Decide: Turning Forecasts Into Fulfilled Orders

Most planning functions run on a cadence that has nothing to do with how fast the business actually moves. Demand forecasts are refreshed monthly. Safety stock levels are set once a quarter and rarely revisited. Staffing plans are built in a spreadsheet two weeks before peak and then defended, rather than adjusted, when reality diverges from the plan.

Meanwhile, the business itself changes hourly: a promotion outperforms forecast, a supplier confirms a delayed container, a competitor drops a price, a heatwave pulls forward demand for one category and kills another. The planning cadence and the operating cadence are mismatched — and the business absorbs the difference as excess inventory, missed sales, or expensive last-minute labor.

Planning is not a document produced once a month. It is a decision — inventory, staffing, budget — that has to be right for the conditions on the day it is executed, not the conditions three weeks earlier when it was written.

This is the second post in our Agentic Operations series. Our first post introduced the Decide → Do → Deliver model. This one goes deep on Decide — the layer responsible for turning forecasts and constraints into a continuously current operating plan.

The planning problem: right answer, wrong moment

Traditional planning tools aren't wrong so much as they're slow relative to the business. Demand planning software forecasts well against historical patterns; the gap opens when conditions shift between refresh cycles. Inventory allocation rules are usually static — a fixed replenishment point, a fixed safety-stock percentage — set at a point in time and rarely revisited until something breaks.

• CFOs — working capital sits in inventory that's in the wrong place, or the wrong quantity, relative to where demand actually lands. It shows up in carrying cost and markdown, not as a single visible line item.
• COOs — service levels degrade unevenly: some nodes stock out on a fast-moving SKU while others sit overstocked on the same item a few hundred miles away.
• Workforce leaders — staffing plans built weeks in advance turn into expensive overtime or idle labor the moment order volume diverges from the average they were built against.

From periodic planning to continuous decisioning

The shift here is structural, not incremental: forecasts re-evaluated continuously against live demand signals instead of monthly or quarterly; inventory positioning rebalanced as conditions change instead of a safety-stock number set once and left alone; staffing and budget plans adjusted to the specific period ahead instead of built against a historical average; scenarios simulated on demand, before a decision is made, instead of a manual exercise reserved for major events.

The goal isn't to forecast further into the future with more precision. It's to shorten the distance between a changing signal and an updated plan — so the plan in use is always the best available one, not the most recently scheduled one.

Two digital workers, one continuously updated view

Datanomous Planning is built from two purpose-built digital workers that share one continuously updated view of demand, inventory, and capacity — part of the wider Operations Brain described in our first post.

• Operations Planner — forecasts demand at the granularity operations actually run on, by SKU and by location, not just category or network level. Plans staffing and budget against that forecast, and simulates a promotion, a weather event, or a new node opening before it happens.
• Inventory Planner — positions stock across locations ahead of demand rather than reacting once it arrives, and continuously rebalances as conditions shift, including rotating date-sensitive or regulated stock so older inventory moves first.

Same decision engine, two timescales

Operations Planner and Inventory Planner run a single decision logic at two different speeds — not two separate systems working in parallel. Before an order exists, they run continuously in the background: forecasting demand, planning staffing, and positioning stock ahead of it. The moment a real order arrives, that same continuous work becomes the input to an instant, five-part decision — sourcing, dispatch, labor, timing, and cost — evaluated together for that specific order.

Before it happens, the system continuously simulates the order scenarios a product could plausibly face over the coming days — different regions, different days, different demand intensities — and decides whether it's best placed in a store or a warehouse, and on which shelf. That decision refreshes itself continuously as demand shifts.

When an actual order arrives, five agents run the same calculation again, this time for a real order:

• Sourcing agent — checks where the product currently sits in stock, and if supply is short, finds the fastest supplier or upstream warehouse to replenish from.
• Dispatch agent — determines which location the order should ship from and by what method. This is the fulfilment decision Datanomous WES then commits and executes on the floor.
• Labor agent — determines which shift, which team, and how much capacity will prepare the order.
• Timing agent — calculates when the product will be ready and whether the delivery promise made to the customer can be kept.
• Cost agent — works out the full cost of that combination to the company, including labor, freight, packaging, and opportunity cost.

These agents don't run one after another — they run at the same time, targeting roughly 400 milliseconds from order received to a confirmed plan: fast enough to sit inside the order's own flow, not after it. The result is a single plan — the product ships from here, this team prepares it, it's ready by this time, and it costs the company this much. Major decisions are still approved by a person; this routine combination comes together in a fraction of a second.

The question was never just whether there's stock. It's this: what's the way to fulfill this order at the right time, with the right team, at the lowest cost?

That's a flawless plan on paper. Things on the floor rarely go exactly to plan — a picker doesn't show up, a truck is delayed, a product isn't where it's supposed to be. Turning this plan into reality on the floor, and instantly managing the moments it deviates, is the job of the Do layer: Datanomous WES.

A peak-season example

Consider a fulfillment operator heading into a promotional peak. The traditional approach: a staffing plan built four to six weeks out from a historical average, submitted for budget approval, and largely locked in. If actual volume comes in 20% above plan, the response is emergency overtime and agency labor, both at a premium and both arranged under time pressure. If volume comes in below plan, the business has already paid for labor it didn't need.

With continuous decisioning, the same event plays out differently: Operations Planner simulates the promotion against three demand scenarios before it launches; as early signals — site traffic, early order pacing — diverge from the base case, the forecast and staffing plan update automatically; Inventory Planner repositions stock toward the nodes where the surge is concentrated, ahead of the order, not after a stockout; and recommended staffing adjustments are surfaced to the operations leader for approval, with the cost and service trade-off shown alongside the recommendation.

The plan a business executes on peak day is the plan built for peak day — not the plan built five weeks earlier and defended ever since. For operators running more than one facility, this same logic extends across the network: one live view of which facility is on pace to hit its numbers today, and where it makes sense to shift capacity or inventory before a gap opens, rather than after a customer notices.

The agents behind the decision

Datanomous Planning isn't one model making one forecast. It's a small set of specialized digital workers, each with a narrowly scoped read on the operation and a concrete, individually attributable output.

• Demand & Signal Agent — watches historical sales, upcoming marketing campaigns, regional weather, and competitor pricing. When a region's weather is about to turn and a winter-apparel campaign is about to launch, it raises the regional demand forecast for the affected SKUs ahead of the order landing, not after.
• Inventory Placement Agent — watches warehouse fill levels, automation cost, freight lead time, and carrying cost. Given a demand signal, it decides what share of incoming stock ships to which node, before the order exists.
• Channel & Margin Allocation Agent — watches product margin, channel contracts, and stockout risk. When stock on a high-margin SKU runs low, it closes it to high-commission third-party marketplaces and reserves the remaining units for the company's own channels.

Two more digital workers exist specifically to catch what drifts after the plan is live:

• Re-balancing & Transfer Agent — spots that one facility is sitting on stock that hasn't moved in weeks while another is losing sales to a stockout of the same SKU, and drafts the inter-facility transfer order for a human to approve.
• PO Adjustment Agent — when sales run slower than forecast, proposes reducing the quantity or pushing the delivery date on an outstanding purchase order, and routes the recommendation to procurement rather than acting on it unilaterally.

None of these five agents work without the readiness standard we covered in our first post: one master product ID. An Inventory Placement Agent that sees “AYK-BLK-42” in one warehouse and “Black Sport 42” in another isn't making one good decision — it's making two disconnected bad ones, on data that only looks like it agrees with itself. That's why Datanomous builds the shared master ID and the live, event-driven data feed as the first step of onboarding, not a phase-2 to schedule once the agents are already live.

Why this matters to leadership

• CFO — continuous inventory rebalancing and per-order, margin-aware fulfillment decisions are two of the more direct levers for improving working-capital efficiency and protecting margin, without changing supplier terms or service commitments.
• COO — a plan that updates itself against live conditions is what makes it possible to hold service levels steady through demand volatility, rather than choosing between stockouts and excess buffer stock.
• CIO — Datanomous Planning deploys as a decision layer on top of existing forecasting, ERP, and WMS systems. It doesn't require replacing them, and every automated action is confirmed against policy, logged for audit, and reversible.

A correct plan still has to be carried out on the floor. Our next post covers Do — how Datanomous WES orchestrates picking, packing, and labor in real time as the shift unfolds.