Industries
Why software employees leave and how to understand why
Engineers and product staff leave for many reasons beyond salary. How to run an exit conversation in a software company, read the pattern and act on it.
6 min readRead article
SaaS & Product Companies · AI product startups
HR for AI product startups: Scarce Researchers, Fast Pivots and Equity Promises
An AI product startup bets on a handful of engineers and researchers who are hard to find and easy to lose. Pay, equity, goals and growth are decided in a hurry, then questioned a year later.
Not sure what you need? Tell us about your business in six short steps and we will come to the first meeting prepared.
Share your requirementsAn AI product startup builds a product whose core is a model, a data pipeline or an agent, sold to businesses or consumers. The team is small. A few machine learning engineers, data engineers, product engineers and a founder or two carry most of the value, and one resignation can stall a release. The direction can change in a quarter when a new model arrives or a customer asks for something else. Offers are made against rival startups and large companies, often with equity as the sweetener. Titles, pay and expectations get set one hire at a time and drift apart. The wider IT and software hub covers what all software firms share, and the SaaS and product companies hub places this type among its neighbours.
- Plan
- Hire
- Onboard
- Manage
- Develop
- Perform
- Engage
- Retain
- Exit
- Improve
When HR is weak in AI product startups, you may see
- A few people hold the whole model
- Offers are made in a hurry and never compared
- Goals change faster than the review cycle
- Equity is promised in conversation
- Researchers and product engineers pull apart
- Juniors are thrown at live problems
- Titles stretch to keep people happy
These are the places the people side of this kind of business tends to give way. The rest of this page works through them.
Who does the work in AI product startups
HR here has to work for each of these groups, not only the office. The set-up below is written with all of them in mind.
- Machine learning engineers
- Research scientists
- Data engineers
- Product engineers
- Product managers
- Applied AI and prompt specialists
- Data labelling and quality leads
- Customer solutions engineers
- Founders' office and operations staff
Where HR strains in AI product startups
A few people hold the whole model
Knowledge of training runs, data choices and evaluation sits in two or three heads. If one leaves, the rest cannot explain why the model behaves as it does.
Offers are made in a hurry and never compared
Each hire is negotiated against a rival offer, so two engineers doing similar work can sit on very different pay, and they find out.
Goals change faster than the review cycle
A pivot or a new model makes last quarter's targets pointless, yet people are still judged against them.
Equity is promised in conversation
Founders say a grant in a call, and the paperwork and records follow late, which breeds doubt about what was agreed.
Researchers and product engineers pull apart
One side values papers and experiments, the other shipping dates, and appraisals tend to favour whoever is louder.
Juniors are thrown at live problems
Fresh graduates join, get little guidance and either learn fast alone or quietly burn out.
Titles stretch to keep people happy
Senior and lead labels are handed out to hold a hire, then become hard to defend when the company grows.
HR topics for AI product startups
Start with the topics that matter most here: Compensation, Payroll and Benefits; Performance Management; Retention and Attrition Management. Then work through the rest. Each topic is written for information technology and software, not in general.
Compensation, Payroll and Benefits
Salary structure, payroll inputs and a month that closes on time.
Performance Management
Goals, reviews and ratings that hold up in a conversation.
Retention and Attrition Management
Who leaves, when, why — and which of them mattered.
HR Strategy and Workforce Planning
How many people, in which roles, by when — and what they will cost.
Organisation Design
Structure, reporting lines and who is allowed to decide what.
Recruitment and Talent Acquisition
Requisition to joining, the same way every time.
Onboarding and Probation
A first day that is the same for the fifth joiner as the first.
Employee Administration
Records, attendance, leave and the day-to-day HR that must simply work.
Learning and Development
Training planned from need rather than from the calendar.
Career and Succession Management
Named backups for the roles that cannot be vacant.
Employee Engagement and Experience
Communication, recognition and the conditions people actually work in.
Employee Relations
Two routes — one for complaints, one for conduct — kept separate.
Policies, Compliance and Governance
Policies a manager can apply, and obligations with owners and dates.
Health, Safety and Wellbeing
Induction, records, incident reporting and follow-through.
Exit and Alumni Management
Resignation to relieving letter, in a defined order.
HR Technology and Automation
Records that survive people, and approvals that do not queue.
HR Analytics and Management Reporting
A short monthly pack management actually reads.
HR Operations and Governance
An HR function that runs the same way every month.
What a working HR set-up looks like
The pieces a AI product startups business puts in place, in an order that makes each easier. Open any card for the detail.
Write a short band for each kind of role
Machine learning, data, product engineering and applied roles each get a pay range and a clear step up.
What this coversHide
Write a short band for each kind of role
Machine learning, data, product engineering and applied roles each get a pay range and a clear step up.
What this coversHideMachine learning, data, product engineering and applied roles each get a pay range and a clear step up. See pay benchmarks.
Keep a record of every grant and promise
One sheet per person showing what was offered, when, and the paper that backs it, held by someone besides the founder.
What this coversHide
Keep a record of every grant and promise
One sheet per person showing what was offered, when, and the paper that backs it, held by someone besides the founder.
What this coversHideOne sheet per person showing what was offered, when, and the paper that backs it, held by someone besides the founder.
Set goals for a quarter, not a year
Short goals reviewed monthly, with a simple way to retire one that a pivot has made pointless.
What this coversHide
Set goals for a quarter, not a year
Short goals reviewed monthly, with a simple way to retire one that a pivot has made pointless.
What this coversHideShort goals reviewed monthly, with a simple way to retire one that a pivot has made pointless. The performance process can carry this.
Write down how the model is built and judged
A short note per project on data, evaluation and open risks, so a leaver does not take the story with them.
What this coversHide
Write down how the model is built and judged
A short note per project on data, evaluation and open risks, so a leaver does not take the story with them.
What this coversHideA short note per project on data, evaluation and open risks, so a leaver does not take the story with them.
Pair every junior with a named senior
A buddy, a first project of the right size and a check at two weeks and six weeks.
What this coversHide
Pair every junior with a named senior
A buddy, a first project of the right size and a check at two weeks and six weeks.
What this coversHideA buddy, a first project of the right size and a check at two weeks and six weeks. The onboarding process sets it out.
Agree what a title means
A one-page ladder describing scope at each level, used for promotions and hiring alike.
What this coversHide
Agree what a title means
A one-page ladder describing scope at each level, used for promotions and hiring alike.
What this coversHideA one-page ladder describing scope at each level, used for promotions and hiring alike.
Start by checking where you stand
Before changing anything, find out what is true today. These checks take an afternoon and tell you which of the topics above to start with.
Share your requirements- Ask each engineer to name the two people they would hate to lose, then compare the lists.
- List every current employee with their pay, title and equity promise side by side.
- Check which models or pipelines only one person understands.
- Count offers lost in the last six months and the reason given.
- Read last quarter's goals and mark how many still make sense.
- Confirm with a qualified professional the rules on employment terms, equity grants and employee records that apply to a company like yours.
Confirm with a qualified professional. What applies to you depends on your business, your state and your arrangements, and it changes. This page describes practice. It does not state a legal position.
How GullyHR helps
The services below fit this kind of business best. Each starts by recording where you stand, and every later report compares against that. We do not promise outcomes.
- Recruitment process consulting
A hiring route for scarce engineers, with a practical task and an honest account of the company's direction.
- Compensation benchmarking
Bands for machine learning, data and product roles set against what comparable teams pay.
- Performance management process
Short goals and reviews that cope with a pivot instead of punishing it.
- Employee retention and attrition management
A look at who is likely to leave, why, and what a counter-offer would really cost.
Start with a free conversation, or see the paid HR Diagnostic.
Our HR transformation approach
- 1
Assess
Understand the business, its workforce, current processes and HR challenges.
- 2
Identify gaps
Find the missing, weak, inconsistent or person-dependent processes.
- 3
Design
Create the structure, processes, policies, SOPs, dashboards and systems needed.
- 4
Implement
Help HR and department managers actually adopt the new system.
- 5
Automate
Digitise repetitive and approval-based workflows where it is worth doing.
- 6
Measure
Track hiring, attendance, employee cost, performance, training and attrition.
- 7
Improve
Strengthen the people-management system continuously, not once.
Blogs worth reading first
From the GullyHR blog: one on HR in information technology and software, and one on each of the topics this page points to.
Industries
Incentives that reward quality, not just output, in industrial component makers
How to design incentives for operators, setters and inspectors in fastener, gear, spring and valve units so that accepted lots, not piece counts, earn the bonus.
7 min readRead article
Industries
Measuring performance in chemical and specialty chemical makers: what to track and what to ignore
Measures that show how reactor, console, lab and gantry staff are really doing, the ones that mislead or bury near misses, and a monthly sheet an in-charge can run in twenty minutes.
7 min readRead article
HR Management
How to recognise and address employee workload problems
How to spot an unmanageable workload before it becomes an exit or an error, what to look at, how to talk about it and what to change, without guessing.
5 min readRead articleMore on the GullyHR blog.
Questions owners ask
Be clear about the problem, the ownership and the learning on offer, and set pay inside a sensible band. Use a practical task over a puzzle quiz. Strong people often choose scope over a small salary gap.
No. Bands should follow the role and level, not the person's negotiating skill. Machine learning, data and product engineers can sit in different ranges, with the same logic for stepping up inside each.
Use quarterly goals with a monthly check, and agree up front that a pivot can retire a goal without a penalty. Judge people on what they did with the information they had.
Note who was offered what, when, and under what conditions, and keep the signed paper beside it. Ask a qualified professional about the form. The aim is that nobody argues later about what was said.
Give them shared goals tied to a customer result, review them in the same meeting and recognise both papers and shipped work in appraisals. Shared outcomes cool most turf quarrels.
We would ask which two people you cannot afford to lose, list pay and titles side by side and check how equity promises are recorded. A role ladder with pay bands is a natural first job, with scope written down for your approval.
More on HR for information technology and software
All 6 business types in SaaS & product companies · Find another business type
B2B SaaSVertical SaaSEnterprise SaaSB2C SaaSSubscription software businessesCompensation, Payroll and Benefits in information technology and softwarePerformance Management in information technology and softwareRetention and Attrition Management in information technology and software
See all HR topics for information technology and software, or another industry.