Each bullet starts with a strong verb, names the artifact or system, and lands a quantified outcome. The examples below show the structure across 12 roles, side by side with the weak version.
Weak: 'Worked on the recommendations system and improved it for many users.' Strong: 'Led a 4-engineer rewrite of the recommendations service in Python and PyTorch, lifting CTR 18% across 12M weekly users.'
Weak: 'Helped reduce server costs.' Strong: 'Migrated the search cluster from EC2 to spot instances on EKS, cutting compute spend from $48k to $19k per month at the same p99 latency.'
Weak: 'Was responsible for the onboarding feature.' Strong: 'Identified 38% of new signups churned at the import step; ran a 6-week effort across design and engineering to add CSV upload, lifting day-7 retention 14 points.'
Weak: 'Helped prioritise the roadmap.' Strong: 'Killed feature X after 3 weeks of beta based on 28% adoption and 4 negative usability tests; redirected 4-engineer team to feature Y, GA in 6 weeks at 71% adoption.'
Weak: 'Grew the user base via marketing campaigns.' Strong: 'Ran a lifecycle marketing program across HubSpot and Customer.io that grew MAU from 38k to 112k in two quarters at a CAC of $7.40.'
Weak: 'Improved email open rates.' Strong: 'Rebuilt 14-segment lifecycle program after audit revealed 28% list churn; lifted open rate from 18% to 31% and cut unsubscribes 41%.'
Weak: 'Consistently exceeded sales quota.' Strong: '$1.8M quota at 124% attainment, $42k ACV, 32-day cycle, ramped in 5 months; top 10% of AE cohort in FY24.'
Weak: 'Built strong relationships with customers.' Strong: 'Closed 4 enterprise expansions totaling $640k ARR by partnering with CS on a coordinated QBR motion across 12 strategic accounts.'
Under 22 words. Recruiters scan, they don't read. Past 22 words the bullet wraps to a third line on most templates and reads as a paragraph.
Target 55 to 75% of bullets with numbers. Below 55% the outcome-density score drops; above 75% reads as performative and recruiters get suspicious. A bullet without a number should still describe a specific change.
Within a role, no. Mix verbs to keep the eye moving. Across roles, repetition is fine when accurate.
Quote the input you controlled: team size, system scale, decision frameworks, process changes. 'Built X to handle 12k req/s' is a defensible outcome even without a downstream business metric.
Three to six. Five or six at the most recent role. Three at roles older than 5 years. Anything older than 12 years gets two bullets or rolls into 'Earlier experience'.
Free AI rewrites against a target JD.
Try it free →