Beyond the Hype: Where AI Agents Belong in Your IBP Cycle (and Where They Don’t)

Meet Sana, the portfolio manager.

By the time Sana opens her laptop, an agent has re-segmented the range by velocity and variability (ABC/XYZ), flagged three products whose margin slipped below threshold, and caught two bill-of-materials errors that would have overstated a launch’s cost. Nobody asked it to. That is the point.

For the planned launch, a reasoning agent built on a time-series foundation model matched the new product against past launches and estimated how much it would cannibalise the existing range. It returned a range with its assumptions, not a verdict.

Then comes the real question: should the company exit a product that a strategic customer buys at a thin margin? The agent can tell Sana what it costs. It cannot tell her what the relationship is worth.

Meet Priya, the demand planner, and Rahul, Head of Sales.

Priya used to spend Tuesdays cleaning point-of-sale files. Now a sensing layer ingests daily POS scans, distributor stock, weather and logistics signals, and adjusts the short-term picture on its own. Baselines for fast-moving SKUs come from gradient-boosted models that nobody touches.

Priya’s time goes to a reasoning agent instead. It tells her the northern promotion uplift is overstated and that one region has forecast high for six months running, and it supplies P10, P50 and P90 ranges to size safety stock.

Then Rahul walks into the consensus meeting. His joint business plan with the largest retailer commits the company to a volume the statistical model considers optimistic. That is a negotiation between a customer relationship and the company’s appetite for risk. The agent can show the range. Rahul and the leadership team decide what to promise.

Meet Marcus, the supply planner, and Deepa, the plant manager.

Marcus inherits a netted demand plan. Inside the time fences, execution agents net the forecast, run lot-for-lot replenishment for standard components and flag a bottleneck on Line 3 the moment it appears. Time fences set the agent’s leash: it executes inside the frozen zone’s rules and proposes, never imposes, in the liquid zone.

A reasoning agent then runs rough-cut capacity across the bottleneck work centres, trading off labour, machine hours and supplier limits, and offers three ways to smooth production. Marcus picks one in an hour rather than building it over a weekend.

What the agent cannot do is sit with Deepa and the works council about a shift change, decide whether a second filling line deserves capital, or renegotiate a key supplier’s contract. Those turn on trust, law and long memory.

Meet Elena, the finance business partner.

Elena’s two-day ritual of turning tons and cases into revenue, COGS, logistics cost, gross margin and EBIT now takes minutes. A reasoning agent then splits the variance to plan into price, volume and mix, and answers questions like “what if raw materials rise 10% and the port is shut for five days?” with ranked options scored by EBITDA.

But which option do we take? Give up margin to protect service to a key customer, or the reverse? Commercial, Supply Chain and Finance each have a different incentive, and an agent cannot broker that. Elena and Anita bring a gap-closure proposal to the executives, and people argue it out.

Meet Vikram, the COO and executive sponsor.

The deck is assembled and the 30-day KPI scorecard is ready before Vikram sits down. In the room, a co-pilot connected to the digital twin answers “what if we move 15% of volume to the second plant?” in seconds, instead of the traditional “we’ll come back to you next month”.

This is what makes the well-known 30-40-30 rule achievable: 30% of the meeting on backward-looking KPIs, 40% on forward trade-offs and gap resolution, 30% on long-term strategy. Without agents, the first 30% usually swallows the rest.

The decisions stay with Vikram and his peers: approving the plan, releasing unbudgeted capital, settling the trade-offs between functions.