At 9 p.m. the night before a board meeting, you're staring at three MRR figures. Salesforce shows one number, Stripe suggests another, and finance has a third number in the ERP-backed spreadsheet. Your VP of Sales insists the CRM is right. Finance trusts the ledger. The CEO wants a forecast by morning.
Nobody has changed the business overnight. The business has changed in three different data systems, according to three different definitions.
That's what data quality issues look like in practice. They aren't abstract governance problems reserved for large enterprises. They're the reason forecasts change every Monday, board slides require manual reconciliation, and smart operators lose confidence in the numbers they're paid to manage. For companies with 20 to 200 employees and no data team, the central decision isn't whether data quality matters. It's whether to hire, patch, or buy a reliable operating layer before the next reporting cycle breaks.
Table of Contents
- The Night the Numbers Stopped Matching
- What Data Quality Issues Actually Look Like at a 50-Person Company
- The Cost of Bad Data in Dollars and Decisions
- Why the Problem Keeps Coming Back Even After You “Fix” It
- The Fully-Loaded Cost of Your First Data Hire
- A Decision Framework for Founders and COOs
- What a Semantic Layer and Trusted Metrics Actually Mean
- What to Do This Week If Your Numbers Don't Match
The Night the Numbers Stopped Matching
The founder opened the finance workbook first. It showed recurring revenue that was lower than the sales dashboard, because finance excluded accounts that had not cleared payment. The sales dashboard counted contracted subscriptions, including a customer whose expansion had been signed but not invoiced. The product dashboard used active accounts, which included a few free users that sales considered prospects.
Three figures. Three defensible explanations. No shared answer.
By 9:30 p.m., the team was no longer discussing growth. They were debating definitions. One person searched old Slack messages for the original MRR formula. Another opened a spreadsheet with a tab called “Board Metrics Final,” then found another called “Board Metrics Final 2.” The CEO had to decide whether to delay the deck, choose one number, or expose the disagreement to the board.
The operational test: If your team needs a meeting to decide which dashboard is right, you don't have a reporting problem. You have an ownership and definition problem.
The damage extends beyond an uncomfortable presentation. A revenue forecast built from inconsistent stages can change hiring plans. A CAC figure built from incomplete spend data can push the marketing team toward the wrong channel. A churn analysis that joins customers differently across systems can send product leaders after the wrong cohort.
The tools usually work as designed. Salesforce records what salespeople enter. Stripe records payments. The ERP applies its accounting rules. The product database records events. The failure occurs when leaders ask those systems to answer one business question without agreeing on what the question means.
That's why data quality issues belong in the operating conversation, not just the technical backlog. When numbers disagree, someone spends time reconciling them, someone delays a decision, and someone eventually makes a choice based on a number they don't fully trust.
What Data Quality Issues Actually Look Like at a 50-Person Company
At a 50-person company, data quality issues are usually visible as friction. They appear when a report takes too long, a customer appears twice, or two competent teams describe the same business event differently.
The five problems founders see first
Duplicate records: A prospect enters through a webinar form, gets imported into HubSpot, and later appears again after a sales rep creates a record manually. Marketing counts two leads, while sales sees one conversation.
Inconsistent definitions: Product calls a customer “active” if they logged in, while finance calls the account active if it paid during the period. Both teams can produce a plausible active-customer report with different results.
Late or missing event data: A product tracking change stops sending usage events for part of the customer base, so the retention dashboard shows a healthier or weaker pattern than the product experienced.
Broken source connections: A tool migration changes a field name or integration behavior, and the pipeline continues with missing values. The dashboard still loads, but several figures lose their source data.
Conflicting revenue numbers: Stripe reflects collections, the ERP reflects accounting treatment, and the finance spreadsheet applies manual adjustments. The result is three revenue views that answer different questions without saying so.
These categories overlap. A migration can create duplicates and break a source connection. An inconsistent definition can make two accurate datasets appear contradictory. The important move is to name the failure precisely, because “the data is messy” gives nobody a useful decision.
If you want a practical way to compare the checks different systems can support, review these data accuracy metrics tools. The point isn't to collect every possible quality score. It's to identify which records and definitions affect the decisions your team makes repeatedly.
Use business language, not data-team language
A founder rarely needs a lecture on completeness or lineage during a board-prep crisis. They need to know whether pipeline value is duplicated, whether net revenue means the same thing everywhere, and whether product events arrived before the report was generated.
Write the problem in the language of the decision it can distort. “The CRM has duplicate contacts” is useful. “Our customer dataset has a uniqueness issue” may be technically accurate, but it won't tell a sales leader what to do next.
The Cost of Bad Data in Dollars and Decisions
Gartner estimates that poor data quality costs organizations an average of $12.9 million per year, a benchmark discussed in Gartner's data quality research. A 20-person company should not copy that figure into its budget. The useful lesson is the mechanism: unreliable data creates financial waste through incorrect decisions, duplicated work, and delayed action. For a 20 to 200 person company, that waste also exposes a hiring decision. Bringing in a first data employee may take longer and cost more than using a done-for-you semantic-layer service to establish trusted definitions and metrics.
Independent coverage of industry research has cited poor data quality consuming 15% to 25% of revenue in some organizations, as documented in this research review on data quality costs. That range does not describe every company, but it explains why the issue reaches finance and executive meetings. A small CRM error can affect account assignment. A broken spend connection can distort CAC. A missing event stream can change how leadership interprets retention.
Where the loss shows up
The first cost is often labor. In one survey, data professionals reported spending roughly 40% of their time evaluating or checking data quality. Organizations also experienced about 61 data-related incidents per month, with each incident taking an average of 13 hours to identify and resolve, according to TDWI's survey coverage. A smaller company may not have data professionals to absorb that burden. The work lands on the COO, finance lead, RevOps manager, or analyst who should be producing insight instead of validating exports. Hiring adds recruiting time, onboarding, management overhead, and the risk of building internal processes before the business has stable metric definitions.
The second cost is decision distortion. Research reported that poor data quality affected 26% of company revenue, while more recent survey data found the average percentage of impacted revenue increased to 31% from 26% in 2022, as reported in the State of Enterprise Data Quality report. These figures do not mean your company is losing exactly that share. They show why data quality belongs in the P&L conversation.
A wrong CAC number can justify ad spend in an inefficient channel. A flawed churn cohort can direct product investment toward customers who are not leaving. A forecast built from changing definitions can make hiring look urgent one week and reckless the next.
Boardroom rule: The cost of bad data is not spreadsheet cleanup. It is the decision made before anyone discovers the spreadsheet was wrong.
Why the Problem Keeps Coming Back Even After You “Fix” It
Many teams respond to recurring data quality issues by buying another tool or asking an analyst to clean another export. That treats the symptom as a software defect. In smaller companies, the deeper failure is usually semantic inconsistency, meaning different teams attach different meanings to the same business terms.
Sales defines a qualified lead by rep judgment. Marketing defines it through campaign behavior. Finance recognizes revenue under accounting rules. Product measures activation through usage. Each definition can be reasonable inside its own function. The problem begins when leadership expects one dashboard to combine them without an agreed meaning.
The skills gap makes this worse. SYNQ's 2025 survey identified insufficient knowledge of how to test well as the top data quality challenge, and nearly 40% of companies were increasing investment in data quality and observability tools, according to its 2025 State of Data Quality Survey. More tools don't solve a team that hasn't agreed what should be tested or who owns the result.
Ownership fails before software does
A new dashboard can display a metric. It can't decide whether “active customer” means paid, logged in, onboarded, or contractually live. An integration can move fields. It can't decide which source wins when Stripe and the ERP disagree.
The data observability guide is useful for understanding visibility into pipeline behavior, but observability alone doesn't create business agreement. Someone still needs authority to define the metric, approve the formula, and resolve disputes when teams use the term differently.
The shortage of staff is also a recognized barrier. A Drexel and Precisely outlook reported that 42% of organizations identified a shortage of skills and staff as the biggest barrier globally, according to the cited SYNQ research summary above. For a small company, that means the person asked to “fix the dashboard” may also own CRM administration, forecasting, finance requests, and operational reporting.
The contrarian answer: If the definition is unstable, another dashboard only makes the disagreement easier to distribute.
A durable fix requires a process that keeps definitions, ownership, and testing aligned as the business changes. That's why the first question shouldn't be “Which BI tool should we buy?” It should be “Who is accountable for the meaning and reliability of the numbers executives use?”
The Fully-Loaded Cost of Your First Data Hire
The default advice for a growing company is simple: hire an analyst. That advice ignores the period before the hire becomes useful and the surrounding cost required to make the role work.
A first data hire needs recruiting support, compensation, benefits, and often equity. They need access to every source system, time with finance and RevOps, and enough context to understand why two teams use the same metric differently. They'll spend part of the ramp learning the business, part discovering undocumented logic, and part rebuilding reports that already exist in fragile form.
The cost isn't just the offer letter. It's the delay before trusted reporting exists, plus the mistakes made while a new employee learns which data is authoritative. If your board deck is due before that foundation is ready, the company still relies on spreadsheets and manual reconciliation.
The decision in one view
| Dimension | First Data Hire | Done-For-You Agentic BI |
|---|---|---|
| Upfront commitment | Recruiting process, compensation package, benefits, and internal management time | $5K per month flat |
| Business context | Built gradually through interviews, documentation, and recurring exposure | External operator brings a defined delivery process |
| Definition alignment | Depends on internal authority and cross-functional cooperation | Establishes agreed metric meanings as part of the reporting foundation |
| Time to useful output | Delayed by recruiting, onboarding, access, and discovery | Trusted metrics and AI-answerable data delivered in 30 days |
| Ongoing ownership | Requires management, prioritization, and retention | Service owns the agreed reporting work |
| Risk profile | One person may become a bottleneck or leave with undocumented knowledge | Delivery risk is distributed across a service model |
The comparison isn't “employee versus software.” It's internal capacity versus an outcome delivered by an operating partner. A service priced at $5K per month can be evaluated against the full cost of waiting, recruiting, managing, and correcting the first hire, not against salary alone.
A company that needs one stable report may not need either option. But if the pain is conflicting dashboards across CRM, billing, product, and finance, a generalist analyst can become the human patch between systems. The role may eventually be necessary. It's rarely the fastest way to establish trustworthy numbers.
For a clearer distinction between analytical and engineering responsibilities, review this guide to data engineer versus data analyst roles. The title matters less than whether someone can own definitions, source relationships, and executive-grade answers.
A Decision Framework for Founders and COOs
Don't frame the choice as “hire or do nothing.” There are three sensible paths, and each fits a different operating condition.
Path one, fractional analyst
Choose a fractional analyst when you have one clear recurring report, the source tools are stable, and someone inside the business can make decisions about definitions. This works when the problem is consistent ownership of a limited reporting process, not a disagreement across the company's operating data.
The warning sign is scope creep. If the analyst immediately inherits CRM cleanup, product events, revenue reconciliation, and board reporting, you don't have a fractional-reporting problem. You have a data foundation problem.
Path two, full-time internal hire
Hire internally when data work is becoming a permanent part of planning, multiple teams need embedded support, and leadership is prepared to change processes around the role. A full-time hire makes more sense when the company has multiple revenue lines, complex planning cycles, and enough operational maturity to keep definitions current.
The hire should own more than dashboards. They need a mandate to resolve conflicting business logic, establish accountability, and influence how teams enter and interpret data. Without that authority, the company pays for reporting production while preserving the source of the inconsistency.
Path three, done-for-you agentic BI
Outsource when the immediate symptoms are conflicting dashboards, slow board reporting, and plain-English questions that cross multiple tools. That is the dominant pattern for many companies in the 20-to-200 employee range. They don't need another visualization layer first. They need agreed metrics, connected sources, and answers leadership can audit.
Use this diagnostic:
- One report, stable tools, clear owner: Start fractional.
- Several planning cycles, expanding data needs, internal mandate: Consider a full-time hire.
- Multiple systems, conflicting definitions, immediate executive reporting pain: Buy a done-for-you service.
The right answer depends less on company size than on the shape of the failure. If your team can't agree which number is correct, adding headcount without decision rights may create another interpreter of the same mess.
What a Semantic Layer and Trusted Metrics Actually Mean
A semantic layer is the shared definition layer between raw business systems and the reports people use. It records what terms such as MRR, active customer, qualified lead, and net new ARR mean, then applies those definitions consistently across dashboards, board materials, and conversational analytics.
Without that layer, every report can write its own formula. One dashboard may count contracted subscriptions. Another may count collected payments. A third may filter out accounts with recent cancellations. The numbers can all be calculated correctly and still fail to answer the same question.
Trusted metrics are a small operating contract
Trusted metrics are the named KPIs the company agrees to use for recurring decisions. They have an approved meaning, a consistent calculation, and enough documentation that finance, sales, product, and leadership can recognize the same result.
That doesn't mean every number in the business must be perfect before anyone acts. It means the metrics used for hiring, forecasts, board reporting, and resource allocation must have a visible owner and a stable interpretation. The semantic layer makes that agreement reusable instead of forcing every analyst to recreate it.
This is also why agentic BI depends on more than a chat interface. If a founder asks, “What was our net new ARR last week?” the system needs to know which accounts qualify, which revenue source is authoritative, and how the time period is defined. Plain-English access becomes reliable only after the underlying business language is governed.
Leaders considering an internal capability can also explore resources on how to hire top data talent. That may be the right long-term move when the company needs permanent ownership. It doesn't remove the immediate need to define the metrics that person will own.
For the operating principles behind shared definitions and accountability, see this guide to metrics governance. The purchase decision should be about the reliability of the operating contract, not the novelty of the interface.
What to Do This Week If Your Numbers Don't Match
Make three decisions before you open another dashboard.
First, name the three metrics that must agree. For most SaaS companies, that may be MRR, active customers, and qualified pipeline. Write down the business question each metric answers and identify the system currently treated as authoritative. Don't attempt to standardize every field at once. Start with the numbers that influence board reporting, forecasts, and hiring.
Second, calculate the fully-loaded cost of waiting for a first hire. Include recruiting effort, compensation, benefits, equity, management time, ramp-up, and the reporting work that remains unresolved during the transition. Compare that with a $5K-per-month done-for-you service that delivers trusted metrics and AI-answerable data in 30 days. The relevant comparison is not salary versus subscription. It's delayed trust versus an operating result.
Third, stop calling the bottleneck a tooling problem. If sales, finance, and product use different definitions, a new BI tool will distribute disagreement faster. Decide who owns the meaning of each executive metric, how disputes get resolved, and whether you need internal capacity or an external service to establish that foundation.
Your goal this week isn't a perfect data estate. It's a reliable answer to the questions your leadership team already asks. If the same question produces a different number depending on who opens the report, the company needs governed metrics before it needs more dashboards.
HelpWithMetrics connects your business data, establishes a semantic layer, and delivers trusted metrics that leaders can question in plain English. Book a call with HelpWithMetrics to get a free first dashboard and see what reliable reporting looks like in your business before committing.