way you would treat a cash-flow breach. A number you do not measure cannot be governed.

Why does bad data cost more during hurricane season or a downturn?

When conditions are stable, a wrong forecast or a duplicate customer is an annoyance. When you are pricing under storm risk, restocking after a disruption, or deciding credit in a tighter economy, the same error drives a decision with a much larger downside. Bad data does not get more common in a crisis. It gets more expensive.

Can AI fix a company's data quality problem?

AI can find duplicates, flag outliers and standardise messy fields far faster than a person, which lowers the cost of correction. It cannot decide what a record means, who is accountable for it, or what an error is allowed to cost. Those are governance decisions. Point AI at an ungoverned data set and you get faster, more confident wrong answers. Govern first, then automate.

Related reading across the Caribbean AI network