The cloud is not always the answer

LusoFicta Foods, Indústria Alimentar, S.A. is a food production and distribution company based in Sabrosa, in the heart of the Alto Douro region. What began three decades ago as a family agricultural operation vineyards and olive groves inherited from previous generations gradually evolved into an integrated business covering the entire value chain: cultivation, harvesting, processing, packaging, and distribution of food products for both the domestic market and export.

The company employs around 85 permanent staff, ranging from agricultural technicians and processing plant operators to administrative, financial, and logistics teams. But that number tells only part of the story. During harvest campaigns, between September and November, LusoFicta Foods hires up to 200 temporary workers for grape harvesting and olive picking. In the peak processing and shipping season, between November and February, dozens more production line operators and drivers join the distribution fleet. At peak activity, the company may have more than 300 people working simultaneously, many of them for only a few weeks, under seasonal contracts, temporary insurance arrangements, and payroll processes that cannot be delayed by a single day.

All of this operation managing permanent and temporary staff, agricultural production control, batch traceability, refrigerated warehouse management, invoicing, shipping, communication with the Tax Authority, and payroll processing was supported by an integrated management system (ERP) installed on a physical server located in a small technical room inside the administrative building, between the accounting office and the staff kitchen.

The server was not new. In fact, it had already been purchased second-hand when the ERP was first implemented. But it worked. Every month, reports were generated, salaries were processed (including those of temporary workers with variable-duration contracts and campaign overtime), transport documents were issued on time, and Mr. Teixeira, the company’s founder and managing director, could review the week’s numbers on his computer while his breakfast was still warm.

Until the day it stopped working not the server itself, but the software support.

The ERP vendor announced the end of life of the installed version. No security updates, no fixes, no technical support. LusoFicta Foods needed to migrate to the latest version or continue entirely at its own risk.

A specialized consultancy was hired to lead the process. After the initial assessment, the diagnosis was clear: the existing hardware did not meet the minimum requirements of the new version. The recommendation was to invest in a new server before proceeding with the upgrade.

Mr. Teixeira listened, ran the numbers, and decided it was not the right moment. The company was coming off a difficult campaign. The previous summer’s drought had reduced production, margins were under pressure due to rising energy and fuel costs, and there were urgent investments needed to modernize the irrigation system. Spending thousands of euros on a new server was not part of the plan.

"The current server can handle it. It always has," insisted Mr. Teixeira.

The consultancy pushed back. They presented benchmarks, explained the risks, and detailed potential failure scenarios. But the decision had been made. LusoFicta Foods signed a liability waiver, accepting the risks of installing the new software on hardware that did not meet the minimum requirements, and the upgrade moved forward.

The installation went smoothly, at a convenient time: it was August, the quietest period of the year, before the start of the harvest campaign. The new version was configured, the data migrated, and validation was carried out with two users simultaneously: Ms. Conceição from accounting and Nuno from human resources. Both navigated through the menus, opened a few reports, and entered test transactions. Everything was fluid, everything responsive.

"See? It works perfectly." exclaimed a satisfied Mr. Teixeira, pleased that his sharp strategic vision had saved the company thousands of euros while still securing the new version of the application.

The migration was declared complete. The old software was removed. The consultancy closed the project, delivered the documentation, and moved on to the next client.

The old server remained there, between the accounting office and the staff kitchen, humming quietly. As always.

The problems did not appear in August, when the company was operating at holiday pace. They appeared in September, when the harvest campaign began and with it, chaos.

In just a few weeks, the 85 permanent employees grew to more than 250. The human resources department, which usually managed a stable workforce (even during previous years’ campaigns with the old system), found itself registering dozens of new temporary contracts per week, each with its own specifics: varying durations, shifting schedules, campaign overtime, temporary work insurance, and tax information from workers coming from different regions, and some from abroad. Each of these records had to go through the ERP.

Every payroll calculation, every social security contribution, every tax withholding depended on a system that was now bearing a load it had never been designed for.

• Payroll processing became a nightmare. Ms. Conceição, who normally processed salaries during the first five working days of each month, saw what used to take a morning turn into a three-day marathon. With more than 250 active records (many with overtime, variable meal allowances, and short-term contracts), the system would freeze mid-calculation, forcing her to start over. In a company where temporary workers rely on that payment for the following week, the delay was not just administrative it was personal. Result: For the first time in thirty years, LusoFicta Foods’ salaries were delayed. A severe blow to Mr. Teixeira’s pride, who had never once delayed paying all his staff.

• The application crashed constantly. With production line operators, agricultural technicians, drivers, and the administrative team all logged in at the same time (easily more than 80 active sessions), the server reached its memory limit and the ERP simply stopped responding. Drivers using handheld devices to register deliveries were blocked mid-route, unable to confirm transport documents. Processing unit operators could not log batches in production, compromising traceability, a legal requirement for food products.

• Production reports had to be generated at night. Generating the daily dispatch report (essential for planning the next day’s delivery routes) was no longer possible during working hours. It had to be run manually at night, one by one, by the company’s only IT employee, who started at 10 PM to launch the processes and monitored the screen until dawn. Result: During the day, there was no one available to support users or address the growing number of issues with the new application version.

• The invoicing module became unpredictable. Invoices took minutes to issue instead of seconds. On high-volume shipping days (and during peak season, when LusoFicta Foods shipped to dozens of clients per day, including retail chains with tight delivery windows), the system failed to generate documents, returning timeout errors. Clients began to complain about delays in receiving invoices, which in turn delayed payments. For a company that needed liquidity to pay hundreds of temporary workers, this was more than an inconvenience.

• Integrations with the Tax Authority failed. Automatic SAF-T submission and transport document reporting began to fail due to excessive response times. The company received notifications from the Tax Authority, causing panic over potential fines at a time when document volume was at its highest.

• Batch traceability was compromised. With the system losing sessions mid-entry, failures in the product traceability chain began to appear. In the food sector, where every batch of olive oil, every box of fruit, and every pallet of wine must be traceable from origin to destination by law, this was not just an operational issue. It posed a compliance risk that could cost certifications and access to export markets.

• Stock control in refrigerated warehouses became chaotic. Discrepancies between physical stock and system records multiplied. Products with short expiration dates were forgotten in cold storage because the system did not reflect reality. Result: food waste and financial loss.

• Management of temporary contracts went out of control. The HR module, overloaded, could not process the entry and exit of seasonal workers in time. Contracts that should have been closed remained active; work insurance that should have been activated was not processed on time. Legal and labor risks quietly accumulated.

Mr. Teixeira no longer had breakfast while looking at the week’s numbers. He had it while listening to complaints, wondering how a company that had survived droughts, frosts, and market crises was now being brought to its knees by a computer.

It was in this context of desperation that a second consultancy appeared with an enticing proposal:

"Your problem is the hardware. But you don’t need to buy anything; you need to migrate to the cloud. Forget servers, forget maintenance, forget limitations. In the cloud, everything is elastic: the system scales automatically according to demand. In August, when things are quiet, you pay less; in October, at the peak of the campaign with 300 people on the system, the infrastructure grows on its own and you don’t even notice. Plus, you get real-time monitoring, task scheduling, automatic backups, high availability..."

The presentation was flawless. Polished slides, projected cost charts showing a perfectly manageable monthly expense, promises of fast and painless onboarding. The seasonality of LusoFicta Foods was, in fact, the main selling point: "Why pay all year for a server sized for the peak when in the cloud you only pay for what you use, when you use it?"

Mr. Teixeira, exhausted from the problems of the past months and just weeks away from the start of the next campaign, said yes.

The migration to the cloud was carried out in a few weeks. The ERP was moved to the infrastructure of a major cloud service provider using a lift-and-shift approach, meaning the system was moved as-is, without optimization, refactoring, or careful sizing of the required resources.

And to ensure that "nothing failed without being detected," the consultancy activated all available monitoring options… A L L: application logs, operating system logs, network logs, database logs, access logs, CPU, memory, disk, and latency metrics. Everything was captured, stored, and topped off with the cherry on the cake: fed into an artificial intelligence pipeline that consumed thousands of tokens per hour to analyze logs in real time and trigger custom alerts.

Every CPU spike generated an alert. Every slow database query generated an alert. Every expired session generated an alert. The IT employee’s phone at LusoFicta Foods became a notification machine, hundreds per day, most irrelevant, burying the few alerts that actually mattered. By the second day, he was already ignoring all alerts they were just noise.

But the system worked. Fast, stable, available. Mr. Teixeira returned to having breakfast while looking at the numbers. Ms. Conceição processed the salaries of 280 employees in one morning. Drivers confirmed delivery documents in seconds. Batch traceability worked flawlessly again. Reports were generated on demand, with no queues, no waiting.

During the first few weeks, everything seemed perfect.

At the end of the first month in the cloud, the invoice from the service provider arrived.

The amount was ten times higher than the consultancy had estimated: ten times, 10x, 1000%.

Auto-scaling, which was supposed to be the main advantage, had become the biggest problem. Without careful configuration of limits, the system scaled aggressively with every usage spike, and in a company with hundreds of users during peak season, spikes are the norm, not the exception. Every payroll run, every heavy report, every production data import triggered additional resources that were billed by the minute.

Log storage, which seemed harmless in the proposal, turned out to be a cost sink. Gigabytes of logs were accumulated daily, retained without an expiration policy, stored on high-performance tiers instead of cold storage.

And the artificial intelligence pipeline, that premium feature promising "intelligent observability," consumed thousands of tokens per hour, 24/7, analyzing logs that mostly didn’t need analysis. The cost of this “intelligence” on the logs rivaled the cost of the infrastructure itself.

But there was another problem that no one had anticipated or that the consultancy had conveniently omitted. Sabrosa is not Lisbon. In the interior of the Alto Douro, Internet connectivity is not gigabit symmetric fiber. LusoFicta Foods relied on a connection that, on good days, was acceptable, but on bad weather days, with heavy rain or wind in the hills, degraded significantly. And when the Internet connection failed, everything failed. Without access to the cloud, the ERP simply did not exist. There was no offline mode, no plan B. Processing unit operators stopped working, drivers didn’t receive delivery notes, and invoicing froze. In a company that depended on shipping perishable goods on tight schedules, every hour without the system was direct financial loss.

On a single stormy day (common in the Douro between October and March), the company lost a whole day of dispatch. Mr. Teixeira calculated the loss and realized that this single day of downtime had cost more than a month of the server payments that had been proposed to him.

Mr. Teixeira looked at the cloud invoice, looked at the accountant, and said simply:

"Turn it off."

LusoFicta Foods returned to on-premises. But this time, they did it the right way.

The decision was not made out of stubbornness or aversion to technology. It was made because, finally, someone sat down with Mr. Teixeira and did the numbers correctly.

The company did not need the cloud. They did not need auto-scaling, AI pipelines analyzing logs, or elastic infrastructure spread across multiple availability zones. LusoFicta Foods needed one thing: a new server.

A server properly sized for the requirements of the new ERP version, capable of handling more than 300 simultaneous users during peak months, with enough disk space for years of operation, and with a support and maintenance contract.

The cost? Significant, yes, but finite. The server was acquired through a bank loan, with monthly installments the company could afford. For three years, LusoFicta Foods would have a fixed monthly payment to the bank. After those three years, the server would be paid off, and the monthly cost would disappear. Only the maintenance contract remained, a fraction of what they had been paying in the cloud.

In the cloud, the cost would be perpetual. Every month, the invoice from the cloud service provider would arrive, and even if optimized, even with excesses corrected, it would represent a recurring and permanent expense that, after three years, would have far exceeded the cost of the physical server. And it would continue. In the fourth year, the fifth, the tenth. Forever.

The story of LusoFicta Foods is not a story against the cloud. The cloud is an extraordinary tool for those who need it. Companies with truly unpredictable workloads, startups that need to scale from zero to millions without upfront investment, organizations with globally distributed teams, SaaS platforms serving clients across multiple time zones: for all of these, the cloud is often the right answer.

But not every company is a startup. Not every workload is unpredictable. Not every performance problem is solved with more infrastructure. And not every location has the connectivity that the cloud requires.

LusoFicta Foods had a seasonal workload, yes, but it was predictable. Every year, the campaign starts in September and ends in February. Every year, the staffing peak is between October and November. There are no surprises. No unpredictable spikes at three in the morning from users on the other side of the world. There is an agro-industrial company in the Douro that, like all others, follows the rhythm of the seasons.

It did not need elasticity; it needed capacity. It did not need AI monitoring; it needed a system that worked without being watched 24/7. And it did not need to depend on an Internet connection for 300 people to be able to work.

The mistake was not the cloud. The mistake was in the diagnosis.

The first consultancy correctly identified the problem (insufficient hardware) but could not convince the client to invest. The second consultancy sold a solution disproportionate to the real problem, wrapped in promises of modernization and digital transformation. Neither effectively solved the fundamental issue: LusoFicta Foods needed a new server, and someone needed to help Mr. Teixeira understand that this investment, though painful in the short term, was the most sensible path in the long run.

Before migrating to the cloud, ask yourself:

• What is the real problem I am trying to solve? If the answer is "my hardware is insufficient," the solution may be as simple as buying new hardware.

- Is my workload truly unpredictable? Seasonal is not the same as unpredictable. If you know exactly when the peaks will occur, you can size for them once.

• Does my location support total dependence on the Internet? For companies in areas with limited or unstable connectivity, keeping critical infrastructure on-site is not a step backward. It’s resilience.

• Have I done the long-term calculations? The monthly cost of the cloud may seem manageable until you sum it over three, five, or ten years and compare it to a one-time investment in on-premises infrastructure.

• Is the proposed solution proportional to the problem? An AI pipeline analyzing logs for a 300-person agro-industrial company is like using a military surveillance drone to check if the grapes are ripe.

The cloud is not the answer to everything. Sometimes, the answer is a new server, a bank loan, and a good breakfast overlooking the vineyards, without worries.

And if you need help doing the calculations or answering these questions, Aubay is ready to assist.

This article depicts an entirely fictional situation. All company names, people, and entities are invented and do not correspond to real organizations or individuals. The narrative was constructed based on recurring patterns observed in technology migration projects, with the goal of encouraging reflection on infrastructure decisions.

This article was written by Anderson Leite, Alumni of Aubay Portugal, and can be read on his blog here.