Today's Feature

Diesel has passed $5 a gallon and chip shipments are jittery after the Iran conflict — and boards are not asking for prettier slides so much as for dollars and contingencies. At the same time capital is pouring into AI infrastructure: C2FO this week reported it has funded more than $500 billion in working capital, BlockchAIn signed a non‑binding LOI worth roughly $400 million for AI infrastructure, and market forecasts put the embedded‑AI market at about $42.3 billion by 2033. The result is a C‑suite that demands immediate, modelled answers — and too often gets them from dashboards it cannot audit. Two concrete currents make this a crisis for FP&A. First, reporting is moving from month‑end spreadsheets to live Power BI canvases: decomposition trees, scenario toggles and real‑time pipelines tied to ERP, CRM and procurement feed single‑pane executive views. Social posts on X show analysts publishing executive dashboards built in hours — even a pharmacy operations dashboard analysing 10,000+ prescriptions has circulated as a template — which explains the speed but not the trust. Second, shocks are arriving faster and with bigger price tags: diesel above $5/gal, GPU and chip tightness, and the impending surge of semaglutide generics into India after the patent expiry on March 20, 2026 — more than 50 branded entrants are expected, with clinical data showing roughly 15–17% weight loss over 68 weeks and meaningful metabolic gains in over 80% of users. Those are margin threats and demand signals at once. That combination elevates three tasks for FP&A. The first is disciplined decomposition: every executive KPI must show its source, refresh cadence and dollar sensitivity. The second is scenario engineering for a short list of needle‑movers — fuel, chips and key suppliers — with toggles that quantify P&L, cash and margin outcomes. The third is auditability: AI that flags anomalies must carry an explainable trail. Opting for speed without provenance hands executives actions that are fast and fragile; a briefing that says “trust the model” is not an answer. This is also a vendor‑management problem. Contractual fragmentation — too many suppliers, inconsistent SLAs and no single system of record — turns supply shocks into bargaining losses. AI can surface supplier risk and accelerate discovery, but the closing move remains human: renegotiation, hedging inventory and SKU prioritization. FP&A people must therefore be bilingual — fluent in data engineering and fluent in negotiation — and able to translate model outputs into concrete savings or costs. Podcast and industry commentary from recent tech forums underline the point: vendors and platforms are building “AI factories” and new model primitives (attention‑residual ideas and block attention in LLMs) that will change what analytics can do, but they do not replace contract tacticians. There is an upside. The same dashboards that expose cost vulnerability can also spot demand pivots. The flood of cheaper semaglutide, for example, is a revenue signal for wellness‑positioned CPG SKUs; analytics should be used offensively to reallocate promotion budgets, reweight assortment and capture migrating margin pools. And the market’s money — large AI infrastructure commitments and rising embedded‑AI forecasts — means tools to do this more quickly are available; the problem is governance, not capability. Practical next moves are few and specific. Demand provenance tags on every executive KPI; instrument scenario toggles for fuel, chips and top suppliers; require a one‑slide human‑verification protocol paired with every AI alert; model semaglutide’s revenue and margin impact now, not in six months; and capture models, assumptions and post‑mortems in a searchable “second brain.” FP&A can either remain a factory for pretty charts or become the nerve centre that converts transient data into durable judgment. In this market, being fast is table stakes — being right fast is the product.