The anti metric: when a winning test is quietly losing you money

A test can win on the number you chose and lose the business money on a number you did not. The protection against that is an anti metric: a guardrail you agree in advance must not degrade while you chase your primary measure. Pick it before launch, write down the level at which you would stop, and read it beside the result. Without one, a variant that doubles form fills while halving lead quality gets rolled out everywhere.

What does a losing winner actually look like?

Arithmetic first, because the shape of the problem is easier to see with numbers in it. The ones below are illustrative.

Your landing page produces 40 enquiries a month and your sales team accepts a quarter of them as worth working. Ten accepted leads. You test a shorter form, dropping the company size question and the budget question, and enquiries go to 80. The dashboard shows a 100% uplift. The test is declared a win and the shorter form goes onto every page you own.

Then the accepted rate settles at one in ten. Eight accepted leads. You are two down on where you started, and your team has spent its month disqualifying 72 enquiries instead of 30. At twenty minutes a disqualification call, that is around fourteen extra hours spent proving that people who were never going to buy are not going to buy.

Volume went up. The business went backwards. Every part of it was visible in the data, provided somebody was looking at the second column.

Why does nobody catch this for a quarter?

Three reasons, and they stack.

The good news arrives first. Volume reads in days. Quality reads at 14 to 30 days, once leads have been worked. Revenue reads at 60 to 180. So the uplift is on a slide roughly three weeks before the damage is visible anywhere, and the rollout decision gets made inside that gap. Nobody is being careless. They are acting on the only evidence that exists yet.

The two numbers live in different meetings. Marketing owns conversion rate and reports it on Monday. Sales owns accepted rate and reports it in the pipeline review. The two are rarely on the same page in the same room, so the trade between them is never seen as a trade. It is seen as one team doing well and another team complaining.

Sales complains in a register marketing discounts. "The leads are rubbish" arrives as an opinion, at roughly the volume it arrives every quarter, so it gets filed as friction. Six weeks later somebody runs the number and it turns out to have been a report.

By the time the pattern is obvious, the change is live on every form on the site. Because it won.

How do you choose an anti metric?

Ask what your change is most likely to break, then measure that.

Almost every improvement in B2B marketing has a natural opposite, because most of them work by making something easier. Easier for the right buyer tends to mean easier for the wrong one too. Shortening a form. Dropping a qualifying question. Broadening a keyword match type. Widening an audience. Softening an offer from a demonstration to a guide. Each one buys volume with a small amount of qualification, and the trade is usually worth it right up until it is not.

Three tests for a guardrail that will actually do its job.

It reads on a comparable clock, or one rung later. If your primary signal reads at seven days and your guardrail reads at 180, the guardrail is decorative. Sales accepted rate at 14 to 30 days is the workhorse for exactly this reason: close enough behind the result to catch it before rollout.

Somebody outside marketing owns the number. A guardrail defined, measured and reported by the team whose test is being judged is not a guardrail. Sales owning the quality number is the cheapest governance available to you, and it costs nothing to arrange beyond one agreed definition.

It has a written threshold and a stated consequence. "We will keep an eye on lead quality" is not a guardrail. "If sales accepted rate falls below 20% for two consecutive weeks, we revert" is. Write the sentence before launch, when you have no stake in the answer.

Which anti metrics work in B2B?

Six that earn their place, and what each is for.

Lead quality, scored as ICP fit. Rate every enquiry against a written ideal customer profile: sector, size, role, technology in place. Use it whenever you test something that changes who sees the offer. It is the broadest guardrail available and the one worth building first, because everything else can be derived once the scoring exists.

Sales accepted rate. The share of marketing leads that sales agrees to work. Use it on form, offer and landing page tests. Its advantage is practical: it already sits in your CRM and it is already owned by somebody who has no reason to flatter your test.

Average deal size. The guardrail for any test that lowers acquisition cost. Halving cost per lead by attracting smaller organisations is a real result, and it is a different result from the one being claimed on the slide. Watch it whenever you widen targeting.

Cost per meeting held. Use this whenever you test outbound or a booking flow. Cost per meeting booked can be improved simply by making booking easier, which mostly increases the number of people who do not turn up. Held is the honest version of that number.

Unsubscribe and complaint rate. The guardrail on frequency and list expansion. Email programmes rarely fail suddenly. They degrade across four quarters, and the unsubscribe curve is the only place it shows early enough to act on.

Sales hours lost to unqualified conversations. The hardest of the six to measure and the most persuasive in front of a board, because it converts a marketing argument into a cost line. Even a rough figure, minutes per disqualification multiplied by disqualifications, changes the temperature of the conversation.

What does a genuine quality signal look like?

TouchWood Play designs and manufactures bespoke playgrounds. The buyer is a school, a trust or a local authority, the specification is unique to the site, and the decision takes months and several people.

The full funnel programme ran across paid search, SEO, email automation, the website, martech and call tracking. It produced £1.1m in confirmed funnel revenue and £927k of tracked return against £135k of cost, a 1200% return over two years.

The figure worth pausing on is a smaller one. Proposals linked to the funnel rose 31.8%.

Proposals, rather than enquiries. That distinction is the argument of this whole post. A proposal for a bespoke playground costs the business real time to produce: a site to assess, a design to sketch, a price somebody has to stand behind. Nobody writes one for an enquiry that is not worth writing for. Which makes the proposal count a quality signal the business polices itself, without anyone having to agree a scoring model first.

If quality had degraded, the shape would have been obvious and unattractive. Enquiries up, proposals flat. Cost per enquiry falling while cost per proposal climbed. That is what a losing winner looks like in two columns, and it is why the second column has to be on the same report as the first.

Call tracking is worth a line of its own here. In a business where a serious buyer picks up the phone, an enquiry model that counts only form fills is measuring the least committed part of your demand and calling it the whole.

What do you do when the guardrail trips?

You have three options, and "carry on and watch it" is not one of them.

Revert. The default. The variant loses, the card closes, the outcome goes in the log with the reason attached.

Keep it where it worked. Quality sometimes holds in one segment and collapses in another. A shorter form can work well for returning visitors and badly for cold traffic. Running the winner only where it won is a legitimate outcome, as long as you write down where that was and why.

Put the friction back somewhere else. If the volume gain is large enough to be worth having, the question becomes where qualification goes instead: a question later in the process, a routing rule, or a separate follow up sequence for the leads who no longer answer the question you removed.

Log it either way. A tripped guardrail is one of the more valuable things a test register accumulates, because it marks a ceiling. You now know where that trade sits in your business, from your own data rather than somebody else's benchmark.

Where to start

Three things, and the first one takes an afternoon.

Put a quality column next to the volume column on whatever you report weekly. One is enough to begin with. Sales accepted rate, on the same slide, in the same meeting. Most of this problem disappears the moment the two numbers cannot be looked at separately.

Go back to your last three declared wins and check them at 30 days. Pull the leads each one produced and find out how many were accepted. In our experience at least one of the three tends to look different under that light.

Write the threshold down for whatever launches next. One sentence, before launch, naming the guardrail number and the level at which you would stop. That sentence is worth more than any dashboard you could buy this year.

Our B2B lead generation work is built around qualified volume rather than volume, and our conversion and attribution intelligence work puts the quality column and the volume column on the same page, where a board can see both.

Frequently asked questions

What is an anti metric? An anti metric is a guardrail: a number you agree in advance must not degrade while you improve something else. It is chosen before a test launches, given a written threshold, and read at the same time as the primary result. Its job is to catch changes that improve one number by quietly damaging another, which is the most common way a marketing test wins and still costs money.

Is lead volume or lead quality more important in B2B? Neither works alone, which is why they belong on the same report. Volume without quality burns sales time and inflates a metric nobody can bank. Quality without volume starves the pipeline. The practical answer is to run volume as your primary measure and quality as your guardrail, with a written level at which a volume gain stops being worth having.

How do you measure B2B lead quality? The most workable measures are sales accepted rate, ICP fit scored against a written ideal customer profile, meetings held rather than booked, and average deal size. Sales accepted rate is the easiest to start with, because it already exists in your CRM and it is owned by somebody with no stake in the test. Agree the definition with sales in writing before you measure anything.

What is a good sales accepted lead rate? There is no universal figure worth quoting, and any benchmark you are offered will come from a mix of businesses that do not sell what you sell. The number that matters is your own baseline over the last two quarters. Establish that first, then set your guardrail threshold as a fall from it rather than as a fixed percentage borrowed from elsewhere.

Why do more leads sometimes mean less revenue? Because most changes that increase volume do so by removing qualification. Fewer form fields, broader targeting or a softer offer all let more people through, including people who were never going to buy. If the accepted rate falls faster than the volume rises, you finish with fewer real opportunities and a sales team spending more hours to reach them.