At Miro, the digital whiteboard company, a signup was a useful event with an inconvenient flaw: it was free. Someone could open an account, draw a few boxes, and disappear. Someone else could bring an entire department. In an advertising report, both could begin as the same small victory. The money arrived later. The algorithm needed instructions now.
- The job: predict which customers will become valuable, then teach ad platforms to find more of them.
- The distinction: forecasts are engineered into signals with platform-specific timing and values.
- The lesson: judge a campaign by the business it produces, not merely the conversions it collects.
The signup that did not pay
Miro had been optimizing Google campaigns around corporate signups. Paid conversions often came after seven days. Voyantis built forecasts from engagement, onboarding and transaction data, then sent predicted conversion likelihood to Google. The test split the budget equally across five non-branded US search campaigns, comparing the old approach with predictive value-based bidding.
Voyantis reports 30% more paying corporate customers at the same media spend. That qualification matters. This was a change in what the budget bought. The case study also reports a shift toward annual plans, and larger teams at renewal. A useful experiment had made the company’s desired customer more legible to Google.
“We brought more annual plans (compared to monthly plans).”Robbert Smit, Miro’s Head of Paid Marketing
The auction needs a translator
Voyantis sits between a business’s customer data and the systems spending its advertising money. Its expertise combines machine learning with an understanding of how ad networks respond to incoming information. The company predicts future value, adjusts how that value is expressed, sends signals to advertising platforms, and updates them as behavior unfolds.
The interesting claim concerns the middle step, which it calls signal engineering. An accurate forecast can still be an awkward instruction. Google, Meta and TikTok have their own learning mechanisms. A score must reach the appropriate system at the appropriate time, expressed in a form that encourages useful bidding. Otherwise, a small forecasting error can attract a large advertising budget.
Actual outcomes return to the learning loop ↺
A grocery basket is not a habit
Swiggy’s Instamart offers another illustration. Its own data scientists already had a predictive model. The obstacle was turning that knowledge into effective auction signals across local markets. Voyantis built separate models for groceries and food delivery, rather than assuming that the two businesses shared one definition of a valuable customer.
Its published account describes signals delivered after a first order, refined as behavior accumulated, with values capped to keep inflated predictions from dominating the auction. A geographic comparison tested Voyantis against Swiggy’s internal model. The reported Instamart result was 51% lower acquisition cost and 34% more new customers. These are vendor-published outcomes from a particular implementation, rather than promises for the next advertiser.
A business built between two clocks
The company began in 2020 with Ido Wiesenberg and Eran Friendinger interviewing dozens of growth leaders. Wiesenberg’s earlier company, Tvinci, had been acquired by Kaltura. Friendinger had co-founded audience-segmentation business Adience, later acquired by MarketTech. They arrived with experience in advertising and data, then spent roughly two years developing Voyantis before its public launch.
The timing was part of the opportunity. In their launch account, the founders described the pressure to improve unit economics alongside changes in marketing privacy. Companies wanted profitable customers, while early campaign events offered imperfect clues. Voyantis proposed using company-owned data to bring a longer view into immediate decisions.
The 2022 launch came with $19 million in seed funding. In February 2025, an Intel Capital-led $41 million round brought announced funding to about $60 million. The business had also reported tripling annual recurring revenue during 2024. Those are different measures: investor capital funds the operation; recurring revenue reflects customers buying its services.
Buying the machinery, keeping the steering wheel
The current product names describe two points of intervention. Acquire feeds predictive signals into advertising. Engage applies customer intelligence to activation, retargeting, retention and win-back. Its proposition includes identifying where an intervention can make an incremental difference, rather than distributing effort or incentives to everyone who looks likely to buy.
This is a sales-led enterprise software business with implementation and ongoing operational support. Growth teams use it alongside existing marketing systems. Data integrations include tools such as Snowflake and BigQuery; campaign delivery runs through established networks. The company’s market includes subscription software, fintech and consumer services. Miro, Rappi and Shippo make the range tangible.
The alternatives include building the machinery internally, using other predictive marketing vendors, or relying on ordinary conversion optimization. Voyantis’s argument is that the enduring work lies in activation and upkeep. Models, pipelines, customer behavior and network algorithms keep changing. Its promise is to take responsibility for that continuing technical burden.
Copy the experiment before the result
A marketer can borrow the reasoning before choosing any software. Define the outcome that pays: a renewal, a repeat order, a profitable account. Find early behaviors that anticipate it. Compare the new signal against the existing campaign under controlled conditions, and evaluate both using business outcomes. Keep spending and measurement comparable.
The cost calculation should include the experiment itself. Even when media budgets are held steady, a team is committing attention, data work and live auction exposure. The sensible question is whether additional customer value exceeds the full cost of achieving it. An acquisition dashboard alone cannot answer that. Nor should a forecast of lifetime revenue be treated as cash already collected. Waiting for cohorts to mature is inconvenient, but it gives a business the chance to check whether its early confidence was deserved.
The conditions matter. This approach depends on usable customer histories, reliable data delivery and enough campaign activity to evaluate a change. A sparse dataset or shifting definition of value makes the forecast harder to trust. A broken pipeline can make a sound prediction arrive too late. The operational discipline belongs in the buying decision.
In August 2026, ADWEEK recognized Voyantis Acquire in its Tech Stack Awards. The more transferable idea is smaller than an award: tell the algorithm what your business actually values. A signup opens the door. The customer who returns is the reason to keep the lights on.