Case Study: The Invoice Automation Decision

From pitch to pilot to reforecast

In the worked example, a vendor pitched "automate invoice processing and save $23,400 a month". Task decomposition and review time cut labour savings to $13,716, realisation to about $8,230, while error savings added $7,500 and costs subtracted $3,016. The point NPV looked excellent, but Monte Carlo showed a 17% chance of loss, driven mainly by realisation and volume. The team funded a pilot that measured 2.6 minutes saved per invoice (with a comparison group), agreed a no-backfill plan to improve realisation, and reforecast before scaling. All numbers are illustrative.

The decision trail

How the estimate evolved.

vendor pitch           $23,400/month saved (whole job automated)
task decomposition     $13,716/month labour time (2.69 min remain per invoice)
realisation 60%        $8,230/month realised labour value
+ error savings        +$7,500/month   - costs  -$3,016/month   = $12,714 net
point NPV (3 yr)       ~$343k, payback month 5
Monte Carlo            median ~$81k, 17% chance of loss
pilot (with control)   2.6 min saved/invoice -> reforecast; realisation plan agreed
decision               limited rollout, quarterly reforecast

Keep the trail

Documenting how the estimate changed builds credibility for the next business case.

Quick check: In the case study, what did the pilot measure with a comparison group?

• The discount rate
• The vendor's profit
• Minutes saved per invoice attributable to the AI tool
• The number of slides