Spend but no purchases: what Meta's connector returns

The question is the right one. It is the first thing to check in any account and usually the most expensive thing you will find: money that went out and brought nothing back. Old prospecting left running, a geography that never converted, a test somebody forgot.

What almost nobody checks is the answer. You get a clean list of campaign names, it has the shape of a finding, and you start pausing things. This page is about what that list is actually made of, because on 25 August 2026 I asked the connector this exact question against live accounts and the answer was wrong in a way that leaves no trace on the page.

Three different ways of saying nothing

Every reporting row that comes back has to represent an absence somehow, and the connector has three separate conventions for it. Here is a real campaign row, values removed, from a purchase- optimised campaign that never recorded a purchase:

"name":          "…",
"amount_spent":  "…",
"omni_purchase": null,
"results": {"indicator": "actions:offsite_conversion.fb_pixel_purchase",
            "value": "Not available"}

Purchases come back as null. The result count comes back as the words Not available. And in a window where a campaign did not deliver at all, the spend field is not zero either — the key is simply not in the row, so a request for spend returns a name and an id and nothing else. Three renderings of nothing, and not one of them is the digit that the question was about.

The field name is worth noting too, because it is not the one you would guess. Ask the connector's own field-metadata tool about purchases and it does not resolve; the canonical name is omni_purchase, and purchases is only its display label. It counts purchase events from connected Meta business tools and from Meta itself, and its description excludes messaging-ad purchases delivered to and from Europe and Japan.

The filter is not the safety net

The field metadata says omni_purchase is filterable and offers the full set of comparison operators, so asking for zero-purchase campaigns looks like a solved problem. It is not, because a filter compares values and a null is not a value. Rows with nothing to compare pass straight through, in both directions.

It is not inert everywhere, and it would be dishonest to say it is. Narrow the same query to the last seven days on the same account and the spend filter returns an empty list, correctly, because there is nothing to return. The failure is specific and it is worse for being specific: the filter works when the data is there and quietly waves everything through when it is not — which is exactly the condition you were asking about.

The habit that catches this is asking for the number rather than the verdict, every time, on every question you would spend money on. See what the weekly routine checks →

Why "results" is not the shortcut either

The obvious workaround is to stop asking about purchases and ask about results instead, since every campaign has some. The connector will give you that, and it tells you plainly why you should not trust the column: each row's results object carries an indicator naming which action it counted.

Those all came out of live accounts on the same afternoon. Ranked in a single table they look comparable and they are not; one campaign's result costs what a page load costs and another's costs what a customer costs. The companion field is more honest about it — cost_per_result comes back with the meaning printed in brackets after the figure, (Website leads) or (Website purchases). That bracket is the part to read first, and it is the part that gets dropped when a table is summarised for you.

How to ask it so the answer is worth something

1. Ask for the spend column, not the list START HERE

   "Which campaigns bought nothing" invites a list of names. "Show every campaign with its spend and its purchases for this window, sorted by spend" invites a table you can audit, and the leak is whatever you can see and add up. A row whose spend is blank is not a small leak, it is a campaign that did not run. The trade: a longer answer with rows you have to read rather than a verdict you can act on — which is the point.

2. Ask the mirror-image question

   Run the question and its opposite: which campaigns had zero purchases, then which had at least one. If both lists come back the same, the filter did not filter and neither answer means anything. This takes one extra prompt and it is the only check here that catches the failure from the outside, without knowing anything about how the connector works.

3. Read the indicator before the result

   Whenever a results or cost-per-result column appears, ask what each row's result actually is before you compare any two of them. The trade: it kills the tidy account-wide "cost per result" number some reports lead with — there is no single one, and the connector will not produce it at account level for the same reason.

4. Total it against Ads Manager once

   Take the leak figure and check it in Ads Manager with the same two dates before you pause anything. This is the one step that cannot be delegated and the one that makes the rest worth doing — the same arithmetic check that catches invented numbers.

A null is not a zero, and a list of names is not a finding. The money is in the column you did not ask for.

• Accepting a list of campaign names as the answer. Without spend attached, a trivial test and the biggest line in the account look identical.
• Assuming a blank cell means zero. It far more often means the campaign did not deliver in your window, which is a different fact with a different action attached.
• Pausing on the first pass. A campaign optimising for something other than purchases will always show zero purchases and may be doing its job perfectly.
• Letting the window drift. Two figures from two windows do not subtract, and the connector will happily produce both without mentioning it.

Try this tonight Ask for zero-purchase campaigns, then ask for the ones with at least one purchase, and compare the two lists. If they overlap at all, stop and re-ask with the spend and purchase columns shown — and the total you get from that second pass is the number worth taking into tomorrow.

This is question one of the four-question audit, taken apart. If you have not connected an account yet, setup runs to about three minutes.