Assignment cost projections and scenario modelling that finance will sign

How projections are sold today

Cost projections are a product, and they are metered. Mercer sells Total Cost Estimate reports, produced using the AssignmentPro tool under a licensing agreement with Equus; ECA runs a cost-estimate service on its own data; AIRINC sells an assignment calculator built on a 900+ location database. On the software side, Topia's cost-estimate product is sold in tiers by estimate volume, starting at up to 60 estimates per year. Read that pricing model as a market signal: projections are treated as a premium artifact to be rationed, which is precisely why programs run fewer scenarios than their decisions deserve.

The demand side wants the opposite. In KPMG's 2025 benchmarking, running complex calculations such as cost projections and payroll reconciliations is the second-ranked AI priority, named by 47% of organizations. Programs do not want one estimate per assignment; they want to interrogate the decision.

Why estimates diverge from actuals

An estimate is a static snapshot of assumptions: benchmark cost-of-living tables, standard policy parameters, a tax rate at a point in time. Actuals accumulate somewhere else entirely: home and host payroll, vendor invoices, FX at settlement dates, and tax gross-ups that only finalize after year-end. The two never meet in one model; 55% of organizations name data spread across multiple systems as their top analytics obstacle. So the projection is not wrong so much as unaccountable: when the final cost lands 40% over, nobody can decompose the miss into rate changes, policy exceptions, FX and estimation error. A projection that cannot be audited against its outcome teaches the organization nothing, and finance prices that in by trusting the next one less.

The corridor is the decision

Scenario modelling matters because the swing factors are corridor-shaped. AIRINC's route-level analysis shows allowances alone can be about half of total assignment cost, UK immigration fees run near USD 30,000 and London school fees approach USD 90,000 per year. The questions that actually precede an assignment (this candidate to London or Amsterdam; long-term assignment or permanent transfer; tier 1 or tier 2 policy; family or unaccompanied) are six-figure questions, and even the enterprise vendors' own content teaches scenario planning as the core discipline. The constraint has never been the concept. It is that each scenario costs a report, or a week of spreadsheet work.

What "finance will sign" means

A projection earns a signature when it stops being a number and becomes an argument. Four properties do the work. Your data, not generic assumptions: computed from the program's own ingested payroll, vendor, policy and benchmark feeds, so the baseline is what you actually pay, not what a survey average pays. A visible calculation trail: every line expands to its basis (which benchmark table and version, which policy rule, which tax methodology), the anti-black-box property. Versioned assumptions: when the projection is revisited in month 14, the original assumption set is reproducible, and variance decomposes into assumption changes versus estimation error. Unlimited comparison: scenarios side by side (host, tier, assignment type), cheap enough to run for every decision, because once the semantic model exists, a scenario is a deterministic computation rather than a purchased report.

This is how the Ask & Chart scenario workspace is built: side-by-side scenarios computed by the same version-stamped engine as the program's actuals, on joins a human confirmed, with provenance on every line. The projection and the reconciliation live in one model, which is the only arrangement under which projections improve over time (see the reconciliation problem).

Frequently asked questions