Buyer guide

Building a TCO that is not fiction

Most ERP business cases compare one vendor's licence to another's. That comparison omits roughly 38% of the actual three-year cost, and it omits the same proportion from both sides — which feels fair and quietly changes the ranking.

Want this modelled properly?

Send your candidate quotes and your requirements. We will build the three-year comparison.

1 / 3
38% of cost is not licenceThree years, not oneInternal time is the biggest omission

Buyer guide

Eight lines a real TCO contains.

A vendor proposal typically contains two. The other six are where business cases are wrong, and they are all estimable in advance.

Subscription over three years

Not year one. Model expected headcount growth, module additions, and annual uplift. Year three commonly runs 30–50% above year one for the same business.

Implementation

Median 1.4× first-year licence, plus a median +27% variance to original quote. Budget the variance rather than hoping to be the exception.

Integration

Per connected system, plus ongoing maintenance. Count your systems honestly — most companies name five and run eleven.

Your team’s time

400–1,200 hours at loaded cost. Usually the largest single line and present in no vendor proposal, because no vendor can invoice for it.

Parallel running

Your existing licence continues during the parallel period, which for a proper migration is several months. Frequently forgotten entirely.

Training and productivity dip

Direct training cost plus the weeks where your finance team is slower than before. Real, awkward to estimate, and better estimated badly than omitted.

Ongoing internal administration

Somebody maintains the configuration, the permissions, and the reports. That is a fraction of a role, permanently.

Cost of leaving

Export capability, data portability, and what a future migration would cost. Rarely modelled and directly affected by which product you choose.

Why three years and not one

Vendor quotes are built around year one because that is when the discount lives. Year two has no implementation cost and a full year of subscription; year three adds seat growth and uplift.

The ranking between candidates frequently changes between a one-year and a three-year view, particularly between per-seat and resource-based pricing models. A product that looks expensive in year one because implementation is included can be materially cheaper by year three.

The discount lives in year one. The decision lives in year three. Quotes are built around the first and evaluated as though they described the second.

Five errors that make a TCO useless

  • Omitting internal time from both sides. It feels neutral and is not — implementations differ in how much of your team they consume, and a shorter one is genuinely cheaper in a way licence comparison cannot show.
  • Using list price for one vendor and a negotiated price for another. Get comparable quotes or use list for both.
  • Ignoring the parallel period. Running two systems for three months is a real cost, and a vendor recommending a shorter parallel period is not saving you money.
  • Counting savings that require headcount reduction you will not make. Most finance teams redeploy recovered hours into analysis rather than reducing headcount. Model it as capacity, not as savings, unless you genuinely intend otherwise.
  • Excluding the do-nothing option. What does staying cost over three years, including the manual hours? Sometimes it is the cheapest option and the exercise should be able to show that.

The do-nothing baseline

The most useful column in a TCO is the one for changing nothing. Current software licences plus the people doing work software could do — that second figure is typically four to ten times the first.

Including it does two things. It shows whether any option is actually better than the status quo, and it reframes the decision from a purchase to a reallocation. The question stops being whether the new system costs more than the old one and becomes whether software plus people goes down.

Where the recovered hours go

Be honest in the model about this. Automation returns hours; it does not automatically return money. If your team redeploys those hours into analysis, forecasting, and business partnering — which is what usually happens and is usually the right call — the benefit is capacity rather than cost reduction.

A business case built on headcount reduction that nobody intends to make is a business case that will not survive its first review. Model capacity, say so, and let the decision-maker weigh it.

Our calculator publishes its assumptions

Every figure behind our TCO tool is stated on the same page — seat costs, implementation multiples, entity fees, and the $96,000 loaded cost per finance person. A calculator that hides its maths is a lead form with a slider. Where our assumptions do not match your situation, change them; where they favour us, say so and we will look at it.

Questions

Common follow-ups.

What proportion of TCO is not licence?
About 38% across our quote sample, and that excludes internal time. With internal time included, licence is frequently under half the total.
How do we estimate internal hours?
400 for a light single-entity implementation with clean data, up to 1,200 for multi-entity with data problems. Weak, and better than omitting it.
Should we model headcount reduction?
Only if you genuinely intend it. Most teams redeploy recovered hours into analysis, which is capacity rather than savings, and a business case built on cuts nobody will make does not survive review.
Why include a do-nothing column?
Because sometimes it wins, and because it reframes the decision from a purchase to a reallocation. Current licences plus manual hours is usually a large number.
Does the ranking really change over three years?
Frequently, especially between per-seat and resource-based pricing. A product that looks expensive in year one can be cheaper by year three.

Model three years and the do-nothing column.

The ranking changes, and sometimes staying where you are turns out to be the cheapest option.