Speero builds an AI growth harness into the tools your team already runs. Your goals, tests, research and rules go into one record. Then anyone, from the CEO to the newest designer, can ask your own AI agent in Slack and get an answer with its sources.
8 roles · 9 jobs · 72 example questions below. Runs in Slack, Teams or Claude.
@SpeeroBot new idea: show the price per delivery instead of the monthly total.
We tried something close in spring.
It was inconclusive because mobile traffic was too thin. Exit surveys back the idea, so run desktop and mobile together. Filed in Jira as VRD-112.
Every answer comes from the same record: your goal tree, past tests, research and the rules your team has set. Choose who's asking, then the job they need done.
The nine jobs SpeeroBot does for the Growth PM. Pick a job to compare all eight roles.
Intake. Ideas come in, get checked against what the program already knows, and get filed where the team works.
Sizing. How long a test takes to read and what a win is worth, worked out before anyone builds anything.
Goal tree. Which business goal each test serves, and which goals nobody is testing yet.
House rules. Brand, legal, privacy and program rules, applied before a test goes live.
Research gaps. The questions no study has answered yet, and the fastest way to answer them.
Learning loop. What past tests and research already proved, pulled up at the moment someone needs it.
Metric drivers. What the program knows, versus only believes, about what moves the numbers.
Self-learning. Tell it once. Results and corrections change how it works from then on.
Artifacts. Briefs, specs, scorecards and decks, drafted from the record with every claim linked.
@SpeeroBot I want to test a cheaper starter plan. Can we?
Logged, and it has backing.
Customers asked for a smaller plan in last month's interviews, so it passes the evidence gate. It's on the Asana board, ready to size.
@SpeeroBot can we test annual billing with a discount?
Logged, with one flag.
It passes the evidence and strategy gates. The flag: discounts have lost twice here, so I attached both tests to the task on the monday.com board.
@SpeeroBot the brand team wants to test a new homepage video.
Logged in Asana. Two gates passed.
It ties to the "more visitors start a plan" goal. It still needs a reason to believe it works. Any research or past results I should attach?
@SpeeroBot new idea: show the price per delivery instead of the monthly total.
We tried something close in spring.
It was inconclusive because mobile traffic was too thin. Exit surveys back the idea, so run desktop and mobile together. Filed in Jira as VRD-112.
@SpeeroBot from today's usability session: move reviews above the plan picker.
Logged in Jira as VRD-131.
Linked to the plan step goal, with the session clip from Dovetail attached. It passes two gates. Pick a metric and it's ready to size.
@SpeeroBot the plan page loads slowly on mobile. Worth a test?
Logged as a fix, not a test.
Speed fixes like this rarely need an A/B test. I filed it in Jira as VRD-133 and attached the mobile drop-off from GA4 so it gets picked up.
@SpeeroBot add-to-cart dropped on Safari last week. Anything there?
Logged as a bug check and an idea.
The drop lines up with the release on the 3rd. I opened a bug in Jira, and logged a test idea in Asana in case it isn't a bug.
@SpeeroBot three people in today's interviews couldn't find the skip button.
Logged. That's the fourth time.
Exit surveys mention it too. I added it to the Miro board with the clips from Dovetail, linked to the retention goal.
@SpeeroBot what's a pricing test worth if it works?
A lot, but it takes time.
A clearer price on the plan page could be worth about $180K a year, readable in 3 weeks. A full price change needs 8 weeks. The model is in Google Sheets.
@SpeeroBot if we fund one more developer, where does it pay back most?
Checkout first, then retention.
Checkout has the most traffic and three sized tests waiting. Retention tests run longer, but each win is worth more. The model is in Google Sheets.
@SpeeroBot can we A/B test the landing page for the fall campaign?
Yes, if the campaign runs 5 weeks.
Paid traffic is enough to spot an 8% lift in sign-ups in that time. A shorter run won't give a clear answer. The sizing is in Google Sheets.
@SpeeroBot plan page or cart. Which first?
Plan page first.
Plan page: 3 weeks to detect a 4% lift, worth about $180K a year. Cart: 6 weeks, and only a 7% lift or bigger shows up. About $60K. Sized on 90 days of GA4.
@SpeeroBot can we test all four plan page designs at once?
Two at most.
Four versions would take 12 weeks to read. Pick the two strongest from the Figma review and you'll have an answer in 6.
@SpeeroBot do I need to build the full meal-swap feature to test it?
No. A fake door is enough.
A button that counts clicks shows demand in 2 weeks. Build the real feature only if 1 in 20 visitors click. The spec is in Jira.
@SpeeroBot how long for the new checkout test?
4 weeks, starting Monday.
That catches a 3% lift in completed orders. Stop at 2 weeks and you'd only see a 6% lift or bigger. End date is set in Optimizely and on the monday.com board.
@SpeeroBot is the FAQ page worth an A/B test, or should I study it?
Study it. Too little traffic.
A test would take 14 weeks to see even a 10% lift. Five user interviews in Dovetail would tell you more, faster.
@SpeeroBot our 2027 goal is retention. Is testing lined up with that?
Not yet. Most of it is on sign-ups.
I drafted a plan in Notion that moves half the queue to retention. It needs your sign-off before the next planning call.
@SpeeroBot which tests support this year's revenue target?
About two thirds of them.
Those sit under first orders and retention. The rest are site polish with no clear link to revenue. The full map is in Airtable.
@SpeeroBot which goal does a new homepage headline help?
New subscribers, top of the funnel.
It sits under "more visitors start a plan." That branch has 9 tests and 2 wins, so the bar is high. The past headline tests are linked in Airtable.
@SpeeroBot which goals is our testing actually working on?
Almost all of it is on new subscribers.
Nothing is running on keeping them, and that's the bigger lever in your goal tree. Three tasks on the Asana board already fit it.
@SpeeroBot why is the cart redesign ahead of my plan page idea?
It's on a bigger goal.
The cart feeds first orders, the top goal this quarter. Your idea helps plan choice, which is already well covered. It's next on the Asana board.
@SpeeroBot I have two sprints for test builds. What first?
The retention tests.
Retention is next year's main goal and has the least testing. Three tests there are waiting on dev. They're at the top of the Jira board.
@SpeeroBot which goals have no tests on them?
Win-back has nothing.
Cancelled customers are a whole branch with no tests and no research. Two ideas from last year's Miro workshop fit it. Want me to add them to the plan in Notion?
@SpeeroBot which goal needs research the most?
Retention, by a long way.
It's next year's main goal, and only one study has looked at it. The other goals have five or more each. I drafted a research plan on the Miro board.
@SpeeroBot is anything we're testing off-brand?
One test, and it's paused.
The countdown timer broke the "no fake urgency" rule in Confluence. I paused it in Optimizely and told the team. Everything else is within the rules.
@SpeeroBot can we test a $14.99 per meal price?
Yes, once finance signs off.
Program rules say any price over $12 needs finance approval. I added you as the approver on the monday.com task. It can start the day you sign.
@SpeeroBot can we test "Cancel anytime, no questions asked" on the plan page?
Not with that wording.
Legal ruled out "no questions asked" in March. It's in the brand rules in Confluence. "Cancel anytime online" is approved. Want me to swap it in?
@SpeeroBot can the pricing test start next week?
Not until December 2.
There's a test freeze over Black Friday in the program rules in Confluence. I moved the start date in monday.com and let the dev team know.
@SpeeroBot can we test a new plan picker this week?
Not on the plan page yet.
VRD-104 is live there until the 21st, and two tests on one page muddy both. I queued yours in Asana for the day after.
@SpeeroBot can the checkout test run on Fridays only?
No. Tests run full weeks.
Program rules in Confluence ask for full weeks so weekday habits don't skew results. I set it to run 14 days straight in Optimizely.
@SpeeroBot can we call the bundle test on clicks instead of orders?
No. The metric was locked.
It was set to first orders before launch, and the rules say we don't switch after. Clicks can go in the Notion write-up as a side note.
@SpeeroBot can I interview people who cancelled last week?
Yes, after a 30-day wait.
Privacy rules in Confluence say we wait 30 days after a cancel. People past that window are ready to invite in Dovetail.
@SpeeroBot what's the riskiest thing we believe?
That price is why people cancel.
The whole retention plan assumes it, but only one survey backs it up. Interviews with recent cancels would test it in two weeks. Want me to set them up in Dovetail?
@SpeeroBot do we have research on price sensitivity?
No. It's a gap.
We've tested prices, but never asked why people pick a plan. A short pricing survey would cover it. I drafted one in Google Forms.
@SpeeroBot do we know why people pick us over the grocery store?
Not really.
Two old surveys touch on it, but nothing since we added bundles. A quick message test would show which reasons land. The brief is drafted in Notion.
@SpeeroBot any research on why people drop at the address step?
Some, but it's old.
A survey from last year in Dovetail says delivery dates confuse people. Nothing since the redesign. A quick five-second test would update it.
@SpeeroBot has anyone usability tested the new cart?
No, and it ships in two weeks.
The old cart had three studies. The new one has none. I set up five sessions in Dovetail and linked the Figma prototype.
@SpeeroBot do we know which errors customers hit at checkout?
Only from support tickets.
Tickets point to older iPhones, but nobody has measured it. Error tracking in GA4 would close the gap. The ticket is in Jira.
@SpeeroBot where is our data weakest right now?
Cancel reasons.
Half of cancels pick "other" in the exit survey, so we can't tell why they left. A better question would fix most of it. I drafted options in Notion.
@SpeeroBot what's the biggest gap in what we know?
Why people pick the smallest plan.
It's the top cancel reason on the retention branch, and no study has looked at it. Five customer interviews would cover it. Want me to draft the guide in Dovetail?
@SpeeroBot what did the program actually teach us this year?
Five things we now know.
Clear pricing wins. Discounts don't. Skip-a-week keeps people. Urgency hurts trust. Mobile needs its own tests. Each one links to its tests in Notion.
@SpeeroBot which wins from this year still show up in revenue?
Two of them.
Clear pricing and skip-a-week both held up in Amplitude three months later. The bundle win is still unproven, because those buyers cancel sooner. The details are in Notion.
@SpeeroBot what messaging has actually won for us?
Clarity beats hype.
Plain price-per-meal copy won three times. "Limited time" urgency lost twice. The winning lines are saved in the brand doc in Notion.
@SpeeroBot before we plan next quarter, what have we learned about the plan page?
Clear pricing wins. Discounts don't.
Across six tests in Optimizely and two studies in Dovetail, price clarity helped every time and both discount tests lost. Urgency has failed twice.
@SpeeroBot anything I should know before I redesign the cart?
Three things from past tests.
Showing shipping cost early helped twice. Trust badges did nothing. The one-page cart lost on mobile. I pinned all three as notes in your Figma file.
@SpeeroBot we're rebuilding the plan picker. Anything from past tests I should know?
Yes. Keep the default plan.
Two tests showed a preselected middle plan helps, and removing it hurt. The specs from those tests are linked in the Jira epic.
Three tests closed. One changes what we believe.
The starter bundle beat the discount, so "bundles beat discounts" moves to Supported. Results are from Amplitude, and the summary's in Notion.
Nice. I'll check the bundle numbers before finance sees them.
@SpeeroBot did any tests confirm what the interviews told us?
Four out of five did.
The one miss: people said they wanted more recipes, but adding them didn't lift orders. I noted it in Dovetail next to the original quotes.
@SpeeroBot what's our biggest growth lever right now?
Keeping people past week six.
That's where most cancels happen in Amplitude, and skip-a-week has moved it twice. Holding people there is worth more than the same gain in new sign-ups.
@SpeeroBot the bundle test won. Did it move revenue?
Not proven yet.
It lifted first orders, but bundle buyers cancel sooner in Amplitude. The net effect is unclear, so I added a 90-day readout to the Jira ticket.
@SpeeroBot can I tell the board free shipping drives retention?
Not yet. Say it drives first orders.
That part has two tests behind it. Retention is still a belief. The holdout in Optimizely reads out in six weeks.
@SpeeroBot does free shipping actually help retention?
We don't know yet.
Two tests show it lifts first orders. Nick (CMO) thinks it helps retention too, based on the last marketing meeting, but nothing has tested that. Want a holdout in Optimizely?
@SpeeroBot do real food photos actually help?
On the plan page, yes.
One test showed real meal photos lifted plan picks. Nobody has tried them in emails or the cart. That's a good next test, and the Figma file is ready.
@SpeeroBot will cutting load time actually help sales?
Probably. We haven't proven it.
Slow sessions convert less in GA4, but that could be the device, not the speed. A small slowdown test in Optimizely would settle it.
@SpeeroBot from what we've tested, what moves retention most?
Delivery flexibility, so far.
Skip-a-week lifted 60-day retention in two tests. Recipe variety tracks with it in Amplitude, but nothing has tested it yet. That's the next best bet.
@SpeeroBot do cancel reasons match what we see in the data?
Mostly, with one gap.
People say "too expensive," but the data shows they cancel right after a skipped week. Five interviews with recent cancels in Dovetail could explain why.
@SpeeroBot are you getting smarter, or just bigger?
Smarter. Here's this month.
Two beliefs changed after tests, one rule was added after a correction, and my forecasts got closer. Every change is logged in Airtable with who and why.
@SpeeroBot reminder: take refunds out before you report revenue.
Got it. Saved for every report.
I re-ran the Q3 numbers without refunds. Two wins got smaller, none flipped. The corrected deck is in Google Slides.
@SpeeroBot stop calling them "meal kits." We say "weekly boxes."
Done. Saved to the brand rules.
I updated Confluence and will say "weekly boxes" in every brief and write-up from now on. I fixed the two open briefs in Notion too.
The urgency banner lost again.
That's twice. I've updated the belief in Airtable, and new urgency ideas now get flagged at intake with both tests attached.
Good. Saves us a third try.
@SpeeroBot please always attach the Figma link when you file a test.
Will do, from now on.
Saved to the team playbook. I also added Figma links to the four open tests in Jira that were missing them.
@SpeeroBot heads up: anything on checkout needs two weeks of QA.
Noted. Timelines updated.
I saved it to the playbook and moved the start dates of two checkout tests on the monday.com board. Sizing now counts QA time.
@SpeeroBot heads up, gift subscriptions shouldn't count when you size checkout tests.
Got it. Saved to the playbook in Notion.
I'll leave gift orders out from now on. I re-sized the open checkout test in Jira too: it now needs 4 weeks, not 3.
@SpeeroBot question 4 in the March survey was broken. Don't use it.
Got it. Question 4 is set aside.
I flagged it in Dovetail. Two ideas leaned on it, so they now need new evidence. They're marked on the Asana board.
@SpeeroBot one slide on the testing program for Tuesday's board meeting.
Done. It's slide 7.
What we learned, what it's worth, and what's next. Every number links to its source.
@SpeeroBot send me the value of every win this year.
It's in Google Sheets.
One row per test, with the estimate, the source and who signed off.
@SpeeroBot turn the pricing results into a brief for the agency.
Brief's ready in Google Docs.
What won, the exact lines that worked, and what to avoid.
@SpeeroBot draft the Q3 program review for Thursday.
Done. It's in Google Slides.
Pulled from Airtable and Amplitude. Every claim links to its test.
@SpeeroBot I'm starting on VRD-112 today. Is there a brief?
Brief's ready in Notion.
Hypothesis, the locked metric, both variants, and the past test it builds on.
@SpeeroBot I need a build spec for VRD-112.
Spec's in Jira.
Both variants, tracking events, the QA checklist and rollout rules.
@SpeeroBot write up VRD-098 for the team.
Scorecard's ready in Google Docs.
The result, what we learned, and the next test it points to.
@SpeeroBot I need a guide for the cancel interviews.
Guide's ready in Dovetail.
Built from the gaps on the retention goal, so you don't ask what we already know.
Verdant is a made-up subscription brand. Every name, number, ticket and tool in these examples is demo data. Your harness answers from your own record, in your own tools.
The model answers. The harness is everything around it. It decides what gets read before an answer, which checks the answer has to pass, and what gets written back afterwards so the next question starts further along.
Slack, Teams or Claude. No new tool and no seat to license. A CFO and a developer ask the same bot and get answers that agree.
The past tests on that page, the goal it serves, the live tests it could clash with, the rules it has to follow. Then it runs the same gates every time.
Results move beliefs up or down. Corrections become rules. Every change is logged with who made it and why, so people keep the decisions.
Specialist skills sit behind them for sizing, research recipes, briefs, scorecards and quarterly reviews.
See it work on one of your own pages and one of your own ideas.
We lay the harness into the store, tools and model you already license. There's no platform to move into, and the record stays yours if we ever leave.
We point the intake agent at one of your pages and one of your ideas, then read one past test together.
We build your goal tree and scenario model with your leads, so every answer can say which goal it serves.
We set up the schema in the store you run and load the tests, research and rules you already have.
We wire in your testing tool, analytics, research tools and chat, then turn on the agents with your house rules.
Ideas in, tests sized, results read, beliefs updated. Then we hand it over, or stay and run it with you.
An advisory retainer. We hand over the schema, the gates and the agent playbooks, then hold weekly sessions while your team builds. We run the first intake and the first read with you.
Best when you have the hands and want the method.
A monthly build program. We lay the record, configure both agents on the model you license, connect your stack, load your history, and run the first full cycle with your team before handover.
Best when you want a running system with a clean record from day one.
Speero strategists and analysts run intake, reads, research and quarterly reviews on the harness with your team. Your people keep the decisions, and the record grows under your name.
Best when the program needs operators as well as a system.
Qualified teams start with a free assessment workshop. We audit your program and tools, then scope the build with you. From there it runs as a monthly build program or an advisory retainer.
A model answers questions. The harness is everything around it. It decides what the model reads first, which checks an answer has to pass, and what gets written back afterwards. That is what turns a chatbot into something that knows your program.
No. The record lives in the Notion, Airtable, Jira or warehouse you already run, and people ask in Slack, Teams or Claude.
The one you already license. The skills and the schema move with you if you switch models later.
Both are yours. The agent runs in your Slack or Teams, on the model you license, and you name it (it's called SpeeroBot in the examples on this page). The record sits in your store, under your name. If you take the program in-house later, nothing has to move.
No. It brings the evidence and flags the risks. People make the calls, and every change to a belief is logged with who made it and why.
It starts with a free assessment workshop for qualified teams. We audit your program, your tools and what you already know, and scope the build together. After that it runs as a monthly build program, or as an advisory retainer if your team does the building. Timeline and price come out of the workshop, because they depend on your stack and scope.
Thirty minutes. We point it at one of your pages and one of your ideas, and you watch what it works out before it asks you anything. If it fits, we pick the way in together on the same call.
Or email [email protected]