I'm Lilach Bullock, a British AI consultant and trainer with 23 years in marketing, and I help product managers use AI in two ways: to clear their own workload of research, specs and notes, and to plan AI features with clear success criteria. You keep the product calls. AI handles the drafting and the digging.
Product managers sit between customers, engineers, sales and leadership, and most of that job is writing and reading. Interview notes, specs, release notes, competitor scans, roadmap updates. AI is good at all of it, as long as someone with judgement checks what comes back.
I come at this from the marketing and go-to-market side. I've spent 23 years in marketing, I built, scaled and sold a digital marketing agency, and I've reviewed more than 300 tools on this site, many of them SaaS products built by teams just like yours.
Product work AI speeds up
These are the tasks where I see product managers win back the most time in the first month.
- Customer research: summarising interview transcripts, support tickets and reviews into themes, with quotes you can check.
- Specs and stories: first drafts of requirements and acceptance criteria from your notes.
- Competitor scans: structured comparisons of public product pages and changelogs.
- Release notes: one change written up for customers, sales and support in their own language.
- Roadmap updates: a short version for leadership and a longer one for the team.
Planning an AI feature: start with success criteria
When the product itself uses AI, the most common mistake I see is shipping without agreeing what good looks like. Anthropic's developer documentation on defining success criteria is the clearest guide I've found. It says good criteria are specific, measurable, achievable and relevant.
Its examples are blunt. 'Safe outputs' is a bad criterion. 'The model should classify sentiments well' is a bad criterion. A good one names the metric, the test set and the threshold, such as a score measured on a held-out set of 10,000 real examples. It also lists the dimensions worth deciding up front:
- Task fidelity: how well the model needs to perform the job.
- Consistency: how similar answers should be for similar inputs.
- Relevance and coherence: whether it answers what the user asked.
- Tone and style: whether the output sounds like your product.
- Privacy preservation: how it handles personal or sensitive information.
- Context utilization: how well it uses the information you give it.
- Latency: how long a user will wait for the answer.
- Running spend: what each answer is allowed to take from your margin.
What stays with the product manager
AI can summarise forty interviews. It can't tell you which three customers matter most to the business, or that the loudest feature request comes from the account you're about to lose anyway. Prioritisation, saying no, and the trade-offs between what customers ask for and what the business needs stay with you. So does the conversation with engineering about what's realistic.
How I work with product teams
I keep it practical and tied to the work you already do.
- We list the writing and research tasks in your product cycle and how long each takes.
- We pick the ones AI can draft with the least risk and build prompts and templates around your past documents.
- For AI features, we write the success criteria together before anyone builds anything.
- Where it helps, I connect the workflow to your existing tools so drafts land where the team works.
- We review after a fortnight and decide what to add next.
Working with SaaS teams
Many of the product managers I work with sit inside SaaS businesses that don't have a data team, and that changes what's sensible to build. My page on working as an AI consultant for SaaS startups without a data team covers that situation, and if your product team also owns growth, my notes on AI for SaaS marketing go into the go-to-market side.
Sessions run over video and in English, for teams in the UK, US, Canada, Australia and beyond. The full set of ways I work, from training to done-for-you builds, is on my work with me page.
Where product managers go wrong with AI
The first trap is trusting summaries without reading the source. AI will happily produce a confident theme from two comments. Always click through to the quotes. The second is pasting customer data into personal AI accounts. Use the business version your company has approved, and strip names when in doubt. The third is building an AI feature because a competitor did, with no criteria for success and no plan for checking it once it's live.
Questions to ask before buying an AI product tool
Every product tool now has an AI button, and most product managers I meet are trialling three or four at once. Before you add another, ask where it stores your customer research, whether your company's data settings apply to it, and whether it does anything your existing AI assistant can't do with a good template. Often the honest answer is that a well-written prompt in the tool you already have does the job.
My advice for product managers
Use AI on your own workload first, for a month, so you learn where it's reliable and where it invents things. That experience makes you far better at specifying AI features for customers. Then, before any AI feature goes into a sprint, write the success criteria in plain words with the team. If you can't say how you'll measure it, don't build it yet.
Frequently asked questions
Who helps product managers use AI?
Lilach Bullock, a British AI consultant and trainer with 23 years in marketing, helps product managers and SaaS teams use AI for research, specs and AI feature planning.
What can product managers use AI for?
Summarising customer research, drafting specs and stories, scanning competitors, writing release notes and preparing roadmap updates for different audiences.
How should product managers define success for an AI feature?
Anthropic's documentation says success criteria should be specific, measurable, achievable and relevant, naming the metric, the test set and the threshold.
Will AI replace product managers?
No. AI drafts and summarises. Prioritisation, trade-offs and saying no still need a person who understands the business and the customer.
Do product managers need to code to use AI?
No. The workflows I set up run in the AI tools and business software your team already uses.