North: making strategy something you can operate

The deck dies faster now

There is a specific kind of quiet that settles over a strategy deck about three weeks after the offsite. The slides were good, the room agreed, and then the deck went into a shared drive while the quarter started without it. The plan lives in slides, the numbers live in a BI tool, and nobody can say on a Tuesday whether the strategy is working. That was the situation when every piece of work was done by a person, and agents make it worse, because agent work doesn’t wait for the quarterly review: churn data gets reconciled overnight, and tickets get triaged before anyone logs in. Once agents do part of the work, human and agent work have to roll up through one structure, so that any Tuesday can answer two questions: are we winning, and how do we know?

North is our probe at that problem. To be clear about what it is: sprint work, a design-sprint prototype with a fictional company’s plan on its screens as placeholder data, and every number in this walkthrough, including the $40M ARR objective below, is sample data. The question the prototype chases is real, though. First we show why the quarterly ritual breaks, then how the plan gets defined and how the work rolls up, then the discipline that keeps the map honest.

Quarterly reviews assume human-paced work

The OKR ritual has a hidden assumption: work arrives at human speed, so checking the plan once a quarter is enough. The standard cadence the OKR tools themselves recommend is annual company goals broken into quarterly team cycles [1], and the newer AI features in those tools point at automating that ritual: pulling live numbers from Jira, Salesforce, and Google Analytics so key results update without manual input, and drafting the review reports [2]. The goal tree in those products records goals and the numbers against them. The work itself, the thing that claims to move the number, lives somewhere else.

Agents break the human-speed assumption from two sides. From one side, people need more direction than a quarterly review provides. Working alongside agents means deciding, week by week, what to delegate, what the numbers mean, and what good looks like, and that guidance has to be available when the question comes up. From the other side, the work agents produce needs checking against intent. An agent that reconciles churn data every night is executing somebody’s reading of the strategy, and whether its output actually serves the objectives is a question a person has to be able to answer. Most companies have no surface where that question can even be asked. In the terms we used in The Tideline, this is the verify-impact stage of the arc, and it stays human. What leaders need, then, is one place where direction flows down and evidence rolls up: objectives that people and agents can both work from, and a live read on whether all of the work is moving the numbers the company committed to.

Defining the plan becomes a conversation

Building that read used to be a consulting engagement: a firm ran the workshops, defined the KPIs, and wired up dashboards, and the result stayed current for about as long as the org chart that commissioned it. In the prototype the primary action is Draft a goal, by voice. Say the commitment the way you’d phrase it in a leadership meeting, “we want to reach forty million in ARR by the end of the year, mostly new-logo bookings, and VP Revenue owns it,” and the software drafts the structured objective, the metrics that would prove it, and the owner. Then it stops. Nothing is saved until a person approves it.

The same reconciliation runs on the messy inputs a real plan is made of. A strategy is never one document; it’s a 32-slide board deck, an annual planning memo, a Slack thread, and a voice note from the offsite, all disagreeing slightly. North reads each one, maps every commitment back to one model, and surfaces the conflicts instead of quietly picking a winner. The deck commits to $40M ARR while the leadership thread settled on $38M, so the screen shows the conflict with both sources attached and asks the one question a machine shouldn’t answer: which is the plan? The person does no reconciling and makes the call, which is the trade a chief of staff would want.

Every piece of work rolls up

The home screen is the strategy map: the mission at the top, stated as a commitment, and one card per function underneath, each carrying an objective, the title accountable for it, and the metrics that would prove it. Each metric also names the system its number comes from. Recurring revenue reads from billing, uptime from monitoring, retention from the customer survey. The map stays the record of what’s committed and who’s accountable; it doesn’t try to be one more dashboard someone has to keep green.

The part built for an agentic company is the rollup underneath the cards.

The rollup, on the prototype's sample plan: work items feed function KPIs, and function KPIs prove company objectives. Every arrow is a claim that this work moves that number.

A renewal call made by a person and churn data reconciled by an agent land in the same place: the customer function’s KPIs, which in turn prove a company objective. When an agentic deployment takes over intake triage, its throughput and error rate join the same tree. From the plan’s point of view a renewal call and an overnight reconciliation are the same thing, work claiming to move a number, so the tree treats them the same. That lets whoever supervises the strategy trace a line from a deployment up to a function KPI up to an objective, on any Tuesday, and see whether the work is moving the numbers the company committed to. The OKR tools we surveyed automate updating the number [2]; the design question North chases is what changes when the work itself, human and agent, lives in the tree.

The discipline that keeps the map honest

Two smaller decisions do quiet work in the same direction. Every metric carries its direction of good: time to first value shows a down arrow because shrinking it is the win, and the sample plan’s most opinionated metric, spreadsheet exports per account, points down at zero because the fictional mission promises to “replace the spreadsheets in between.” A mission that never measures the thing it promised is easy to feel good about and impossible to check, so the map insists on the measurement. Objectives that are standards rather than projects, like 99.95% availability, are marked Ongoing instead of being forced onto a due date. And the map shows two of the five sample objectives at risk, plainly, at the top of the screen, rather than averaging them into a comforting composite score.

Give every metric a direction of good, mark standing commitments as ongoing, and show at-risk objectives at the top instead of averaging them into a composite.

The agent does the legwork, and you make every call

An OKR tool stores a plan a person typed in and leaves that person the exhausting part: chasing updates, reconciling the deck against the memo, and rebuilding the board deck every month. North hands that legwork to the agent, the reading, the reconciling, the chasing, and keeps the person as the decider on everything: which number is the plan, whether the drafted goal is right, whether an at-risk objective changes this week’s priorities. That’s the split to carry into Monday, whatever software you use. List what your strategy ritual actually costs in legwork, hand the reading and reconciling to software, and keep every decision, including the small ones, with a named person.

Are we winning, and how do we know?

One consequence of taking the map seriously is stated in the prototype as a refusal: North will not render an objective that has no owner, and it will not render an objective that has no metric. An objective missing either one isn’t ready to direct anyone’s work, human or agent, so the software declines to show it as part of the plan until a person supplies what’s missing.

The other consequence is a test you can run without us. The office of strategy, the one to three people who own the plan but none of the work, spends most of its time today chasing updates and rebuilding the board deck; a tool like North is the kind of software we’d build as bespoke SaaS, shaped around one company’s plan and vocabulary and changed weekly as the plan changes. Some random Tuesday, ask your own company the prototype’s core question: are we winning, and how do we know? If answering takes four tools and a meeting, that gap is what this design was built to explore. Open the prototype at the top of this page and walk the sample plan with your own in mind.