The Spreadsheet Is the System
Nobody decided that. It just happened, one urgent Tuesday at a time.
We have now run intake conversations with organizations at very different scales. A forty-person specialty firm. A regional operator with a few hundred employees. A division inside something much larger, with a real IT function and a real budget.
They do not look alike. Their problems do.
Somewhere in each of them, a business-critical number lives in a spreadsheet that one person maintains. Not as a working draft. As the system of record. Pricing tiers. Vendor terms. Certification expiration dates. Which customers are on which contract version. Headcount by cost center before it becomes an official report.
Nobody chose that architecture. It accumulated.
How It Happens
Growth does not announce itself. It arrives as a problem you have to solve by Friday.
A new service line needs tracking, and standing up a proper system takes a quarter you do not have. So someone opens a spreadsheet. It works. It keeps working. Six months later it has tabs, formulas, and a color-coding scheme only its author understands.
Or the other version, which is worse: you already own a system that is close enough. So you adapt it. The CRM gets a custom field called “Notes 2” that actually holds renewal risk. The project tool gets a status value that means something specific to one team and nothing to anyone else. The ERP gets a workaround that made sense to the person who built it and has since left.
Both paths feel responsible in the moment. Both are debt. The bolt-on creates a system nobody owns. The adaptation creates a system that lies, because the field name no longer matches the data inside it.
This is not a small-company failure mode. AutoRek’s annual payments survey, covering 500 senior finance managers in the US and UK, found spreadsheets remain integral to financial operations at 90% of firms. KPMG surveyed 550 board members and executives, two-thirds of them at companies above a billion dollars in revenue, and found spreadsheets were the most-used ESG data management system by a wide margin, at 47%. Big companies have more systems. They do not have fewer spreadsheets.
What It Actually Costs
The cost is rarely a dramatic failure. It is a slow tax, collected in three places.
Time. The tax shows up first as hours. Reconciliation, re-keying, chasing the current version, verifying that what you found is the one that counts. None of it appears on a budget line, which is exactly why it survives. You hire five people and four show up to do the work, because the fifth spends the week finding out what is true.
Money. Gartner’s 2020 estimate, still the figure the firm publishes, is that poor data quality costs organizations at least $12.9 million a year on average. Gartner also names inconsistency across sources, the result of data stored in silos with overlaps and gaps, as the most challenging data quality problem. Silos are what spreadsheets are. Every copy is a fork.
Optionality. This is the one that hurts later. Thomson Reuters put numbers on the fragmentation pattern: 54% of organizations lack the context needed to complete cross-functional governance and compliance tasks, and 49% lack a continuous, cross-functional view of their data. Their description of how it starts is exact: sales builds one vendor tracker, procurement maintains another, legal keeps its own contract database, finance manages payment terms separately, none of them speak to each other, and reconciling them becomes a full-time job.
Then someone asks a reasonable question. Which customers are affected by this regulatory change? What is our actual margin on this service line? Can we take on this contract without breaking a certification requirement?
The answer exists. It is just distributed across four files, two systems, and one person’s memory. So the question gets answered slowly, or approximately, or not at all. Worse, sometimes it gets answered confidently and wrong, and the error ships into operations before anyone catches it.
The regulatory version of this is the sharpest. When your obligations live in a tracker and your customer profile lives somewhere else, you cannot answer “does this apply to us” without a human doing the join. That human is not always available, and is never continuous.
This is also where the AI conversation quietly stalls. A Gartner survey of data management leaders found 63% either do not have the right data management practices to support AI or are not sure whether they do. Gartner’s accompanying prediction is that through 2026, organizations will abandon 60% of AI projects that are not supported by AI-ready data. Those pilots do not fail on the model. They fail on the spreadsheet underneath.
The Trap in the Obvious Fix
The instinct is to launch a data cleanup. Inventory everything. Standardize everything. Then modernize.
We have written about why that stalls. It has no definition of done. Every table looks important, every inconsistency looks urgent, and the program becomes a standing IT effort with no owner in the P&L.
The better move is narrower. Do not fix your data. Make one part of it legible, on a path where the money is visible.
Five Things You Can Do Without Hiring Anyone
This is work you can start this month, with people you already have. It is also the work that tends to make everything downstream easier, whoever ends up doing it with you.
1. Inventory the spreadsheets that are actually systems.
Not every spreadsheet. The ones where the answer to “where does this live” is a file. Ask each function leader one question: what do you maintain by hand that other people depend on? Write down the file, the owner, the update cadence, and what breaks if it is wrong. This usually takes a week and usually surprises the executive team.
2. Name a single source of truth for every field that matters.
For each important field, one system wins. Customer name. Contract effective date. Payment terms. Certification status. Write it down. Where two systems disagree today, you have found a decision, not a technical problem. Make the decision. This is the highest-leverage hour a leadership team can spend on data, and it requires no software.
3. Kill the fields that lie.
Every custom field repurposed for something other than its name is a landmine for any future automation. Find them. Either rename them to what they hold, or migrate the content to a field that means what it says. A system that lies to humans will lie to machines faster and at greater volume. Read the last sentence again.
4. Write down freshness expectations.
Data is not clean or dirty. It is current or stale. For each source of truth, state how fresh it has to be for decisions to be safe: real time, daily, monthly, annually. Then check whether reality matches. A field updated whenever someone remembers is not a data source. It is a rumor with a timestamp.
5. Pick one workflow and follow it end to end.
Choose a path where delay or error costs something you can name. Renewals. Onboarding. Invoice reconciliation. Compliance attestation. Walk it step by step and mark every place a human retypes, re-checks, or chases something by email. Those points are where the spreadsheets are load-bearing. That map is worth more than any readiness score, because it tells you which data has to be trustworthy first.
None of this requires a platform. All of it makes the next platform cheaper, faster, and far less likely to fail.
You Do Not Have to Have This Solved
We have not yet met the company that had this solved. It comes up in the first conversation, every time. That is not the exception, it is the condition.
The research agrees. Harvard Business Review Analytic Services surveyed leaders involved in their organizations’ AI data decisions and found just 7% say their data is completely ready. The single most cited obstacle was siloed data and difficulty integrating sources, at 56%. Only 23% have an established data strategy for AI. But 53% are actively building one, which is the more useful number. Most companies are not negligent about this. They are mid-effort, and they know it.
Elution Labs builds production-grade agentic systems for mid-market to enterprise organizations. We combine a managed data foundation with structured delivery methodology, so what we deploy is ready for real production use from day one. The data foundation piece is not a preface to the real work. It is half of what we do.
So take these five steps as a head start, not a checklist you have to pass. Do one of them. Do the inventory and stop there. Whatever you learn in the doing is what makes the first conversation useful, because you will already know where the soft ground is.
If the spreadsheets are still winning, that is exactly the moment to talk. Begin your intake assessment at https://www.elutionlabs.ai/intake
Closing Thought
The spreadsheet was never the mistake. It was the right tool for an urgent Tuesday, and it did its job.
The mistake is letting an urgent Tuesday become the architecture. Data locked in a file cannot be assessed, cannot be governed, and cannot be acted on by anything faster than a person opening it.
Get it out of the file. Not all of it. The part that decisions depend on.
Sources
AutoRek, annual payments survey. 500 online interviews with senior finance managers and above in the US and UK, fieldwork October 2024, published February 2025. Summary · Report
KPMG US, survey of 550 board members, executives and managers on ESG reporting. Coverage · KPMG
Gartner, data quality research, 2020, still published on Gartner’s data quality topic page. Gartner
Gartner, survey of data management leaders and related AI-readiness predictions. Press release, February 2025. Press release
Harvard Business Review Analytic Services with Cloudera, Taming the Complexity of AI Data Readiness. More than 230 members of the HBR audience involved in their organizations’ AI data decisions, fieldwork October 2025, published March 2026. Findings · Report
Forrester Consulting, AI Meets Governance And Compliance, commissioned by Thomson Reuters, October 2025, as reported by Thomson Reuters. Report (PDF) · Thomson Reuters