What is an ERP, when is it worth it, and what changes with AI agents

Key Takeaways

• An ERP is a system that brings a company's core processes into a single database: finance, purchasing, inventory, production, payroll and sales. Its value is that every area works from the same data.
• In Colombia, 20% of companies with 10 to 49 employees used an ERP in 2020, compared with 36% across the OECD in 2025 (OECD, ICT Access and Usage by Businesses). Among companies with more than 250 employees the figure rises to 76%.
• When an implementation runs late, the most cited cause is organizational, not technical. 57.9% of late projects blame governance, resistance to change or process redesign (Panorama Consulting, The 2026 ERP Report).
• An ERP imposes its own logic on the company. Thomas Davenport warned about this in 1998 in Harvard Business Review: the system pushes toward generic processes even when your own are an advantage.
• AI agents don't replace the ERP; they make it more important. An agent needs clean data and reliable transactions to act, which is exactly what a well-implemented ERP holds. What changes is how it gets used and what you should demand when you buy one.

The company invoices in one program, tracks inventory in a spreadsheet, takes orders over WhatsApp and reconciles payroll in another sheet. Every month-end someone spends two days matching numbers that don't agree. At some point a software vendor shows up and the answer seems obvious: buy an ERP.

Sometimes it is. Other times the company spends a year implementing a system that ends up running alongside the same spreadsheets as before. This guide explains what an ERP is, when it is worth it, why so many implementations go wrong and what is changing now that vendors are building AI agents into their systems. We wrote it at Suricata Labs, which doesn't sell or implement software. It is part of our guide to digital transformation for businesses.

What is an ERP?

An ERP (enterprise resource planning) is a set of integrated applications that share one database and one process model to run a company's core operations. Gartner defines it as an integrated suite of business applications with a common process and data model covering end-to-end processes such as finance, HR, distribution, manufacturing, service and the supply chain.

The word that matters is "integrated". When the warehouse ships an order, inventory goes down, the invoice is generated, the receivable shows up in accounting and the manager sees the sale on the dashboard, with nobody typing anything twice.

Not every company needs every module. A distributor can do without production, and a services firm may need very little inventory.

What is an ERP for, and what doesn't it solve?

An ERP lets the whole company work from the same data, with standard, traceable processes. That cuts repeated data entry, transcription errors and closing time, and gives management a single version of the numbers.

There is evidence it can pay off, with caveats. Lorin Hitt, D. J. Wu and Xiaoge Zhou studied companies that implemented SAP R/3 between 1986 and 1998 in Investment in Enterprise Resource Planning (Journal of Management Information Systems, 2002). They found that firms investing in ERP showed higher performance across a wide range of financial metrics, although with a slowdown in performance and productivity shortly after implementation. It is an association, not proof of causation, and the early dip is real.

What an ERP doesn't solve is a messy process. Thomas Davenport wrote this in 1998 in Putting the Enterprise into the Enterprise System: technical challenges are not the main reason these systems fail; the biggest problems are business problems. A system like this imposes its own logic on the company's strategy, organization and culture, and pushes it toward generic processes even when its own processes are a source of advantage.

That cuts two ways. For most administrative processes, adopting the ERP's standard is a gain. For the process that sets the company apart from its competitors, think twice before letting the software redefine it.

How many companies use ERP in Colombia?

In Colombia, one in three companies with 10 or more employees used an ERP in 2020, the latest figure the OECD reports for the country. The gap is in smaller companies: among those with 10 to 49 employees, only one in five.

Source: OECD, ICT Access and Usage by Businesses database, indicator for businesses using ERP software, percentage of enterprises with 10 or more employees.

By sector, use in Colombia ranged from 53% in information technology to 19% in retail, with manufacturing at 34%. A mid-size Colombian company without an ERP is the norm, and that alone is not a sign of being behind. The useful question is whether the cost of not having one is already higher than the cost of implementing it.

When is an ERP worth it?

An ERP is worth it when the cost of running on separate systems (hours of data entry, errors, miscounted inventory, slow closes, decisions made with old numbers) is already higher than the cost and risk of implementing one. That calculation is rarely done, and it is the one that should decide.

A note for Colombia: mandatory electronic invoicing is not, on its own, a reason to buy an ERP. The tax authority (DIAN) regulates invoicing in Resolution 000165 of 2023 and electronic payroll in Resolution 000013 of 2021, and offers a free invoicing solution to businesses registered in the RUT. Many small companies comply with that and good accounting software.

Why do ERP implementations fail?

ERP implementations fail mostly for organizational reasons: scope that grows unchecked, processes nobody redesigned and people who don't adopt the system. Technical failures happen, but they are rarely the main cause.

Panorama Consulting's 2026 ERP Report, with 170 organizations surveyed (mostly large and multinational), found that 30% went over budget and 22% over schedule, with a median duration of nine months. Among late projects, the most cited reason was organizational issues (57.9%), followed by expanded scope (55.3%). Fewer than a quarter of organizations reported an intense focus on change management. Panorama is an ERP consultancy and its sample is not random, so the figures are a reference point.

Gartner goes further with a prediction: by 2027, more than 70% of recently implemented ERP initiatives will not fully meet their original business case goals. It is an analyst projection, not a measurement.

The best-known case of why this happens is Lidl. The German supermarket chain canceled its SAP project in 2018 after seven years, at a cost Handelsblatt estimated at around 500 million euros. According to that report, Lidl valued its inventory at purchase price, while SAP's retail standard uses retail price. Lidl didn't want to change how it worked, the system was customized to fit, and the project became slow and expensive. Its CEO wrote to employees that the original strategic goals could not be reached at an acceptable cost.

The underlying cause is usually the one we explain in our guide to change management: the software gets budgeted and the work of getting people to use it gets left out.

What changes in ERP with artificial intelligence and agents?

AI is changing how an ERP is used and the debate over whether you need one, but not the fact that a company needs a reliable record of its transactions. This is the section most likely to move in the coming years, so we will review it often.

Satya Nadella, Microsoft's CEO, opened the debate in an interview on the BG2 podcast in December 2024. His argument, widely quoted since, is that business applications are essentially databases with business logic on top, and that this logic will increasingly move into AI agents. The media read it as the end of enterprise software, but the argument is more precise: what changes is where the logic lives, and the database stays.

ERP vendors are already moving in that direction:

Gartner estimates that by 2028, 33% of enterprise software applications will include agentic AI, up from less than 1% in 2024. In the same release it predicts that over 40% of agentic AI projects will be canceled by the end of 2027 due to costs, unclear value or poor risk controls, and warns that integrating agents into legacy systems is often complex and costly.

Klarna shows both sides. In 2024 its CEO said the company had dropped Salesforce and would drop Workday, and it was read as proof that AI would replace enterprise software. In March 2025 he clarified publicly that they had not replaced that software with a language model; they had built an internal platform to bring their data together, and he doubted other companies would do the same.

For a mid-size company evaluating an ERP today, this translates into four practical changes:

• Clean data is worth more than before. An AI agent that takes orders or reconciles payments only works if inventory, customers and prices are properly recorded. A poorly fed ERP produces agents that get things wrong quickly.
• The interface matters less. Much of ERP training is learning to navigate screens. If queries and repetitive tasks are done by talking to an assistant, adoption can be easier for people who don't use the system every day.
• Customization becomes less necessary. Adapting the system to your own process used to require custom programming, which is what sank Lidl. Today many of those needs can be covered with automations and agents working on top of the standard system. We explain this in business process automation: when you need AI and when you don't.
• Access to your data becomes a buying condition. If the ERP doesn't let you read and write its data through open interfaces, no outside agent can work with it, and the company is tied to whatever the vendor decides to build.

This continues an idea Gartner has been pushing since 2014, when it described postmodern ERP: a smaller standard core surrounded by connected services. Agents take that logic a step further. The core that records transactions stays, and what surrounds it becomes more flexible.

How to decide whether your company needs an ERP, step by step

These seven steps are for before you ask a vendor for the first demo. They save the company from the most expensive mistake, which is buying before knowing what it needs.

1. Map the processes the ERP will touch

Before looking at software, document how sales, purchasing, inventory and accounting work today, exceptions included. Without that there is no way to judge whether a system fits. The method is in our guide to the business process map.

2. Calculate what not having one costs you today

Add up the hours spent entering the same data several times, reconciling inventory and closing the month, plus the cost of errors: orders shipped without stock, invoices with wrong data, rush purchases. That figure is what you compare against the full cost of implementation, licenses and team time included.

3. Exhaust the simpler alternatives first

If the problem is in a single process, a specialized system or an integration between current tools may solve it for much less. In Colombia, electronic invoicing and payroll can be handled without an ERP. An ERP makes sense when the problem is that areas don't share data.

4. Adapt your processes to the standard before customizing the system

For administrative processes, accept the ERP's way of working whenever you can. Save customization for the process that truly sets you apart, and before programming it, ask whether an automation on top of the standard system could handle it. Every customization makes upgrades more expensive and ties you to the implementer.

5. Demand open access to your data

Ask how data is exported, whether the system has documented application programming interfaces (APIs) and whether it lets you connect agents or third-party tools. Also ask what happens to your data if you switch vendors. With AI agents on the horizon, this answer matters as much as the feature list.

6. Budget for adoption and the temporary dip

Set aside time and money to redesign processes, migrate and clean data, train by role and support people in the months after launch. Expect performance to drop for a while after go-live, as Hitt, Wu and Zhou found. And name the person responsible for adoption from the start.

7. Start in phases and away from your peak season

Begin with the module that solves the biggest pain, stabilize it and then continue. Never go live right before your sales peak, because that is when a shipping error costs the most. The case below shows why.

Real case: the ERP that ran into Halloween

In July 1999, Hershey, the American chocolate maker, launched a new integrated system for orders, warehousing and shipping, with software from SAP, Siebel and Manugistics. According to Computerworld, the project had cost 112 million dollars.

The launch coincided with the back-to-school and Halloween season, the most important of the year for the company. Orders came in, but the system couldn't process and ship them on time. In September, the company's CEO announced it would not be able to deliver around 100 million dollars' worth of products, and the stock fell more than 8% that day, according to CIO.

The cost shows up in the official filings. In its third-quarter 1999 report to the SEC, Hershey reported net sales falling from 1,217.2 to 1,066.7 million dollars versus the same quarter the year before, and net income from 107.5 to 87.6 million. The company attributed it to order fulfillment problems (customer service, warehousing and shipping) since the start-up of the new system. In its annual report it added lost sales, higher freight and warehousing costs and higher inventories.

Hershey fixed the system and kept running on it. What the case leaves is that the timing and the way you go live weigh as much as the software you choose, and that the problems hit exactly the process the customer felt: getting their order on time.

Frequently asked questions about ERP

What is the difference between an ERP and a CRM?

An ERP handles internal operations (finance, inventory, purchasing, production, payroll) and a CRM handles customer relationships (leads, opportunities, sales follow-up). Many ERPs include a sales module, and many companies connect a specialized CRM to the ERP so a closed deal flows straight into invoicing and shipping.

Is an ERP the same as accounting software?

No. Accounting software records financial transactions; an ERP connects accounting with every other process, so shipping, purchasing or payroll create their accounting entries without retyping. For many small companies, good accounting software connected to electronic invoicing is enough.

Are there free ERPs?

There are open-source ERPs with no license fee, such as Odoo's Community edition, which sits alongside a paid Enterprise edition. No license fee doesn't mean no cost: configuration, data migration, support and team time are still the bulk of the investment.

How long does it take to implement an ERP?

The median duration was nine months in Panorama Consulting's 2026 survey, with a sample of mostly large companies. A mid-size company starting with a few modules can take less. What drags projects out is growing scope and processes redesigned along the way.

Will artificial intelligence replace ERPs?

Not on the horizon that matters for a buying decision today. AI agents change how the system is used and reduce the need to customize it, but they need a reliable record of transactions to work, which is what an ERP holds. What you should do is choose a system that lets you connect agents and get your data out without depending on the vendor.

Before buying the ERP, decide what it's for

Companies that come out of an ERP project well arrive at the first demo knowing which processes they want to integrate, what it costs them today to keep them separate and what they will demand in access to their data. Companies that come out badly tend to choose the most convincing demo.

At Suricata Labs we don't sell licenses or take commissions from any vendor. We help you map your processes, decide whether an ERP makes sense now and what to ask of it, and prepare your team to use it.

Schedule a conversation before you buy

Read also: Change management: why people don't use what you bought