Spend is visible. Utilization and value are not.
Finance can usually produce the total. What it cannot easily produce is the denominator: how many purchased seats are genuinely active, which departments drove a consumption spike, whether two platforms cover the same people, and what any of it has returned. Without that, AI budget conversations reduce to two unsatisfying positions — cut something visible, or approve everything and hope.
The difficulty is structural rather than a reporting failure. Seat licenses and metered consumption are billed on different logic, vendors report only their own footprint, and value is claimed in slide decks that are never reconciled with usage.
What finance actually has to decide
- Renew, resize, or exit
- Whether a contract renews at the same seat count, a smaller one, or not at all — and the utilization evidence that supports the position going into negotiation.
- Fund the next expansion
- Whether the next AI request is incremental value or duplicate capability already licensed elsewhere in the organization.
- Set the budget envelope
- How much of AI spend is fixed licensing versus variable consumption, and how much variance the plan can absorb.
- Choose where value gets measured
- Which initiatives carry an explicit value case with documented assumptions, and which are simply cost.
- Decide what the board hears
- One AI investment narrative that survives scrutiny, with sources shown and estimates labeled as estimates.
Four systems, none of which sees the whole
Vendor consoles report their own product and nothing else. Finance systems record the invoice but not who used the seat. Procurement holds the contract terms but not the utilization behind them. Business units hold the outcome stories but not the cost. Every AI review begins by rebuilding the same reconciliation by hand.
Midgentic's role is to hold those pieces together continuously: contracted terms with effective dates, automated adoption where a connector exists, manual and CSV data where one does not, and value assumptions recorded next to the spend they justify.
The finance-relevant AI metrics
- Total AI investment split into fixed licensing and variable consumption
- Seat utilization: contracted seats against genuinely active users, per platform
- Cost per active user, in the currency you are billed in
- Dormant seat exposure and the annualized value of reclaiming it
- Overlap: platforms covering the same people with similar capability
- Consumption variance by period, with the model or project that drove it
- Documented value assumptions, kept separate from measured outcomes
- Spend that entered the portfolio outside central procurement
From line item to investment case
License terms are entered once per product or SKU, with seats, price, currency, and effective dates, so historical periods stay accurate after a contract change. Adoption arrives automatically where a connector exists and manually where it does not. Metered platforms report real consumption cost. The portfolio engine then computes utilization and cost per active user across providers without flattening currency or double-counting a vendor billed two ways.
Value is handled conservatively on purpose. Estimated value is only ever as strong as the assumption behind it, so assumptions are explicit, editable, and attributed; measured outcomes are recorded separately. Nothing in Midgentic manufactures a savings figure on your behalf. See enterprise AI ROI for how that model works, and AI spend intelligence for the cost side.
AI bought outside the process
Some AI spend never passes central procurement: a team subscription on a card, an AI add-on enabled inside an existing platform, a pilot that quietly became production. Finance usually discovers it at renewal. Shadow AI explains how that gap forms and how bringing those tools into the portfolio — manually, honestly labeled — makes the total real rather than convenient.
Common questions
Is Midgentic an expense management or procurement tool?
No. It does not process invoices, manage vendors, or run approval workflows. It is an intelligence layer: it holds your AI license terms and vendor-reported consumption alongside adoption data, so spend can be judged against usage and value. It complements finance systems rather than replacing them.
Where does AI cost data come from?
Two sources, always labeled. Metered platforms such as the OpenAI and Anthropic API report actual consumption cost automatically. Seat-licensed products use license terms you enter — seats, price, currency, and effective dates — because vendor reporting APIs generally do not expose negotiated pricing.
Can Midgentic prove AI ROI?
It can build a defensible ROI position, not a guaranteed number. Investment is factual: contracted and metered cost. Value is separated into documented assumptions, which you set and can defend, and measured outcomes recorded against initiatives. Midgentic never blends the two or invents benchmark savings.
How does it handle multiple currencies?
License terms are stored with their own currency and effective dates, so historical periods stay accurate when a contract changes, and portfolio rollups are computed without silently flattening currency.
What about AI spend finance does not know about?
Departmental subscriptions and tools bought outside central procurement can be added manually or by CSV so they appear in the portfolio. Midgentic surfaces the gap between what is approved and what is actually in use; it does not scan networks or endpoints.