You can build almost anything. This is the one you should not.
Engineering competence is not what decides an ERP program, and the best proof is a company that runs the world's largest cloud platform and still could not land an HCM deployment. Find your stage, read the decision that belongs to it, then run the two-minute check.
The distribution is a power law, and that is the whole problem
Engineering estimation is built for normal curves, where the average describes the typical case and the tails are small. Enterprise systems programs do not distribute that way. Most land near plan, and a meaningful minority go somewhere else entirely.
Illustrative shape, real endpoints. The 27% average, the one in six frequency, and the 200% and 70% figures come from a published study of 1,471 IT projects, with a follow-on establishing the distribution as a power law rather than a normal curve. The bar heights show the shape of that argument; they are not the study's histogram. Separately, Gartner's published expectation is that more than 70% of recently implemented ERP initiatives will fail to fully meet original business case goals by 2027, with as many as 25% failing catastrophically.
The right decision depends on where the company sits today
Five stages, one decision each. Pick the one you are in, or the one you are about to enter, which is usually the more useful read.
Everyone waits for a revenue number to justify a real finance system. The revenue thresholds people quote range from $5M to $30M, which is a clue that revenue is the wrong trigger.
Trigger on legal entity count, not revenue. The practitioner threshold is 3 to 5 entities, the point where spreadsheet consolidation and manual intercompany elimination become an audit finding rather than a workaround.
An entity is not a reporting dimension. It carries its own statutory ledger, its own local filing obligations, its own intercompany relationships, and its own functional currency. Companies that add entities faster than they add finance systems accumulate a reconciliation debt that surfaces at exactly the wrong moment, which is usually the first audit that matters. The other reason to move early: the argument that engineering can build it is strongest here and weakest in hindsight. A company that operates one of the world's largest cloud platforms signed for HCM and later mutually agreed to discontinue the company-wide deployment, a decision that took up to 7.8% off the vendor's share price the day it became public. Engineering capability was never the constraint.
How many legal entities do we have, and how many will we have in twelve months?
Engineering hires wherever the talent is, and the systems conversation happens afterward. By the time finance sees it, there are people in eight countries and a payroll answer for two of them.
Scope the payroll estate as an integration program from the start, and make worker location a first-class dimension on every labor and vendor transaction.
Workday runs native payroll in five countries: the US, UK, Canada, France and Australia. Everything else runs through a connected provider or a partner. Any distributed engineering organization is therefore buying integrations, not modules, and the difference belongs in the business case rather than in a change order. Tax makes it sharper. The 2025 restoration of immediate expensing for research costs was domestic only, and foreign research and experimental spend still amortizes over 15 years. That means the domestic and foreign split has to be derivable from source transactions, using worker location, cost center, project and supplier country as reliable dimensions. Most scale-ups solve this with a quarterly analyst spreadsheet mapping headcount to projects. That approach survives neither an audit of the credit nor a controls walkthrough.
Can we split research spend domestic from foreign at the transaction, without a spreadsheet?
The company moves to usage-based or AI pricing because the market demands it. The revenue accounting built for annual subscriptions does not move with it, and nobody tells finance until the first close.
Design the revenue model and the product catalogue together, before the pricing goes live, and treat the shortcut you are currently relying on as expiring.
Just over half of public SaaS companies now carry a usage-based component, up from roughly a quarter in 2021. The mechanics that break: prepaid credits and token bundles are contract liabilities recognized on consumption rather than purchase; breakage on unused expiring credits needs an estimate that a new AI product has no history to support, which makes the estimate itself an audit issue; and tiered, declining-rate or retroactive volume discounts break the right-to-invoice practical expedient that most SaaS finance teams were leaning on. Once that expedient is gone you are back to allocating across the contract at standalone selling price. There is a quieter one underneath: securities rules require material product and service revenue to be presented separately on the face of the income statement, which is a revenue category and chart-of-accounts decision made during configuration. Get it wrong at design and you re-map the ledger after go-live.
Which of our pricing constructs still qualify for the right-to-invoice expedient?
The window opens, the board wants to file, and the readiness answer everyone quotes is 18 to 24 months. A company that decides in the first quarter and wants to be public by the fourth does not have that.
Sequence the controls work against the first fiscal year end as a public company, not against the filing date, and decide early which systems work is deferred rather than compressed.
The obligations do not arrive together. Executive certification of financial statements and disclosure controls applies immediately on going public. Management's assessment of internal control over financial reporting applies from the first annual report, subject to the usual first-year transition relief. Auditor attestation can be deferred for emerging growth companies for up to five years. That staggering is an opportunity if you plan to it and a trap if you assume everything lands at once. The recurring finding in this sector is not exotic: reliance on system reports without adequate validation, which is the information-produced-by-entity problem in an auditor's own language. Named tech companies have carried material weaknesses across two consecutive year ends while remediating, with market capitalizations in the billions. This is a two-year problem when it goes wrong, not a two-quarter one.
Which reports feeding our close have documented completeness and accuracy validation?
Go-live is treated as the end of the program. The team disbands, the integrator leaves, and the platform keeps changing on a schedule nobody on the remaining team owns.
Staff the steady state before go-live, with a named owner for the twice-yearly release regression and for the standards landing in the next 24 months.
Workday ships two feature releases a year, in March and September, with a five-week sandbox preview window. That is two mandatory regression cycles annually, permanently, and they land on whoever is left. Meanwhile the accounting standards keep arriving: income tax disclosure requirements are already effective, expense disaggregation applies to annual periods beginning after 2026-12-15, and the internal-use software standard applies to annual periods beginning after 2027-12-15 with early adoption permitted. That last one retires the project-stage capitalization model, which never fit agile development in the first place. Each of these is a chart-of-accounts or dimensional question, not a footnote question, and each one is cheaper to design for than to retrofit.
Who owns release regression, by name, and what is their day job?
What those five mean for the chair you sit in
Each seat's sharpest exposure, the early sign it is live, and the one question worth asking this quarter.
The build-it-ourselves instinct
Your team can build this, and that is the problem: the constraint is not engineering, it is the accounting judgment, the statutory edge cases and the release treadmill afterward. A company operating one of the world's largest cloud platforms discontinued its company-wide HCM deployment, which is the cleanest available counter-argument. Then the permanent cost: two vendor releases a year with a five-week preview, forever, landing on whoever remains after the program closes.
The plan has a go-live date and no named owner for the March and September regression cycles.
What is our steady-state team, and does it exist before go-live or after it?
Three standards and a pricing model, all at once
Expense disaggregation applies to annual periods beginning after 2026-12-15, internal-use software after 2027-12-15, and income tax disclosure is already effective. Each is a chart-of-accounts decision. Layer on usage-based pricing, where prepaid credits are liabilities, breakage needs an estimate with no history, and tiered discounting breaks the practical expedient. And the tax split: domestic research expenses immediately, foreign research over 15 years, derivable from source transactions or not at all.
The domestic and foreign research split is produced by a quarterly analyst spreadsheet.
Does our chart of accounts carry the dimensions these three standards need?
Distributed hiring writes obligations you did not sign
Payroll runs native in five countries and through partners everywhere else, so every new country is an integration with its own registration, calendar and compliance tail. Trailing equity withholding can require keeping a foreign payroll registration open for people who no longer work there. Pay transparency thresholds now reach very small headcounts, with one state at five employees, so a 30-person company with one remote hire can be in scope. And remote work is being tested against permanent establishment thresholds, which turns a hiring decision into a tax position.
Hiring in a new country is approved before anyone has priced the payroll and registration tail.
What does the next new country cost us in payroll, registration and exit?
Four items already on the calendar for 2026 and 2027
Accounting standards and vendor release calendars do not negotiate with a program schedule. Each of these lands on the ledger or the platform while the program is still in flight.
Effective for annual periods beginning after 2024-12-15. The rate reconciliation disaggregation threshold is 5% of the amount at the statutory rate, which is 1.05% at a 21% US rate. Quoting the derived number is how you tell whether someone has built the schedule.
Annual periods beginning after that date. This is a chart-of-accounts and dimensional design problem, not a disclosure exercise, and it is far cheaper to configure for than to retrofit after go-live.
Early adoption is permitted. It retires the project-stage capitalization model, which never fit agile development, so the way engineering effort becomes an asset changes underneath every SaaS company at once.
Not a deadline, a treadmill. Two mandatory regression cycles a year, permanently, and they are the single most reliable thing missing from a post-go-live staffing plan.
Five questions worth more than a readiness assessment
Answerable from memory, scored on this page, nothing captured and nothing emailed.
These five are the start of the instrument. A full review also covers entity and intercompany design, the payroll partner map, revenue category structure against securities presentation rules, and access and change controls. Or skip the tooling and book the program review.
Building it yourselves, or buying it badly?
Pre-selection, mid-build, or remediating a material weakness with an auditor in the room. I sell no software and staff no builds, so the answer you get is the one I would give a friend. Tell me where the company is and I will tell you what I see.