What if tomorrow you had to design a finance function with a swarm of intelligent agents at your disposal, how would you redesign the work? Would it end up being a pile of spreadsheets anyway?
This question, sparked by a recent FP&A Gen-AI hackathon, led me to uncover four stubborn seams in current processes along with the reality checks that keep them stubbornly human-heavy.
1️⃣ Self-Serve Reporting → Insight-on-Tap
FP&A pros still burn 45 % of their time collecting/validating data and just 35 % driving insight, basically unchanged despite years of BI roll-outs. (FP&A 2024 Trends Survey). That isn’t just wasted time, it’s evidence of output thinking: ship a static deck, pray someone reads it. Most teams don’t even track what gets opened or consumed.
Challenge: Most reporting workflows are built backwards, from tool to template to output. Design for the reader first. Capture who consumes what, ask “competency questions” for each role and moment, then leverage Ai to tailor the feed beyond a dashboard. Density of artefacts ≠ density of insight, insight comes from relevance, context, and timing.
2️⃣ Scenario-Ready Forecasting → Models Anyone Can Poke
92 % of CFOs admit forecasting accuracy is still a headache. (PwC) The blocker isn’t math, it’s access. Scenario work is usually an Excel bunker exercise that takes days of lead time and slow feedback cycles.
Challenge: Forecasting often lives in isolated models, owned by a few and understood by even fewer. Reality is that is not always about the math but the ergonomics. Giving business owners a “slider board” of internal drivers (price, churn, hiring pace, etc), showing live P&L ripples, and surfacing the why behind each swing, is what drops the black box mystique and creates a learning experience. Lowering that barrier will add tremendous value as more and more sophisticated data feeds come in.
3️⃣ Reconciliation & Revenue Assurance → Plug the Everyday Leak
SaaS research shows 3 – 5 % of revenue quietly slips through data mismatches and manual billing errors (m3ter). Every new SaaS module adds another seam; perfect unification is a myth.
Challenge: Reconciliation is not broken because people are careless (at least not always haha), it’s broken because every new system adds another seam and no one owns the in-between. Treating AI as a reconciliation concierge: stitching CRM ↔ CPQ ↔ ERP, flagging anomalies early, nudging the right owner and speaking the right business language can reclaim some of that revenue and reduce friction. I think about it less as pure automation, but orchestration for the messy middle.
4️⃣ Compliance Intelligence → Know If It Even Applies
Highly regulated enterprises can spend up to 6 – 10 % of revenue on compliance overhead. (Bloomberg) Yet the first, and most common step is still: Does this rule even apply to us?
Challenge: Domain specific GPTs could technical help spotlighting clauses, exposure, and draft action plans, however true productivity will depend on walking users through the right inputs. The real lift is guiding teams to contextualize rules with the business (entity, scope, precedent), then surfacing risk in plain language with a transparent trail. That’s how AI shifts from PDF spelunker to compliance co-pilot.
🔄 Take-away
The real unlock isn’t bolting AI onto yesterday’s workflows, it’s re-thinking finance from first principle product thinking:
- systems that measure what questions users actually ask and learn what they need,
- models anyone can interrogate and play with,
- orchestration that speaks business language vs tool language,
- and feedback loops tight enough to iterate every cycle.
Those possibilities simply didn’t exist for FP&A a few years ago. They do now, if we’re willing to design for a world where Ai and agents are the default teammates.
