Restoring New-Business Mojo

Repeated allocation to a performing channel builds competence in that channel, which improves its results, which justifies further allocation.

Every step is defensible on the evidence available at the time. The loop is self-reinforcing and produces no signal that any capability is being lost.

Foresight does not prevent it and testing does not catch it. Someone doing all of it right still ends up preferring what pays back immediately over what pays back with practice - even when the two are worth exactly the same. That condition is the part doing the real work: you can accept every figure your paid search reports and still act on this. They are two different questions - what a channel returns, and what the business could do without it.

The 1988 definition, verbatim:

favorable performance with an inferior procedure leads an organization to accumulate more experience with it, thus keeping experience with a superior procedure inadequate to make it rewarding to use

Note the mechanism is about accumulated experience, not about the merits of either option - which is why no amount of performance data would have surfaced it.

A 2020 revisit of the same trap showed the bias survives a rational, foresighted learner following an optimal exploration strategy, between options of identical expected value. The authors' own summary: selection systems, even optimally designed ones, are biased against late-bloomers.

Every one of those four habits is exercised only when a team has to go out and find buyers. For as long as paid search kept delivering people who had already put their hand up, not one was required - and a practice that is not required stops being practised.

If marketing's success is what made them unnecessary, marketing's decisions are a large part of what makes them possible again. None of the four is yours to perform; all four are affected by choices made here, and two are decided by them.

Tolerance of non-response depends on the nos being real ones. Nineteen in twenty not buying today is the market, and a team can absorb that. A wrong name, a bounce, a company nothing like the size on the record - those are not nos, they are nothing at all, and in the moment they feel identical. Which of the two a team is mostly getting is a data decision.

Belief in a long horizon depends on what the activity is measured against, and the measurement window is a reporting decision before it is a sales one. Shorter than the pipeline being built, and persistence is recorded as failure.

Composure depends on expectations set before the first contact rather than explained after it. What counts as a normal response rate for this kind of work is a number you own and they do not.

Improving the message is half yours. Testing creative across a campaign is something you already do competently; it has simply not been commissioned for outbound in years. The delivery half belongs to them: presenting to someone who did not ask is a different skill from handling a lead who has already put their hand up, and it sharpens the same way, by doing it and adjusting.

When a team stops trusting what it is working from, it stops working the audience and starts picking through it - which looks like diligence, and is the audience being abandoned one name at a time.

That is not fixed by a better briefing. It is fixed by the quality of what they were given before any briefing happened.

Research into industrial settings established the principle: what an organisation produced in the past systematically overstates what it is able to produce now. Capability depreciates when it is not exercised, and the historical output carries no signal that it has happened.

Your performance data is an accurate record of what this business did. It is not evidence of what it can still do. Those are different claims, and only the first has ever been measured.

Which leaves one question worth putting plainly: if paid search stopped clearing the bar - not stopped, merely stopped clearing it - what would the company do instead, and how long before it could actually do it? Most planning conversations answer the first half confidently and have never tested the second. The testable version is simpler: not how much outbound this business once did, but what it could put into the field this quarter.

Keep the ads running and keep taking the orders they bring. The case for direct access is additive rather than a swap - and it is not purely a deferred-return channel either. Some of the people it reaches are ready now, whether or not they have declared it anywhere, and those orders arrive on an ordinary timescale.

So the comparison is not short-term against long-term. Paid search collects the demand that has declared itself. Direct access reaches the rest, and leaves recognition behind as a by-product of the same contact.

The 95:5 split is only half the argument. The consequential half is what it implies about position: buyers entering a category do not run an open evaluation, they start from a shortlist assembled from existing recognition.

Around four of the five suppliers seriously considered are on that list from day one, and the one ranked first before any seller contact wins more than eight times in ten. Which is why presence beforehand does work that reach at the moment of intent cannot: by the time the moment arrives, the list is largely drawn.

Dawes (Ehrenberg-Bass) for 95:5, widely disseminated through the LinkedIn B2B Institute; 6sense's 2025 Buyer Experience Study for the day-one shortlist.

Together they convert 95:5 from a claim about timing into one about position - being reachable when a buyer enters the market is worth very little if the list was drawn before you got there. We made the timing argument in Your Rented New-Business Pipeline in September.

Think about what each layer of reporting is built to answer. Platform reporting answers how a campaign performed. Attribution answers where a lead came from. Budget review answers how this year compares with last. All three are useful. All three cover a period. None of them can tell you what the business is still able to do.

There is a second limit, and this one is about who sets the reporting rather than what it can say. In September 2020 the search terms report stopped showing terms below a volume threshold, so some of the queries you paid for never appeared in your reporting at all. Seer Interactive measured 30-plus businesses and 5.1 million data points straight after the change, and found search-term visibility on cost falling from 98.7% to 71.0%, and on clicks from 98.3% to 77.9% - roughly 28% of budget reported blind. Google restored some of the historical data afterwards, and the report today carries considerably more than it did in late 2020. It has never gone back to complete.

In a channel you rent, the reporting you get is set by the party you rent from, and it can change without your agreement. A file of named contacts is different: you know who is on it because you can look.

You cannot look at performance data to settle this either, because performance data is the thing that looks healthy throughout. The channel worked, the reporting said so, and the reporting was right. No number in it would have flagged what this briefing is about, which is why it took thirty years of organisational research rather than a dashboard to name it.

It will go on reading healthy right up to the point where it stops, because it measures the channel and not your position in the market. Cost, competition and how buyers go looking all move on their own schedule, and none of them shows up in a monthly report until it has already arrived.

So the reading worth having is the one you form yourself, while everything still looks fine. The business may well carry on doing well - and it is still worth being able to reach a market directly just in case that changes.

Two things compound here. Abandoned options never generate the evidence that would correct the estimate, so the verdict stands unchallenged because nobody tried again.

The second is a measurement error, and it is the one that decides how much of this can be undone: a team fluent in inbound benchmarks will read a normal direct-response rate as failure, because the denominators are not comparable. The channel was not judged badly. It was judged against the wrong scale.

The hot stove effect (Denrell & March, 2001) describes asymmetric error correction in adaptive learning. A 2024 generalisation shows the bias holds when a negative belief merely reduces sample size rather than eliminating it - which is precisely the “one small mailing a year” pattern.

Field support comes from Dittmar & Duchin (2016), tracking 9,000+ managers and finding negative experience influences subsequent policy more strongly than positive. That study concerns financial policy, not marketing; it establishes the effect is real in senior decision-makers, nothing more.

This has a practical consequence for planning. Any capability rebuild has to be budgeted as a rebuild, not a restart - there is no dormant competence to reactivate.

Assume the learning curve starts near the beginning, and resource the first two or three campaigns accordingly rather than expecting them to pay their way.

The 2023 two-part review in The Learning Organization synthesises the empirical literature on turnover-induced knowledge loss and concludes tacit knowledge loss is materially more damaging than explicit, because codification is what makes transfer possible and tacit knowledge resists it by definition.

Sales turns over faster than most functions, which makes this exposure structural rather than incidental.

Framing this as a capability rather than a channel changes which conversation it belongs in.

The question is not whether to spend less on paid search. It is how many routes to market you actually control - and that is a strategy question, not a media one.

Worth following what that means in practice. The target universe is defined and acquired once, and the same file sits behind every one of those routes, however the approach is actually made.

So this is not a decision taken per channel. It is one decision that either makes several routes available or leaves none of them available, which is why it sits a level above the media plan.

The target universe is a procurement decision, resolvable quickly, with quality that can be specified and guaranteed. The operating temperament is an internal capability programme measured in campaigns run, not weeks elapsed. The measurement window is a management decision that costs nothing and gates both of the others.

Attempting the rebuild before the measurement window has moved produces exactly the failed first campaign that ends the initiative - and it will be attributed to the channel rather than to the reporting.

This is the hot stove effect applied to your own reporting.

The asymmetry is that negative early outcomes reduce resampling, so mistakenly low estimates persist while mistakenly high ones self-correct through further use. Later work shows the bias survives even when a negative belief merely reduces the sample size rather than stopping sampling altogether.

A measurement window shorter than the sales cycle manufactures a negative early outcome by construction. The channel is then judged on it, permanently.

You are the supplier of its raw material, and the quality of that material sets the ceiling on what the effort can produce.

The data job has two parts. First, clean what you already hold - goneaways, duplicates, records nobody can trace - which needs no purchase. Second, add data that covers the market as it stands, which your CRM cannot know.

The first is a morale intervention as much as a data one. A salesperson turned down by a real person is doing the job and knows it. A salesperson ringing a number for a company that closed in 2021 is being worn down by something you can remove.

The other thing marketing brings is a force multiplier. Contacts land differently into recognition than into nothing, so brand and content work running alongside makes the outbound easier - which is the 95:5 argument again, seen from the marketing side of the building.

The file is also the nearest thing you have to a lever on the persistence itself. Nobody keeps going because they were told to. They keep going because the work pays often enough to be worth repeating - the quality of the list largely decides whether it does. A team working current records will still hear no far more often than yes, but they will get far enough into enough conversations to keep pressing. A team working a stale file is just collecting evidence that the exercise is pointless.

Neither half of the data job is simple, though. Cleaning what you already hold takes research capacity most teams cannot spare, and finding B2B contact data that is accurate, current and comes with someone who answers when you ring is a job in itself.

The team judges the data as soon as they start calling, on how the calls feel rather than on any analysis - and enough dead records early makes that verdict negative and close to irreversible.

So getting the file right is not an efficiency optimisation. It decides whether the attempt gets a fair test at all.

The one thing you cannot check by looking at a sample is where the data came from. A supplier who documents it has made the purchase defensible. One who waves the question away has left you holding it.

Recency is the half you can check, and it matters just as much. Data verified recently is the most direct way to reduce the friction a team meets - fewer people who have moved on, fewer numbers that go nowhere.

Checking is easier than it sounds. Ask for a sample of companies you already know, or ones you can test in an afternoon. A supplier confident in the file will not hesitate, and a few calls settle more than a page of claims.

This is also the answer to “why now”. Not because anything has broken - nothing has - but because the build has a lead time and the exposure does not.

Any risk with an instant onset and a multi-quarter mitigation has to be addressed before onset, or not at all. That is a commercial framing rather than a channel one.

Early efforts are practice and should be resourced as such - the first campaigns exist to rebuild the skills, not to pay for themselves. Judging them on ROI recreates the hot-stove problem internally.

Whatever the figure is, it will not move on its own. The universe can be defined and acquired - that is procurement, and it resolves quickly. The operating temperament behind it cannot be bought at any speed.

Which is why it is worth establishing now rather than when it is needed. Asked while paid search is performing, the figure is a planning input. Asked after it stops, it is a post-mortem.