The AI Demand Forecasting Agent: How It Drafts Forecasts and Price Changes in Odoo and MYOB Acumatica

An AI demand forecasting agent reads sales history, open orders, lead times and margin from Odoo or MYOB Acumatica, drafts a forecast per product with a written reason, flags the SKUs the built-in statistics will get wrong, and proposes price changes. A planner approves every forecast and a pricing owner approves every price. As at September 2026 neither platform ships an AI forecast; both ship statistical ones.

At a glance

• Who it is for: wholesalers, importers and manufacturers on Odoo or MYOB Acumatica with a few hundred to a few thousand active SKUs and one person, usually the buyer or operations manager, who “knows the numbers” in their head.
• Problem it solves: forecasts that are last year’s figure plus a guess, seasonal SKUs ordered off a flat average, and price lists that lag landed cost by a quarter.
• Platforms: Odoo 19 Enterprise (Master Production Schedule, reordering rules, pricelists); MYOB Acumatica 2025.2 and 2026.1 (replenishment parameters, seasonality, sales price worksheets).
• Trigger: a scheduled run each Monday before the planning meeting, plus an on-demand run when a supplier price notice, FX move or large quote lands.
• Inputs read: outgoing stock moves by SKU and warehouse (24 months), confirmed and quoted orders, supplier lead times and minimum order quantities, last and average cost, current pricelist rules, and the platform’s own suggested forecast.
• What the agent drafts: a forecast review list (proposed quantity, the platform’s number, the difference and a one-line reason), a flagged-SKU list where the statistical model should not be trusted, and pending price changes with an effective date and the margin before and after.
• Approval gate: the planner applies or rejects each forecast line in the ERP; the pricing owner releases or rejects each price change. The agent has no write access to forecasts, pricelists or price worksheets.
• Audit trail: every proposal is stored with its inputs, the reason and who approved it, so a bad forecast can be traced to the data it came from.
• Typical cycle time: 15 to 30 minutes of planner review per week for a mid-sized catalogue, in place of a half-day spreadsheet exercise.
• What it never does: write a forecast or price into the ERP, publish a price to a website or marketplace, set or suggest a resale price for a reseller, or read competitor pricing that was not publicly available.
• Typical cost: a single scoped agent pilot typically starts from a few thousand dollars ex GST for discovery and sandbox build, per our AI agents for ERP in Australia page.
• Typical timeline: 4 to 10 weeks from discovery to a drafts-first go-live, per the same page.
• Last verified: 8 September 2026 against Odoo’s 19.0 Master Production Schedule documentation, Acumatica’s replenishment configuration guide and MYOB’s sales price worksheet article.

Every distributor we work with has a forecast. It lives in a spreadsheet the buyer built four years ago, it takes half a day a month to update, and its central assumption is that next month looks like the same month last year. It’s right often enough to survive and wrong often enough to explain the pallet of seasonal stock in aisle nine. The demand forecasting agent’s job is to make the forecast a weekly, explained, reviewed number instead of a monthly ritual.

What does an AI demand forecasting agent actually do?

An AI demand forecasting agent is software that reads demand and cost data from your ERP, produces a proposed forecast for each product and period with a reason attached, and hands the list to a planner who applies the numbers they agree with. It sits upstream of the AI replenishment agent, which turns an agreed forecast into draft purchase orders, and alongside the AI inventory exception agent, which finds where the last forecast was wrong.

The useful part is not the arithmetic. Odoo and MYOB Acumatica already do arithmetic well. The useful part is the reading: the agent notices that last September’s spike on one SKU was a single project order from one customer, that three open quotes for another SKU add up to twice its monthly average, that a supplier has emailed a price increase effective the first of next month, and that a product is 14 weeks old and has no history for any average to work from. A statistical model can’t read an email or a quote. A language model can, and it can write down why it changed the number, which is the thing a planner needs in order to trust or reject it in ten seconds.

The forecast then feeds two decisions: how much to buy, and what to charge. The agent drafts both. It approves neither.

What do Odoo and MYOB Acumatica already forecast without AI?

Both platforms ship a forecast, and it is worth knowing exactly what it is, because the agent’s first job is to check it.

Odoo 19’s Master Production Schedule (shipped, Enterprise) holds a forecasted demand figure per product per period, and its Suggest Forecasted Demand tool fills that figure from history. The choices are Actual Demand, Previous Year, Last 30 Days, Last 3 Months or Last 12 Months, each scaled by a percentage you type in, so 110% means “last year plus ten”. We have read the v19 Enterprise code behind it: the suggestion is completed outgoing moves for the matching period, prorated and multiplied by your factor. No trend, no seasonality index, no model. The schedule lives in the Manufacturing app; a distributor without it keeps its forecast in the minimum and maximum on each reordering rule, which is where the agent’s proposals land instead. Odoo’s reordering rules then act on forecasted stock, and pricelists hold price rules with a validity period. Odoo 19’s AI features (shipped) work on fields, documents and chat; none of them touch the forecast. Anything about forecasting in Odoo 20 is roadmap until Odoo Experience on 24 to 26 September 2026.

MYOB Acumatica, built on the Acumatica platform, goes further on the statistics. Its replenishment configuration (shipped) uses a moving average demand forecast model, a seasonality table you maintain per season, and a Calculate Replenishment Parameters process that works out average daily sales and average lead time and then sets safety stock, reorder point and maximum quantity per item and warehouse. A planner reviews and applies the calculated parameters on a separate screen, which is already a drafts-first pattern. Prices change through sales price worksheets: pending prices calculated from last cost plus a markup, held until someone releases the worksheet, effective from a date you set. MYOB’s AI Automation prompts (an early access preview, per MYOB) run inside forms and don’t produce forecasts. Globally, Acumatica 2026 R1 (shipped in the global product, March 2026) added in-transit stock to MRP and DRP planning, which sharpens the supply side of the forecast; MYOB Acumatica typically receives global features 6 to 9 months later. Our MYOB Acumatica AI post tracks the ANZ position.

Table: What Odoo 19 and MYOB Acumatica ship for forecasting and pricing today, and what a drafts-first AI demand forecasting agent adds on top.

If the built-in tools aren’t configured, start there. An agent checking a forecast that nobody maintains has nothing to check. We covered the groundwork in our guide to Odoo inventory management.

A worked example: the Monday morning forecast review

A Queensland wholesale distributor with about 1,800 active SKUs across two warehouses runs the agent at 6am each Monday. This is what the planner finds at 8:30.

The agent has pulled 24 months of outgoing moves per SKU and warehouse, the platform’s own suggested forecast for the next 13 weeks, open sales orders and quotes, and supplier lead times. It has produced three lists. The first is 41 SKUs where its proposed forecast differs from the platform’s suggestion by more than 20 per cent, each with one line of reasoning. One reads: “Previous-year September demand of 1,240 units includes a single 900-unit order from one customer whose project completed in November; proposed 380 units, in line with the trailing six-month average.” Another: “Three open quotes totalling 2,100 units expire within the period; the moving average is 640. Proposed 1,200 units, weighted by this customer’s 55 per cent quote conversion over two years.”

The second list is 26 SKUs the agent won’t forecast at all: nine products under 90 days old, eleven with fewer than six sales in the last year, and six where one customer accounts for over 70 per cent of demand. For each it proposes a proxy (a similar product’s curve) or a note to ask the account manager. The third list is pricing, which we’ll come to.

The planner opens the reordering rules or the Master Production Schedule in Odoo (or the Apply Replenishment Parameters screen in MYOB Acumatica), works down the first list, accepts 33 lines, changes four, rejects four, and types the agreed numbers in. The agent has no permission to do that step, deliberately. The supplier PO chasing agent picks up the resulting purchase orders later in the week.

A forecast nobody can explain is a forecast nobody will defend in the planning meeting. The agent’s real output isn’t the number, it’s the sentence next to the number. If it can’t say why, the planner shouldn’t accept it, and we build it so the planner has to click.

Josh Craig, Director, Auboros

Can an AI agent change our prices?

It can propose them. It should not change them, and in Australia there are three lines it must be built not to cross.

The third list from Monday’s run is pricing. The agent has read a supplier’s price notice from the inbox (a 6 per cent increase effective 1 October on one category), the FX exposure on open foreign-currency purchase orders, and the current pricelist rules. It proposes a pending price change for 212 SKUs in that category, effective 1 October, showing gross margin at the current price, at the proposed price, and at the price that would hold margin exactly. In Odoo that becomes a proposed pricelist rule with a validity start date; in MYOB Acumatica it becomes a sales price worksheet on hold. The pricing owner reviews the margin columns, trims the increase on the twelve SKUs that sit in the public web catalogue, and releases. The channel sync agent then checks that the website and marketplace prices moved with it.

The three lines. First, the ACCC’s guidance on setting prices (updated 15 July 2026) says dynamic pricing is legal, but businesses must be clear about the price customers will pay and must not mislead anyone about the reason for a change. An agent that drafts “prices are up because of freight” when the driver was margin has written a misleading claim on your behalf, so the reason the agent records is the reason the customer sees, or no reason at all. Second, prices must be set independently of competitors. Public web prices are fair game to read; a spreadsheet of a competitor’s trade prices that arrived from a mate at the competitor is not, and the agent’s inputs are restricted to your own data and public sources so that the question never arises. Third, if you sell through resellers, the ACCC’s rules on minimum resale prices (updated 12 August 2026) make it illegal to impose a minimum price on them. The agent may draft a recommended resale price list. It must never draft a “do not sell below” condition, a discount tied to holding price, or a supply hold on a discounter, and we write that exclusion into its instructions and test for it.

Excess stock is the other side of the pricing coin. Where the forecast shows a SKU won’t clear in twelve months, the agent proposes a clearance price and notes the effect on year-end stock valuation, because the ATO lets you value trading stock at cost, market selling value or replacement value, and obsolete stock can be valued lower with reasons you can support. That decision is your accountant’s, with the agent’s list as evidence.

Where does the demand forecasting agent go wrong?

• The confident trend. Three good months become a growth curve and a doubled order. Fix: the agent must name the driver of any increase above a threshold, and “trend” on its own is not a driver the planner accepts.
• The phantom quote. Open quotes are counted as demand and the customer never converts. Fix: weight quotes by that customer’s actual conversion history, show the weighting, and never count a quote older than its validity date.
• The stale cost. Prices are proposed off last cost when the next receipt is at a different cost. Fix: read confirmed purchase orders and supplier notices, not just the cost on the product record.
• The permission drift. Someone gives the agent write access to “save the planner clicking”. Fix: role-scoped credentials with no write access to forecasts or prices, checked at the monthly permission review, per our AI agent governance post.
• The reason nobody reads. The review list grows to 300 lines and the planner accepts all. Fix: cap the list, sort by dollar impact, and treat “accept all” in the log as a signal to retune thresholds, not as success.

How to start

Run the agent read-only against a copy of your database for one quarter, with no ERP changes at all, and score its proposals against what actually sold. That gives you a measured error rate for the agent, the platform’s own suggestion and the buyer’s spreadsheet, side by side. Only then move to drafts-first in production, forecasts first and pricing a quarter later, because a wrong forecast costs a stock adjustment and a wrong price costs a customer. 78 per cent of wholesale distribution businesses plan to adopt or expand AI in their ERP, according to MYOB’s ERP trends report; the ones that do it well start by measuring the forecast they already have.

Goes well with

• AI agents for ERP in Australia: the full set of drafts-first agent patterns and the approval gate for each, if your next question is “which agent first?”
• The AI replenishment agent: how an agreed forecast becomes draft purchase orders a buyer approves, if your next question is “what happens after the forecast?”
• Odoo inventory management for Australian businesses: the reordering rules and forecasted report the agent checks, if your next question is “what do we need set up first?”

Thinking about a demand forecasting agent for your Odoo or MYOB Acumatica business?

We build drafts-first forecasting and pricing agents for Queensland and Australian wholesalers and manufacturers, from Brisbane, and we start by measuring the forecast you already have before we change anything. If you want to know how far your current forecast is from what actually sells, book a free consultation. We’ll score last year’s forecast against last year’s sales with you, which is usually enough to decide whether an agent is worth building.