You set OKRs quarterly. You check them maybe twice. At quarter end the results surprise you. It is a visibility problem, and spreadsheets plus quarterly reviews guarantee it. You find out about slippage weeks after it happened, when there is nothing left to do about it.
Two cadences: daily scoring, weekly prediction
The agent wires into your metrics sources and scores continuously. Daily at 4 PM it pulls the latest value for each key result. Tracking "reduce churn by 2%," it grabs yesterday's churn number. Tracking "increase enterprise ARR by $2M," it grabs recent close data from Salesforce.
Then it scores you against the trajectory you need. At week 5 of a 13-week quarter you should be roughly 38% of the way to goal. At 25%, you are off-track, and the agent tells you that morning. Anything trending the wrong way, or with confidence sliding under 50%, gets flagged before you would have noticed on your own.
Friday at 3 PM it switches to prediction. Using current velocity, it models whether you hit each OKR, how confident it is, and what would have to change. The output reads like this: "You are at 65% of target with 4 weeks left. If you maintain this velocity, you will hit 82% of goal. To hit 100%, you need to accelerate 25%."
It also runs an off-cycle opportunities pass. Any OKR likely to blow past 120% of target gets flagged as overachieving, and the agent asks whether that effort should move to an OKR that is drowning. It hunts the quick wins that could drag a lagging one over the line.
What you connect: a metrics platform (Amplitude, Mixpanel, or a custom dashboard) that tracks your OKR metrics daily, your CRM for revenue OKRs like ARR and pipeline, product analytics for activation and retention, your OKR document with the goals and success definitions, and experiment tracking so the agent knows which initiatives are driving which metrics.
The payoff is simple. You catch a decline the day it starts instead of at the end-of-quarter review, and you fix it mid-quarter instead of explaining a miss. Set up the feeds, then ask the agent which OKR has the lowest confidence score today. Start there.
FAQ
What does the weekly prediction report look like? For each OKR: current progress, velocity per week, trajectory if velocity holds, confidence score, and a specific recommended action. The at-risk section lists every OKR below 50% confidence with a root-cause hypothesis and the highest-leverage fix.
What data sources does it need? A daily metrics platform, your CRM, product analytics, your OKR document with targets, and experiment tracking.