When to use this
Reach for an improvement when you have a goal that:- Spans days, not a single chat — a metric you want to move steadily, where progress only shows up over repeated passes.
- Benefits from a tracked baseline — you want to see where you started, where you are now, and whether the trend is going the right way.
- Generates a stream of decisions — you’d rather review a handful of concrete proposals over time than re-investigate from scratch each day.
- Should never act without you — you want the analysis automated but every change kept under your sign-off.
How it works
You start an improvement from a chat conversation. State the goal in plain language; the improvement is anchored to that conversation, which becomes the place you steer it later (see Steering an improvement). From there, Nash takes over on a recurring schedule:It establishes a baseline
On its first pass, Nash figures out the right metric for your goal, reads where that metric stands today, and records it as the baseline. It also forms early hypotheses about what’s driving the number.
It works the goal on a cadence
On each later pass, Nash queries your analytics, updates the current value of the tracked metric, revisits its hypotheses, and reflects on what it has learned so far. It carries its own running notes from pass to pass, so each run builds on the last rather than starting over.
It proposes recommendations
When Nash identifies a concrete action worth taking, it writes it up as a recommendation — a titled suggestion with its reasoning and the impact it expects. Recommendations pile up for your review; Nash never carries them out itself.
You decide
You review each recommendation and approve or reject it. Nash uses your decisions, and any guidance you add in the linked conversation, to shape its next pass.
What Nash tracks
As an improvement runs, you can check in on its progress at any time and see:| What you see | What it tells you |
|---|---|
| Goal | The objective you set, in your own words. |
| Status | Where the improvement is in its life — see Lifecycle. |
| Tracked metric | The metric Nash chose to measure progress against your goal. |
| Baseline vs. current | Where the metric started and where it stands now. |
| Passes completed | How many times Nash has revisited the goal. |
| Start and end dates | When the improvement began and when it’s set to complete. |
| Cadence | How often Nash revisits the goal. |
The metric, baseline, and hypotheses are Nash’s own read of your goal. If it picks the wrong metric or you want it to weigh something differently, say so in the linked conversation — your guidance feeds into its next pass.
Recommendations
A recommendation is Nash’s concrete suggestion for moving your metric — for example, rescheduling a cluster of at-risk deliveries, or shifting volume away from an underperforming provider. Each one carries a title, the reasoning behind it, and the impact Nash expects. Recommendations arrive in one of two states:- Pending — waiting for your review. You approve or reject it.
- Approved — already cleared. This happens when you approve a pending recommendation, or automatically when its action type is one your org has pre-cleared (see Pre-approving routine actions).
Approving a recommendation records your decision and signals your intent — it’s how you tell Nash a suggestion is good. Carrying out the underlying change is a separate step you take in your normal Nash tools; the improvement itself stays read-only and doesn’t execute the action for you.
Pre-approving routine actions
If certain kinds of recommendation are always fine — a category of routine adjustment your team has already agreed on — you can pre-clear that action type for your org. From then on, any recommendation of that type lands already approved instead of pending, so it skips the manual review step. This is a per-action-type allowance you grant and can remove at any time. Everything not pre-cleared still arrives pending and waits for your explicit decision.Cadence
By default an improvement revisits its goal once a day, in the early morning, and runs for two weeks before completing on its own. You can set a different cadence and a different duration when you create it, and the cadence runs in the time zone you choose. If you need more time, you can extend an improvement, which adds days to both its duration and its end date. If it’s done its job early, you can stop it.Each pass does real work and consumes part of your organization’s usage budget. If your org is at its weekly usage limit when a pass is due, that pass is skipped rather than queued — Nash picks back up on the next scheduled pass once budget is available. See Usage for how the weekly budget works.
Steering an improvement
Because an improvement is anchored to a chat conversation, that conversation is your steering wheel. Messages you post there — “focus on the evening shift,” “the real problem is provider X,” “ignore weekends” — are read by Nash on its next pass and factored into its analysis and recommendations. You don’t need a special command; just talk to it in the linked conversation the way you would in any chat.Lifecycle
An improvement moves through a small set of states:| Status | What it means |
|---|---|
| Active | Running on its cadence, revisiting the goal and proposing recommendations. This is where an improvement spends most of its life. |
| Paused | Temporarily stopped. No passes run while paused. Resume to pick back up on the cadence. |
| Completed | Reached the end of its duration and stopped on its own. |
| Cancelled | Stopped early and archived by you. It no longer runs and drops off your active list. |
| Control | What it does |
|---|---|
| Pause | Stops all scheduled passes without losing anything. The improvement reports as paused until you resume it. |
| Resume | Puts a paused improvement back on its cadence. |
| Extend | Adds days to the duration and pushes out the end date so the improvement keeps working. |
| Delete | Stops the improvement and archives it. It stops running and is removed from your active list. |
FAQ
Does an improvement ever change anything in my operation on its own?
Does an improvement ever change anything in my operation on its own?
No. Improvements run read-only. Nash analyzes your data, tracks a metric, and proposes recommendations — but it never cancels, reschedules, reassigns, or sends anything on its own. Every action stays under your control.
What's the difference between an improvement and a custom agent?
What's the difference between an improvement and a custom agent?
A custom agent runs a repeating playbook — the same job, the same way, each time it fires. An improvement holds onto a single goal and keeps working it over days, tracking one metric and accumulating recommendations as it learns. Use a custom agent for repeatable routines; use an improvement for a measurable objective you want pursued over time.
How do I influence what Nash works on?
How do I influence what Nash works on?
Talk to it in the conversation the improvement is anchored to. Recent messages there are read on Nash’s next pass, so you can redirect its focus, correct an assumption, or point it at the real problem just by saying so in chat.
Why did a scheduled pass not run?
Why did a scheduled pass not run?
The most common reason is that your organization hit its weekly usage budget — passes are skipped, not queued, when the budget is exhausted, and resume on the next scheduled pass once budget frees up. See Usage.
Who can create and manage improvements?
Who can create and manage improvements?
Creating an improvement and acting on its recommendations is a managed permission, typically granted to operations managers and admins rather than every user. Viewing an improvement’s progress and recommendations is more broadly available. If you can see an improvement but can’t act on it, you likely have read access without management access — check with whoever administers your Nash account.
Related
Custom agents
Set up a scoped, reusable agent for a recurring job — run it on demand, on a schedule, or on an event.
Usage
See how token usage and the weekly budget that gates improvement passes work.
Memory and learning
Let the agent build context about your team and operation over time.
Knowledge
Ground Nash with your SOPs, policies, and reference links.