We predict a campaign's result before you pay for it.
Kerblabs rebuilds your city as a model: your real streets, your real channels and the size of the audience you can actually reach. A hundred AI marketers write competing plans. Twelve hundred synthetic local people live through them. Then we compare our forecast with what really happens.
Five steps from your street to a forecast.
Each step uses the strongest evidence available and says where every number came from. When we do not know something, it counts as zero, never as a guess in your favour.
Know
We research your business before anyone plans: your website, competitors, market, reviews, and your own social profiles and their real audience size.
Public pages, search results, your own dataPlan
A hundred AI marketers in ten departments each write one plan, with hard numbers: posts, messages a day, flyers, ad spend, partners.
Built only on the facts from step 1Check
Plans that cannot reach at least 100 real people with your actual channels and budget are ruled out before any audience test.
The reach gatePredict
The synthetic people react to each surviving plan. Their reaction, multiplied by the reach you really have, gives people reached, leads and sales as ranges.
Real reach × simulated responseProve
After launch, your real results replace our assumptions. We record the forecast against the actual numbers, and the model learns from the gap.
Forecast versus actual
Synthetic people, built to disagree.
A panel that likes everything is useless. Ours is designed the way behavioural research designs a sample: varied, consistent and hard to please.
Built from local research
Each person belongs to a segment drawn from cited research about your area and category: age, area, occupation, income, the apps they use, what they want and what makes them doubt.
A personality, not just a profile
Every person carries a Big Five personality (openness, conscientiousness, extraversion, agreeableness, worry), so a careful, sceptical shop owner reacts differently from a curious, outgoing one.
A fixed stance mix
Champions, pragmatists, sceptics, blockers and observers in fixed shares, ordered along an adoption curve. Sceptics need proof, blockers have a concrete reason it will not work for them, and observers mostly scroll past.
Memory of your brand
People remember what they saw from you on earlier simulated days. A fourth message lands differently from a first one.
They never see our notes
The panel sees only what a customer would see: the headline, the message, the offer and where it appears. Test plans, tracking, budgets and our hypotheses stay hidden, so they cannot steer the answers.
- Example synthetic person
- Banquet hall owner, 45–49, Kompally, Hyderabad
- Stance
- Sceptic needs proof before believing a marketing claim
- Personality
- Organised and careful, reserved, worried about losing money
- Wants
- One clear answer for the wedding season
- Already saw from this business
- Day 2: a free-preview post, ignored it
- Says
- "Show me one hall like mine where this worked, then we talk."
Illustrative example in the format we use. Synthetic people are not real customers and hold no real person's data.
Every test is checked for flattery.
AI models like to agree. Each audience test gets an honesty score out of 100, built from four checks. A low score marks the result as unreliable instead of letting it pick a winner.
Stance violations
Sceptics and blockers who suddenly say yes with high intent. Above 15% means the panel broke character.
Echo rate
Quotes that repeat the ad's own words back. Real people describe things in their own language.
Yes share
A panel where most people would buy is implausible for a first message from an unknown business.
Spread
Intent that barely varies means the people stopped reacting as individuals.
The real world in the model, and the rules around it.
Your streets
Street maps from OpenStreetMap, road-traffic statistics by hour of the day, and your own street counts and photos when you share them.
Your channels
Your social profiles and their public audience size, found on your website and in search results. Social pages are never scraped; you can add your own insights.
Your reviews
Ratings and review themes for you and nearby competitors, read from search results only. Google is never read by our tools; you add your Google rating yourself.
Your history
Past sales, ticket or footfall exports calibrate the model. Names, phone numbers and emails are removed before anything is stored.
Nothing runs by itself
Every AI job starts only when you press Start. Nothing is posted, sent, bought or spent on your behalf.
One client, one dataset
Your data is used only for your business. It never shapes another client's synthetic people or results.
Every number says where it came from.
Observed or given: your own data, a public page we read, a search result that states the figure.
Cited, but indirect: a number from a source we could not fully verify.
A planning value we had to pick, such as a cost per click or a reply rate. It is shown so you can replace it.
Computed by our engine from the inputs above, such as reach, footfall on a street or a score.
A reaction of synthetic people. Useful for comparing plans; never presented as real customers.
We publish how close we were.
A forecast is only worth something if it is checked. For every campaign that runs, we record the forecast before launch and the real result after it, and we publish the gap here. Our first live check is an event on 22 November 2026.
| Campaign | Forecast before launch | Actual | Gap |
|---|---|---|---|
| Live event, Hyderabad, 22 Nov 2026 | Recorded before launch | After the night | Pending |
What a simulation cannot tell you
- The exact number. We give ranges, and early forecasts are uncalibrated until your own past data is added.
- Events nobody can foresee, such as a rival's surprise launch, the weather or a news story.
- How good your execution will be. A great plan posted badly still fails.
- Anything about real individuals. Our people are synthetic by design.
See your own city rehearse your next campaign.
One Hyderabad business, a free preview within three working days. Pricing after the preview is not published yet.