What Large Companies Do With Their Data (And Why 50-100 Employee Companies Need the Same Thing)

In November 2024 Databricks published a case study on how Corning, Volvo and Cox Automotive automated their data flows. The numbers belong to another world compared with an 80-person construction company, but the problem is the same: data in silos and decisions taken on old reports. This analysis covers what the large companies did, why small companies fall behind and what the same solution looks like on our platform, at your scale.

Corning, Volvo and Cox Automotive solved the same problem as your company: data in silos

The three companies in the Databricks study had data scattered across systems that did not talk to each other, and automated the flows between them. Corning manages approximately 2,500 automated jobs, around 5 petabytes of data, for 900 active users worldwide (Databricks, 2024). Before, the divisions worked in silos: each with its own data, cleaned and processed separately.

Volvo sells nearly 200,000 new trucks a year and tracks millions of spare parts. With real-time processing it monitors inventory continuously and achieved up to 40% greater efficiency in handling large data volumes (Databricks, 2024). Cox Automotive runs about 300 jobs, of which approximately 120 are scheduled to run regularly, without the earlier bottlenecks.

None of the three is talking about alien technology. They talk about data gathered from different sources, updated automatically and shown as it is today, not as it was two weeks ago. The key phrase in the study is the shift from “reactive problem-solving” to “proactive innovation”: you learn that a project is losing money while you can still do something.

Reports steal the time that should be spent on the business

At a company with 50–100 employees, the silos are three Excel files, and their cost is the time lost and the decision taken too late. Accounting has its data in one file, sales in another, the site in a third. On Friday afternoon someone tries to find out whether Project X was actually profitable, and the answer comes on Monday, if it comes.

Nobody counts those hours, because they appear on no invoice. The real cost is not the hour spent copying numbers but the decision that was postponed. A client asks on Wednesday whether you can take on a new project; to answer, you need the team's capacity from three different files.

There is another, quieter cost: the consolidated report reaches the director three to five days after something happened on site. A budget overrun discovered on Monday is already a week of extra orders. The report says what happened, not what is happening.

SMBs fall behind because they are too big for Excel and too small for consultants

The gap between small and large companies is about size, not technology: the same tools exist, but nobody brings them to their scale. Eurostat measures every year how many enterprises in the European Union (EU) use artificial intelligence (AI). In 2025, 17.0% of small enterprises used it, against 30.4% of medium and 55.0% of large ones (Eurostat, Dec 2025).

The smaller the company, the lower the chance that it works on automated data. The explanation is not a lack of solutions: Power BI, Tableau and the open-source tools of the Apache family work at any scale. Volvo uses the same principles as a 50-person distributor; the only difference is who brings them to the door.

Large consultancies do not look at small and medium-sized businesses (SMBs): you are too small for them. Vendors of enterprise resource planning (ERP) systems propose projects sized for corporate budgets: you are too big for Excel but too small for them. The middle path exists, but nobody is selling it.

Automation, concretely, at an 80-person company

Automation means data is entered once, at the source, while the report builds itself and warns you when something goes over budget. The word is frightening because it sounds like a one-year project. In practice, the data already exists: in the accounting software, in the timesheets, in the purchase orders.

What is missing is the link between them and one place where they are read together. Before, the site manager sends an Excel file on Friday afternoon, the project manager reconciles it with two files from purchasing, and the director gets the report on Monday. After, the sources are connected once, the dashboard refreshes itself, and the director sees the situation on their phone.

The difference that matters is not the dashboard but the alert. A pretty report nobody opens saves no project; a notification saying a site has spent more of its budget than of its works does. The AI Agent in the opening scene does exactly that: it reads the three sources and sends the alert to the project manager, with the orders that explain the gap.

What to ask of an automation partner

You do not need Databricks; you need someone who understands what large companies do and brings the same principles to your scale, in weeks. Databricks works with Volvo and Corning and sells at their size; that is normal. An 80-person company does not have 2,500 workflows but five or six reports made by hand.

The right partner starts from those reports, not from a platform. The first two criteria are about understanding: they know what the large companies did and why it worked, and they can reduce the principles to five to ten ongoing projects. The next two are about tools and pace: affordable tools you can pay for and understand, and a go-live in weeks, not months.

The fifth criterion is the one few people ask for: alerts, not just reports. The cost is judged against the alternative, not in the absolute. The alternative is a person gathering numbers every Friday and a project on which you learn you lost money after it is finished.

In practice: the Friday report becomes a dashboard with alerts

A fictitious construction company with 80 employees and seven projects connects three sources and catches its first budget overrun in the second week. The scenario is the one in the opening scene: accounting software with a purchasing module, electronic timesheets and a tracking Excel file for each site. The Friday report took three hours and reached the director on Monday.

Fictitious data.

Connecting took four weeks, the first two spent cleaning project names written differently in the three sources. From the third, the dashboard answers questions asked in plain language: “which projects went over their materials budget this month?”. In the second week of operation, the AI Agent sent its first alert: Northgate warehouse had spent 92% of its materials budget at 68% of the works.

The alert came with the three orders that explained the gap: PPR pipe, overtime from week 36 and a rejected return on a valve. Robert T., the project manager, saw it at 08:14 on his phone, opened the orders and stopped the fourth before it left. In the Friday report, the overrun would have appeared ten days later, with the order already delivered.

Where it does not apply

Connecting sources and alerts make sense when the data already exists and someone takes decisions on it; in three situations it is not the moment.

• When the data does not yet exist in any system. If hours and orders are known from memory or on paper, there is nothing to connect; first you digitise, then you automate.
• When you have one or two ongoing projects. A director who sees everything from the site does not need alerts; a good Excel file and the Monday meeting are enough up to five or six projects.
• When nobody has time to act on the alert. The alert catches the overrun, but a person takes the decision; if the project manager has no mandate to stop an order, the dashboard becomes one more unread report.

What next

This month, count the hours your team loses on manual reports: who gathers, from how many files, how long it takes and when the result reaches the person who decides. That number is the answer to the right question, “how much does it cost us NOT to automate?”, and it is the first thing a serious partner asks for. If you want to see which processes usually get automated first, the eight automation ideas for companies that have outgrown Excel start exactly from manual reporting.

Frequently asked questions

Sources

1. How automated workflows are revolutionizing the manufacturing industry · 2024
2. Use of artificial intelligence in enterprises (Eurostat, Statistics Explained) · 2025