Jeff Brokaw

AI Can Find Demand. It Cannot Create Capacity.

Jeff Brokaw · CMO and Certified Chief AI Officer ·

Before I spend another dollar finding buyers, I want to know when we can deliver what we're selling.

At a U.S. specialty manufacturer, my work covered demand generation and the case for expanded domestic production. The manufacturer engagement required a commercial plan that accounted for the company's ability to serve the buyers we pursued.

That is the check I want before an AI recommendation reaches the ad account. Who will quote the work? When can we ship it? What happens to the customers already waiting?

Can AI recommend a marketing action the business can't deliver?

Yes. An AI marketing recommendation can generate demand the business cannot fulfill. A system working from campaign performance cannot reliably check sales availability, inventory, production schedules or service hours unless it receives those inputs and uses them as constraints. Before approving spend, a named human owner should confirm the capacity, delivery promise and stop conditions.

This is the capacity check in my AI Marketing Operating Model series. The recommendation needs an answer from the people who will have to do the extra work.

AI can help with forecasting, scheduling and administrative work. That can free time or improve the use of existing resources. Count that improvement after someone verifies it. A proposed automation saving is a poor reason to promise an earlier delivery date today.

The problem is easier to see in an ad account. Google's Maximize conversions documentation describes bidding for the most conversions while spending the campaign budget. If submitted forms are the campaign's optimization goal, that is the event you have told the system to pursue. The setting does not establish whether your team can quote, sell or deliver the resulting work.

Give the recommendation a capacity check before it becomes a budget change.

Where the extra work actually lands

Capacity means the work the business can take on in a defined period while meeting its existing commitments. A headcount, a stock total or an annual production figure is too broad to approve the next campaign.

Sales: time to handle the opportunity

A lead creates work. Somebody has to check the fit, answer questions, prepare a quote and follow up. A complicated sale may also need an applications engineer or a specialist who is already supporting other deals.

I would look at the active queue and the age of opportunities waiting for a response. Then I would ask the sales leader how much additional work the team can handle during the campaign period. A rep count tells me less than a clear account of their remaining time and the work that already has a claim on it.

If AI removes some of that work, measure what changed. Faster email drafting does not automatically create more time for technical quotes or customer calls.

Fulfillment: what can be promised, where and when

The number in an inventory report may include stock already committed to another customer. For a manufacturer, the limiting resource may be a particular process, a qualified operator or a component with a long lead time. For an installer, it may be the calendar of a crew that serves one region.

Microsoft's order-promising documentation, updated April 21, 2026, makes a useful distinction. Available-to-promise checks material availability. Capable-to-promise also considers production capacity. An item being available does not settle when the business can produce and deliver the customer's order.

The backlog deserves its own check. In its October 2, 2026 manufacturing report, the U.S. Census Bureau reported August unfilled orders of $1,609.6 billion, up 0.6 percent from July. That national total does not establish whether your plant is overloaded or a delivery is late. Your own scheduled work and available resources must control the promise.

For ecommerce, check the exact product variant and delivery region. Google's Merchant Center availability rules require consistent availability in the product data, landing page and checkout. A campaign should not advertise stock that disappears when the buyer selects a size or enters an address.

Service: the work after the sale

The first purchase can start a larger commitment. Onboarding, implementation, training and support all need time. Selling an account that requires extensive setup can consume capacity well beyond the sales call.

I would ask the service leader which work this offer creates and whether the team can absorb it alongside current customers. Count overdue implementations and unresolved customer issues before assuming a new account can start immediately.

A profitable sale can still be badly timed if its delivery would require breaking an existing customer promise.

Cash: paying for the work before the customer pays

An order can require materials, labor or an outside supplier before the payment arrives. More demand can increase that cash requirement even when the expected margin looks good.

Finance should check payment terms, deposits and the cash needed to fulfill the proposed work. My Five Systems, One Customer article explains why a form submission, a booked deal, recognized revenue and collected cash need separate definitions. A campaign report cannot substitute one for another.

Turn capacity into a rule that can stop the spend

“Check with Operations” is too vague. It leaves the marketer waiting for a reply and the software with nothing to enforce.

For each campaign, identify the work it could create and the resource that limits it. Use the same product, region and period as the offer. Subtract existing commitments and an agreed reserve before calling the remainder available.

Keep these facts beside the recommendation:

  • The additional work expected, with the assumptions behind that estimate.
  • The remaining capacity for the relevant product, team, location and period.
  • The source, last update and person responsible for confirming that capacity.
  • The limit on the offer or spend, and what happens when the limit is reached.
  • The evidence and authority needed to restart after a pause.

Marketing's demand estimate is uncertain. Allow for more response than the forecast expects and for disruption to delivery. The owner of the constrained resource should agree to that allowance. Do not let the model invent the amount of spare capacity it needs to make its recommendation look acceptable.

There are also two different kinds of control. An inventory reservation or a booking limit can block another commitment. A budget limit only controls media spending. It does not guarantee a particular number of orders or prevent customers arriving through other channels. Use the controls that can enforce the actual constraint.

Check the combined load, too. Two campaigns can each appear to fit the same available installation slots. Approving them separately does not make those slots available twice. Sales, paid media and lifecycle campaigns need to draw against the same view of committed work.

Hold a spend increase for review when the capacity report is too old or unavailable. Someone needs to confirm the limit before the workflow proceeds.

For automatic actions, enforce the boundary in the connected application or workflow. A sentence in a prompt is not enough. The person responsible needs a working pause control and a record of what was already changed or queued.

When the business is full, change the offer

A capacity limit does not mean marketing has to disappear. It means the campaign needs to respect what the business can deliver.

The team may be able to shift demand to an available product or region, offer a later delivery date, or collect interest for a future opening. Operations should confirm the alternative, and the customer should see the timing before committing. An honest waitlist can be useful. It is not a completed sale.

Sometimes the extra demand is valuable enough to justify adding capacity. That is a decision about hiring, equipment, outsourcing or a different way of delivering the work. The people responsible should approve the cost and timing. Marketing can help make the case; it cannot treat the proposed expansion as capacity already available.

Microsoft's finite-capacity planning documentation describes moving a production job and its delivery date when the necessary resources are unavailable. That is the kind of constraint I want reflected in the offer. Buying more attention does not remove it.

Name the person who can approve the extra work

The marketing owner is responsible for the campaign. The sales leader confirms follow-up capacity. Operations or fulfillment confirms delivery. Service confirms onboarding and support. Finance reviews the funding requirement when the campaign creates a material cash commitment.

That does not require a committee to approve every ad. Give routine work a standing limit. Bring the relevant owner into the decision when a recommendation approaches that limit, changes a delivery promise or creates work outside the agreement.

Who Owns an AI Marketing Decision? covers the named owner, software authority and approval record. For a capacity-dependent recommendation, record which constraint was checked, which version of the data was used, what the owner approved and when that approval expires.

NIST's AI Risk Management Framework Playbook calls for defined roles and responsibilities for people using and overseeing AI systems. My operating rule is that the person responsible for the constrained work must have a say before a campaign creates more of it.

Then review delivery alongside acquisition. Did the team respond when it said it would? Did the orders ship on the agreed date? Did new work increase the backlog or disrupt existing customers? Keep those results with the campaign's costs and commercial outcomes. Proving AI Pipeline to Your Board covers the evidence needed to connect activity to later outcomes without claiming that attribution proves incremental growth.

Approve the delivery plan before approving more spend.

Frequently asked questions

Can AI recommend a marketing action the business can't actually deliver?

Yes. A recommendation based on campaign performance can pursue demand without checking the resources needed to serve it. Supply current capacity data, define limits for the relevant product, location and period, and require the responsible owner to approve any material increase in work before spend changes.

How do you add capacity limits to marketing AI?

Identify the resource that limits the offer, subtract existing commitments and an agreed reserve, and give the workflow a current source for the remaining capacity. Define the spending or booking limit, stop condition, data freshness requirement and restart authority. Enforce hard limits in the connected systems rather than relying on the prompt alone.

What happens when marketing generates more demand than sales can handle?

Unworked opportunities can accumulate and responses can fall behind the agreed schedule. Check the queue, assign the work and reduce or redirect acquisition if the team cannot meet its commitments. Improving the handoff or using automation may help, but verify the additional capacity before assuming it can handle more demand.

Which teams should approve AI-driven demand campaigns?

Name one marketing owner and involve the people responsible for the resources the campaign uses. Sales confirms follow-up capacity, Operations or fulfillment confirms delivery, Service checks onboarding and support, and Finance checks material funding requirements. Routine actions can run within approved limits; exceptions need approval from the affected owner.

Sources

Research checked October 13, 2026. The manufacturing statistic above concerns August 2026. Product documentation supports the described controls, not a claim that those controls are enabled in every company. The capacity rules and review process are my operating recommendations.