The Orphan Finding

Somebody in your company is about to write a goal in an area where two or three similar goals have already been tried. They will not know. The people who tried are still employed, some of them are in the building, and the information will not reach the person writing.

That is the usual argument for writing things down, and it is a good one as far as it goes. Here is the part that should worry you more. Even if somebody did write it down, the person writing still probably will not know. Finding it requires them to go looking, and you cannot search for an answer you do not know exists.

Capture was never the hard part. Matching is.

The category you create by accident

The moment you set a measure, you create a second category without meaning to. Everything that is real, but does not count. Not wrong. Not noise. Just orthogonal to the thing you happened to be chasing.

Melted chocolate is not a radar result. A weak adhesive is not a strong adhesive.

Spencer Silver at 3M was trying to make a tougher glue and made one that barely held, peeled clean and could be reused. Against his objective, a failure. It sat for five years before Art Fry connected it to the bookmarks falling out of his hymnal. Twelve before the product launched.

Twelve years. There is no planning cycle on earth that holds that.

And notice what saved it. Not a process. A bloke who would not let it go. Every story like this is the same: somebody carried the orphan personally, because there was nowhere to put it.

The problem was never the goal. It is the absence of anywhere for a finding to live when it does not serve this quarter’s outcome.

When somebody says goals killed their inventiveness

Almost always they are wrong, and it is worth being able to say why without being smug about it.

The serendipity list is survivor bias. Penicillin, Viagra, Post-it Notes, the microwave. Nobody writes a listicle about the transistor, or GPS, or an mRNA vaccine at pandemic speed, and all of those came out of brutally directed programmes with hard measures and impatient funders. Accidents get the headlines and become a wisdom of sorts, because accidents make better copy.

And the accidents were less accidental than the retelling suggests. Fleming’s objective was antibacterial agents; he noticed the contaminated plate because he had stared at thousands of clean ones. Sildenafil emerged from trial measurements rather than in spite of them. Goodyear spent years obsessively trying to stabilise rubber before the stove did him a favour.

So what is the thread worth keeping? Direction built almost everything. But you also need eyes and a brain that can spot the anomaly.

Here is where companies actually miss out. They have one of these three things.

• Weak measures that do not distinguish signal from anything.
• No route for an anomaly. Somebody finds something real and there is nowhere to put it.
• A culture where reporting something odd looks like admitting you missed your number. If this did not work, but it did something interesting, and saying so costs you status in the room, nobody says it. Ever.

The fix is much less romantic than twenty per cent of the time. It is writing down what you found in properties rather than verdicts. It is somebody owning the list. It is a manager whose first reaction to an anomaly is curiosity rather than a raised eyebrow about focus.

Properties, not verdicts

This is the practice that makes a finding retrievable, and it costs nothing.

A verdict is closed. It failed. It did not work. We stopped it.

A property is open.

The adhesive held at about a tenth the strength we wanted, peeled off without residue, and could be reused four or five times.

Write what it did and under what conditions. Never just whether it passed.

A verdict answers one question, once. A set of properties answers questions nobody has asked yet.

Three questions worth asking instead of the big one

The big one is usually “how do we balance exploration and exploitation?”, which nobody has ever answered in a meeting. These three are answerable this week.

• Where do findings go when they are real but do not serve the current objective? If the answer is in somebody’s head, or a Slack channel nobody reads, you have your answer.
• Who owns that list, and when does anyone look at it? An unowned list is a landfill. Somebody has to be accountable for reviewing it against live problems, on a cadence, or it rots.
• What is the longest useful idea our cycle length can survive? If everything has to pay back inside a quarter, you have decided, without ever discussing it, that you are not in the business of anything that takes longer.

Every company that tried this built a database

Innovation portal. Lessons learned repository. Ideas bank. They all die the same death, which is write-only. People put things in. Nobody ever takes anything out.

It is not laziness. Retrieval needs somebody with a live question to go looking, and they do not know the thing they need is in there.

Which is the part that has changed, fairly quietly, in about the last three years. Semantic matching against a live problem statement is precisely what this generation of AI is good at.

Not an innovation brain for your organisation. Something duller and more useful. Findings written as conditions and properties, and something watching new problems go past that says: this was tried in 2023, here is what it actually did, it just was not what we wanted at the time.

That is the Art Fry job. The half that creates the value, and the half no filing cabinet ever did.

Worth being straight about the catch. It only pays when the second question turns up, and you cannot forecast when that is. Which makes it hard to build a business case for and very easy to kill in year one.

It is cheap now, though. And the alternative is the one most companies are running today, which is hoping somebody in the building is stubborn enough to be Spencer Silver.

Where this lives in a working system

Nothing extra, which is the design constraint that makes it survive. Look at what the other five things already produce. The Focus has the choices with the hardest thing named. The Numbers has each measure with its evidence next to it. The ideas have the tests that were run and what they showed. The Goals have resourcing and what came off. The Work has weekly confidence with the reason, and endings with a grade and what they taught. The Rewards have a record of who was recognised for what.

All of that is already in the memory. Nobody writes a retrospective document, because the reasoning was captured as a byproduct of deciding. Which is why capture has to be a field rather than a task. Anything that needs somebody to sit down afterwards and write it up will happen for two cycles, then stop.

The orphan gets one more field, in the ending: anything you found that was real but did not serve this goal. The ending is the only ritual in the method where it can be caught. If it is not caught there, it is gone.

How it goes wrong

• It becomes a records system nobody reads. Prevented only by the read-before-you-write rule.
• The capture becomes a chore, so it stops. Fields, not homework.
• The reasoning gets left out. “We stopped the project” is a record. “We stopped it because the leading measure moved and the outcome did not, so the relationship we assumed was not there” is a memory.
• Meaning stored where only software can see it. Confidence as a cell colour. A threshold with no direction. If a person scanning back cannot read it, neither can the AI.
• Verdicts instead of properties. It failed answers one question once.
• Nowhere for the orphan. A finding that is real and does not serve this quarter has no home, so it survives only if one person carries it personally. That is luck wearing a lanyard.
• An unowned list. Nobody is accountable for reading it against live problems, so it becomes a landfill.
• Write-only. Everything goes in and nothing comes out, because retrieval needs somebody with a live question who does not know the answer is in there.

The test

The tags do the work: choice one, two, three. Ask this and see whether you get an answer.

What are we doing about C2? How confident are we this week, what have we learned, and what would make us change the strategy?

Answering that needs the hardest thing, two experiments, four team goals, the resourcing on each, this week’s confidence with the reason, the evidence behind the leading measures, and the stop condition. In most companies that is three documents and somebody’s afternoon.

If your installed AI can answer that question, you have a memory. If it cannot, you have files.

Questions worth asking

• Could somebody who joined last month find out why you stopped the last thing you stopped?
• If you asked your AI what you are doing about your second strategic choice, would it know?
• Where do findings go when they are real but do not serve the current objective?
• Who owns that list, and when does anybody look at it?
• What is the longest useful idea your cycle length can survive?
• When did anything last come out of your lessons-learned system?

From the ZOKRI OKR Handbook, the methodology we install and maintain. Written by Matt Roberts.

// about the author

Matt Roberts, co-founder, ZOKRI

A UK-based strategy and OKR consultant and two-time SaaS founder with a venture-backed exit, Matt turns strategy into execution for teams scaling from tens to thousands. He co-founded ZOKRI in 2018, having previously co-founded Linkdex, a venture-backed enterprise SaaS platform he led to a trade sale. He writes the methodology behind these notes.

Read Matt's profile →Book Matt →