1. Home
  2. Business
  3. History of ERP

The evolution of ERP: from punched cards to AI agents

How a 1913 formula for ordering stock became the system that runs the world’s companies: MRP, MRP II, ERP, the cloud and now AI agents, with the vendors that shaped each era and the market as it stands in 2026.

  • MRP
  • MRP II
  • ERP
  • Cloud ERP
  • AI agents

What is ERP?

Enterprise resource planning (ERP) is the software that keeps a company’s core processes, such as finance, purchasing, inventory, production, sales and people, on one shared set of data. A sale, a goods receipt and a journal entry become three views of the same event instead of three records kept by three departments.

ERP did not appear from nowhere. Each generation answered a bigger question than the one before: how much to order (1910s), which materials to buy and when (MRP, 1964), whether the plant and the money can support the plan (MRP II, 1983), how to run the whole company on one system (ERP, 1990) and now how much of the work the system can do by itself.

One idea runs through all of them: record each fact once, plan from it, and let every department see the same numbers.

The big picture: eight eras, one question each

Pick an era to see the question it answered, what it added and the names to know. The sections below go deeper into each one.

Era
1890 – 1963

Inventory control

The question it answered

How much should I order, and when?

Tabulating machines, then mainframes running nightly batches.

What it added

  • Economic order quantity (1913) and a reorder point for each item
  • Punched cards for stock records, payroll and billing
  • The first business computers (LEO I, 1951) take over the clerks’ ledgers

Names to know

  • Herman Hollerith
  • Ford W. Harris
  • LEO I
  • IBM
1964

Material requirements planning

The question it answered

Which materials do I need, how many, and by when?

IBM mainframes and weekly batch regeneration.

What it added

  • Demand for parts calculated from the bill of materials, not guessed
  • A time-phased plan: net requirements offset by lead time
  • Regular planning runs on the mainframe

Names to know

  • Joseph Orlicky
  • Black & Decker
  • IBM
1970s

Closed-loop MRP

The question it answered

Can the plant actually build the plan?

Online terminals reach the shop floor and the purchasing office.

What it added

  • Capacity requirements planning for each work centre
  • Shop-floor and purchasing feedback into the next run
  • The master schedule managed as a firm commitment

Names to know

  • APICS
  • George Plossl
  • Oliver Wight
1983

Manufacturing resource planning

The question it answered

Do the plan, the plant and the money agree?

Minicomputers and midrange systems such as IBM System/38 and AS/400.

What it added

  • Sales and operations planning at the top of the loop
  • Costs and budgets taken from the same plan
  • What-if simulation before committing

Names to know

  • Oliver Wight
  • SAP R/2
  • JD Edwards
  • MAPICS
1990

Enterprise resource planning

The question it answered

Can the whole company run on one system?

Client-server computing and relational databases.

What it added

  • Finance, HR, sales, purchasing and production on one database
  • Standard processes delivered as configurable software
  • Logistics and accounting posted together, in real time

Names to know

  • Gartner
  • SAP R/3
  • Oracle
  • PeopleSoft
  • Baan
  • JD Edwards
1998 – 2010

ERP II and the first cloud ERP

The question it answered

Can customers, suppliers and the web join in?

Web browsers and multi-tenant cloud software.

What it added

  • CRM, supply chain and e-commerce around the ERP core
  • Software as a service: no servers to run
  • Mergers reshape the vendor landscape

Names to know

  • NetSuite
  • Workday
  • SAP ECC 6.0
  • Microsoft Dynamics
2011 – 2019

In-memory and postmodern ERP

The question it answered

Can we see it in real time, and keep only what we need?

In-memory columnar databases and browser-based apps.

What it added

  • In-memory databases replace aggregates and overnight reports
  • Simpler data models, such as one journal for finance and controlling
  • A core suite plus loosely connected cloud applications

Names to know

  • SAP HANA
  • SAP S/4HANA
  • Oracle Fusion
  • Dynamics 365
2020 – today

Composable and agentic ERP

The question it answered

Can the system act, not just record?

Public and private cloud, large language models and agent runtimes.

What it added

  • Cloud subscriptions and a clean core
  • Composable building blocks joined by APIs
  • AI assistants, then agents that reconcile, match invoices and help close the books

Names to know

  • RISE with SAP
  • Joule
  • Oracle Fusion Agentic Applications
  • Microsoft Copilot

1890 – 1963

Before MRP: order points and punched cards

For the first half of the twentieth century, planning meant managing each item on its own: keep enough in stock, reorder when it runs low. Machines helped count, but not think ahead.

  • 1890

    Punched cards

    Herman Hollerith’s tabulating machines counted the 1890 US census. His company later became part of IBM, and punched cards ran stock records, payroll and billing for decades.

  • 1913

    Economic order quantity

    Ford W. Harris published the order size that balances the cost of placing an order against the cost of holding stock.

    EOQ = √(2DS / H)D annual demand · S cost per order · H holding cost per unit per year
  • Order point

    Reorder point and safety stock

    Each item is reordered when stock falls below a set level: the usage expected during the lead time, plus a safety margin. Simple and robust, but blind to what the factory will build next.

  • 1951

    The first business computer

    British caterer J. Lyons & Co. ran LEO I, the first computer put to routine office work, on valuations, payroll and stock. Through the 1950s, mainframes took over inventory files in nightly batches.

  • Early 1960s

    Bill of materials processors

    Programs such as IBM’s BOMP stored product structures on the computer: which parts and quantities go into each product. That data made the next step possible.

The limit of order points. They treat every item as if its demand were independent. But nobody buys a bicycle wheel on its own: the number of wheels needed depends on how many bicycles are planned. Calculating that dependent demand is what MRP does.

1964

Material requirements planning (MRP)

In 1964 Joseph Orlicky, an engineer at IBM, set out the principles of MRP, and Black & Decker became the first company to put them to work. Instead of watching each item’s stock level, MRP starts from what the company plans to build and works backwards to every part it needs.

It takes three inputs and produces one plan:

  • Master production schedule: what will be built, how many, and when.
  • Bill of materials: what each product is made of.
  • Inventory records: what is in stock and what is already on order.

The output is a time-phased list of planned orders: what to buy or make, how much, and when to start so it arrives on time.

MRP in four steps: 100 bicycles for week 6

  1. Master schedule100 bicycles are due in week 6. Assembly takes one week, so the parts must be there in week 5.
  2. Explode the bill of materialsEach bicycle needs 1 frame and 2 wheels: gross requirements of 100 frames and 200 wheels in week 5.
  3. Net against stock30 frames and 40 wheels are in stock, so the net requirements are 70 frames and 160 wheels.
  4. Offset by lead timeFrames take 3 weeks to arrive, wheels 2 weeks: order 70 frames in week 2 and 160 wheels in week 3.
The resulting time-phased plan
ItemLead timeIn stockWeek 1Week 2Week 3Week 4Week 5Week 6
Bicycle1 week0Start 100Due 100
Frame (×1)3 weeks30Order 70Need 100
Wheel (×2)2 weeks40Order 160Need 200

The same logic, 60 years later. In SAP S/4HANA it runs as MRP Live (MD01N), and its output is still planned orders and purchase requisitions.

1970s – 1983

Closing the loop: from MRP to MRP II

The first MRP systems assumed the factory could build anything on time. In the 1970s the APICS “MRP crusade”, led by Orlicky, George Plossl and Oliver Wight, pushed for closed-loop MRP: check the plan against the capacity of each work centre, and feed what really happened on the shop floor back into the next run.

In 1983 Oliver Wight extended it into manufacturing resource planning (MRP II), adding sales and operations planning, master scheduling, rough-cut capacity planning and the money: costs and budgets from the same plan. The “II” only tells it apart from the first MRP, which shares the initials.

The MRP II loop

  • MRP (1964)
  • Closed loop (1970s)
  • MRP II (1983)
  1. Sales and operations planningMonthly volumes agreed by sales, operations and finance (MRP II (1983))
  2. Master production scheduleWhat to build, week by week (MRP (1964))
  3. Rough-cut capacityIs the schedule realistic for key resources? (MRP II (1983))
  4. Material requirementsPlanned orders for every part (MRP (1964))
  5. Capacity requirementsLoad on each work centre, period by period (Closed loop (1970s))
  6. Purchasing and shop floorOrders released, work confirmed (Closed loop (1970s))

Feedback: receipts, confirmations and delays flow back, and the next planning run starts from reality.

Finance: standard costs, budgets and cash projections come from the same plan, so operations and finance finally speak the same numbers.

By the end of the 1980s MRP II was a billion-dollar software market in the United States alone, sold by vendors such as IBM (MAPICS), JD Edwards and, in Europe, SAP with R/2.

1990s

1990: ERP, one system for the whole company

By 1990 the research firm Gartner was calling the next generation of MRP II enterprise resource planning: one system and one database for finance, human resources, sales, distribution and purchasing, not just the factory. Client-server computing and relational databases made it affordable, and SAP R/3, released in 1992, became the reference design.

The decade ended in a rush. Facing the year 2000 problem, many companies replaced home-grown systems with packaged ERP instead of fixing two-digit years. That turned the late 1990s into a boom and the early 2000s into a slump, and taught the industry hard lessons about projects that change processes, not just software.

The vendors of the 1990s, and what became of them

  • SAP

    1972 · Germany

    Founded by five former IBM employees. R/2 (1979) on mainframes, R/3 (1992) on client-server.

    Today: independent; SAP S/4HANA since 2015.

  • Oracle

    1977 · USA

    The database company moved into financial applications in the late 1980s, later sold as Oracle E-Business Suite.

    Today: bought PeopleSoft (2005) and NetSuite (2016); Fusion Cloud ERP.

  • JD Edwards

    1977 · USA

    ERP for IBM midrange computers (System/38, then AS/400), popular with midsize manufacturers and distributors.

    Today: PeopleSoft (2003), then Oracle (2005); still sold as JD Edwards EnterpriseOne.

  • Baan

    1978 · Netherlands

    A strong rival to SAP in manufacturing during the 1990s.

    Today: Invensys (2000), SSA Global (2003), Infor (2006).

  • PeopleSoft

    1987 · USA

    Started with human resources, then finance; known for client-server applications that users liked.

    Today: Oracle, after a hostile bid, in January 2005.

  • Great Plains and Navision

    USA · Denmark

    Accounting and ERP for small and midsize companies, sold through partners.

    Today: Microsoft (2001 and 2002), now Dynamics 365.

MRP, MRP II and ERP side by side

MRPMRP IIERPCloud and AI ERP
Main questionWhat to buy and make, and whenCan the plant and the money support the planCan the whole company run on one systemCan the system act on its own
ScopeMaterialsManufacturing: materials, capacity, costsEvery departmentThe company, its partners and AI agents
Main usersPlanners and buyersProduction, planning, financeEveryone in the companyEveryone, with agents doing routine work
FinanceNot includedCosts and budgets from the planIntegrated general ledger, posted in real timeContinuous close and predictive figures
TechnologyMainframes, batch runsMinicomputers and midrange systemsClient-server, relational databasesCloud, in-memory databases, language models
Planning rhythmWeekly regenerationDaily net changeReal-time posting, overnight MRPLive MRP on demand

1998 – 2019

ERP II, the cloud and in-memory computing

Once the whole company ran on one system, the next questions were how to open it to customers and suppliers, how to stop running servers, and how to see the numbers the moment they happen.

  1. 1998

    NetSuite, founded as NetLedger, offers the first ERP delivered over the internet.

  2. 2000

    Gartner describes ERP II: ERP opened to customers, suppliers and the web, with CRM, supply chain and e-commerce around the core.

  3. 2005

    After Oracle takes over PeopleSoft, its founder Dave Duffield and Aneel Bhusri start Workday, a cloud-only suite for HR and finance.

  4. 2006

    SAP ERP 6.0 (ECC 6.0), the long-lived release behind most SAP systems before S/4HANA.

  5. 2011

    Oracle ships Fusion Applications, and SAP’s in-memory database SAP HANA becomes generally available.

  6. 2013

    Gartner coins postmodern ERP: the monolith splits into a core plus loosely connected cloud applications.

  7. 2015

    SAP S/4HANA arrives, built for HANA, with one journal for finance and controlling: the Universal Journal, ACDOCA.

  8. 2016

    Microsoft launches Dynamics 365, and Oracle buys NetSuite.

What the cloud changed

The vendor runs the software for many customers at once and updates everyone on the same schedule. Companies trade some freedom to customize for lower running costs and regular releases.

What in-memory changed

With data held in memory, totals can be calculated on the fly. Aggregate tables and overnight reports disappear, and the data model gets simpler: one line item feeds every report.

2020 – 2026

Today: cloud subscriptions, a clean core and AI agents

The large vendors now sell ERP as a cloud subscription. SAP packaged the move as RISE with SAP (2021) and GROW with SAP (2023), and in 2025 renamed its cloud editions SAP Cloud ERP (public) and SAP Cloud ERP Private. The watchword is a clean core: standard processes in the ERP, extensions built beside it.

Gartner’s newest label, composable ERP, describes the same shift: business capabilities assembled from building blocks joined by APIs, mixing cloud and on-premise. On top of that core, AI moved from assistants that answer questions (SAP Joule and Microsoft Copilot, both 2023) to agents that do the work.

The agentic turn in 2026

  • SAP

    May 2026

    At Sapphire, SAP presented the SAP Autonomous Suite: more than 50 Joule Assistants across finance, supply chain, procurement, HR and customer experience, coordinating over 200 specialised agents, such as an assistant for the financial close.

  • Oracle

    October 2026

    Oracle released 25 Fusion Agentic Applications across Fusion Cloud ERP, HCM, supply chain and manufacturing, and customer experience, running on a new agent runtime.

  • Microsoft

    2026 release wave 1

    Dynamics 365 Finance gained an Account Reconciliation Agent; Business Central ships Sales Order and Payables agents and lets customers design their own.

The SAP ECC dates that matter

For the many companies still on SAP ERP 6.0, three dates frame the move to SAP S/4HANA.

  1. End of 2027Mainstream maintenance endsThe standard support of SAP ERP 6.0 under the regular maintenance fee.
  2. End of 2030Extended maintenance endsOptional, at an extra fee, for customers who need more time.
  3. End of 2033Transition option endsThe SAP ERP private edition transition option: a cloud subscription for large, complex landscapes on their way to S/4HANA. Not an extension of maintenance.

The ERP market in 2026

ERP software sales grew 11% in 2025, to about $146 billion. Oracle and SAP are almost level at the top, yet the ten largest vendors hold less than a third of the market: ERP is still served by thousands of regional and industry specialists.

  • Market 2025$145.6 bnERP applications, worldwide
  • Growth+11%from $131.6 bn in 2024
  • Top 10 share31.9%a fragmented market
  • Forecast 2030$204.7 bn7.1% a year from 2025
Top 10 ERP vendors by 2025 revenueERP suitesAccounting and industry software
  1. Oracle6.7% · $9.7 bn (ERP suite)
  2. SAP6.6% · $9.6 bn (ERP suite)
  3. Intuit4.5% · $6.5 bn (accounting or industry software)
  4. Constellation Software4.0% · $5.8 bn (accounting or industry software)
  5. Microsoft2.2% · $3.2 bn (ERP suite)
  6. FIS2.0% · $2.9 bn (accounting or industry software)
  7. Roper2.0% · $2.9 bn (accounting or industry software)
  8. Sage1.7% · $2.5 bn (ERP suite)
  9. Infor1.3% · $1.9 bn (ERP suite)
  10. Workday1.0% · $1.4 bn (ERP suite)

Source: Apps Run The World, August 2026 (estimates). It counts every ERP application, so accounting software (Intuit) and industry systems (Constellation Software, FIS, Roper) rank next to the ERP suites. Counted by suites alone, the order changes.

How many companies use each vendor’s ERPCustomers each vendor reports for its ERP products. SAP counts about twice as many companies as Oracle, mostly thanks to SAP Business One.Large-enterprise ERPSmall and midsize ERPAll products, not split
  1. SAP118,000+SAP S/4HANA and SAP ERP (ECC) about 35,000 · SAP Business One 83,000+
  2. Infor60,000+All Infor products 60,000+ (CloudSuite ERP plus HCM and asset management)
  3. Oracle55,000+Oracle Fusion Cloud ERP 11,000 · Oracle NetSuite 44,000+ · E-Business Suite and JD Edwards not published
  4. Microsoft55,000+Dynamics 365 Business Central (online) 55,000+ · Finance and Supply Chain, and on-premise Business Central and NAV, not published
  5. Epicor23,000+All Epicor products 23,000+ (Kinetic, Prophet 21, Eclipse and others)

Sources: SAP Business One 83,000+ (SAP, December 2025) · SAP ERP (ECC) about 35,000 customers, most of them moving to S/4HANA (Gartner, March 2025) · Infor 60,000+ (Infor, 2026) · Oracle NetSuite 44,000+ (Oracle, 2026) · Oracle Fusion Cloud ERP 11,000 (Oracle, October 2025) · Dynamics 365 Business Central online 55,000+ (Microsoft, April 2026) · Epicor 23,000+ (Epicor, 2024).

Each vendor counts its own customers, in different years and not always per product. A company that runs two products is counted twice. These are companies, not users.

Gartner 2025 Leaders · product-centric

  • Oracle Fusion Cloud ERP
  • Microsoft
  • SAP Cloud ERP
  • IFS
  • Oracle NetSuite
  • Epicor
  • Infor

Magic Quadrant for Cloud ERP for Product-Centric Enterprises, October 2025

Gartner 2025 Leaders · service-centric

  • SAP Cloud ERP
  • Workday

Both named Leaders for the fourth year in a row in the Magic Quadrant for Cloud ERP for Service-Centric Enterprises, October 2025

Who sells ERP to whom

Company size still decides most shortlists. Pick a segment to see the main products in it.

Company size

Large enterprises

Global groups with many countries, legal entities and industries. Long projects, deep localization and industry solutions.

ProductVendor and originBest known forDeployment
SAP S/4HANA (SAP Cloud ERP, SAP Cloud ERP Private)SAP · Germany, 1972Finance, controlling and manufacturing in depth; industry solutions and country versions for multinational groupsPublic cloud, private cloud, on-premise
Oracle Fusion Cloud ERPOracle · USA, 1977Cloud-native finance, procurement, projects and performance management, updated every quarterPublic cloud
Microsoft Dynamics 365 Finance and Supply Chain ManagementMicrosoft · USAClose ties to Microsoft 365, Power Platform and CopilotCloud
Workday Financial ManagementWorkday · USA, 2005Finance and HR on one platform for services, education and the public sectorPublic cloud
Infor CloudSuiteInfor · USA, 2002Suites by industry: distribution, manufacturing, healthcareCloud
JD Edwards EnterpriseOne and Oracle E-Business SuiteOracle · USALong-running on-premise suites with large installed basesOn-premise or hosted

Midsize companies

Growing companies and subsidiaries of larger groups. Faster projects, cloud first, strong fit for one or two industries.

ProductVendor and originBest known forDeployment
Oracle NetSuiteOracle · USA, 1998The original cloud ERP; fast-growing companies and subsidiariesPublic cloud
Epicor KineticEpicor · USA, 1972Discrete manufacturing and distributionCloud or on-premise
IFS CloudIFS · Sweden, 1983Asset-heavy industries: aerospace, energy, field serviceCloud or on-premise
AcumaticaAcumatica · USA, 2008Cloud ERP priced by resources rather than users; owned by Vista Equity Partners since 2025Cloud
Sage Intacct and Sage X3Sage · United Kingdom, 1981Intacct for cloud financials, X3 for manufacturing and distributionCloud or on-premise
Unit4 ERPUnit4 · Netherlands, 1980People-based services: professional services, public sector, non-profitCloud
PlexRockwell Automation · USAManufacturing ERP joined to execution on the shop floorCloud

Small businesses

Companies that outgrow accounting software. Quick to start, priced per user or per app, often sold and run by a local partner.

ProductVendor and originBest known forDeployment
SAP Business OneSAP · GermanySmall companies and small subsidiaries of SAP groups, sold through partnersOn-premise or partner cloud
Microsoft Dynamics 365 Business CentralMicrosoft · USA, 2018Successor of Navision, for firms already on Microsoft 365Cloud or on-premise
OdooOdoo · Belgium, 2005Open-source core with dozens of apps; version 19 released in September 2025Cloud or self-hosted
TOTVS (Protheus, RM, Datasul)TOTVS · Brazil, 1983Leader among smaller Brazilian companies, with Brazilian tax built inCloud or on-premise
ERPNextFrappe · IndiaFully open-source ERP with a large communityCloud or self-hosted
QuickBooks and XeroIntuit · USA; Xero · New ZealandAccounting software where most small firms start before an ERPCloud

What six decades of ERP teach

  1. Record each fact once

    Every generation removed duplicate records, from card files to the Universal Journal. Fewer copies mean fewer reconciliations.

  2. Master data decides the plan

    MRP only worked because the bill of materials described the product. Bad master data still breaks the best ERP.

  3. Process before software

    The projects that failed in the 1990s mostly failed on people and processes, not code. Standard processes and a clean core age better than custom ones.

  4. Each era keeps the last

    MRP still runs inside every ERP, and AI agents act on the same integrated data. The foundations matter more, not less.

Frequently asked questions

What is the difference between MRP and ERP?

MRP plans materials: what to buy and make, and when, from the production schedule, the bill of materials and stock. ERP runs the whole company on one system, including finance, sales, purchasing and HR, and contains MRP as one of its functions.

What does MRP II stand for?

Manufacturing resource planning. Oliver Wight introduced it in 1983 as MRP extended with sales and operations planning, capacity planning and finance. The “II” distinguishes it from material requirements planning, which has the same initials.

Who coined the term ERP?

The research firm Gartner, around 1990, for systems that extended MRP II beyond manufacturing to the whole enterprise. The method underneath goes back to MRP in 1964.

Which is the largest ERP vendor?

It depends on what is counted. By revenue, Apps Run The World puts Oracle first with 6.7% of 2025 ERP revenue and SAP second with 6.6%. By number of companies, SAP leads: about 118,000 customers for its ERP products (83,000+ on SAP Business One), roughly twice the 55,000+ that Oracle publishes.

When does support for SAP ECC end?

Mainstream maintenance of SAP ERP 6.0 ends at the end of 2027 and optional extended maintenance at the end of 2030. The SAP ERP private edition transition option runs to the end of 2033 for large, complex landscapes moving to SAP S/4HANA.

What is composable ERP?

Gartner’s term for building ERP from interchangeable business capabilities connected by APIs, instead of one monolithic suite. In practice it means a stable core plus cloud applications and extensions that can change without touching it.

Will AI agents replace ERP?

Not on current evidence. The agents that SAP, Oracle and Microsoft released in 2026 work inside the ERP, on its data and its rules. They change who does the routine work, not where the records live.