The troublesome question arrived from a customer’s CFO. What would happen to revenue if the company switched off paid social for a month? Lifesight’s software could trace the touches before a purchase. According to co-founder Tobin Thomas, it could not answer that. A dashboard full of numbers had reached the edge of what it knew.
- The job: estimate what marketing adds, then help decide where money goes.
- The method: connect models, geographic experiments, and adjusted attribution.
- The catch: convincing evidence needs suitable data, enough signal, and a willingness to change the plan.
The question the dashboard couldn’t answer
Founded in 2017, Lifesight started in location intelligence. Its present business sells marketing measurement to consumer brands and agencies. The company’s website names more than 300 brands, spanning retailers, restaurants, and ecommerce businesses. Its proposition speaks to the person spending the budget and the person asking whether the spending worked.
“I didn’t have a good answer.”
Tobin Thomas, on the CFO’s question
The Decision Desk, May 2026
Thomas’s account gives that proposition some teeth. He says Lifesight chose to rebuild around causal measurement, giving up standalone attribution. Customers left. Roughly a fifth of revenue disappeared within two quarters, he wrote. Those figures are his account, rather than audited disclosures. The commercial lesson is uncomfortable: a measurement company can lose business by questioning numbers its customers like.
Three ways to cross-examine a sale
Consider someone who sees an advertisement and buys a shirt. Attribution can allocate credit along that journey. But perhaps the shopper already wanted the shirt. Rewarding the advertisement for being nearby can make an ordinary purchase look like a marketing triumph.
Lifesight connects three methods. Marketing mix modeling examines spending and business outcomes over time, accounting for influences such as promotions and seasonality. Geographic experiments change advertising exposure in selected markets and compare outcomes with suitable controls. Adjusted attribution brings the calibrated findings into more detailed campaign decisions.
The distinction is how the methods inform one another. An experiment challenges a model; the model informs attribution and the next allocation. Lifesight competes with specialists and broader platforms, including Measured, Haus, Recast, and Northbeam. Buyers should inspect that connection in practice. Three modules on a menu do not establish that their answers agree for sound reasons.
The shirtmaker changes the numbers
Seidensticker supplies a revealing example. In Lifesight’s published case study, the apparel company faced isolated channel reports and spending increases that did not reliably produce proportional revenue gains. Finance wanted evidence for upper-funnel spending. Marketing needed a way to defend worthwhile channels while reducing waste.
The team consolidated measurement, compared incremental returns, simulated allocations, and used geographic experiments. Lifesight reports that July 2025 revenue finished 11.5% above the monthly target while advertising spend landed 11.7% below the planned budget.
Those denominators matter. “Above target” is different from year-on-year growth; “below budget” is different from a measured saving against a control group. The useful takeaway is the sequence: question reported efficiency, compare options, test uncertain channels, and revise the allocation. Copy the discipline before copying anyone’s percentages.
The customer who never clicked
Obvi’s problem was more physical. The wellness brand wanted to know whether connected-TV advertising through Universal Ads moved products inside Walmart. Impressions were visible. The connection to a shopping basket was harder to establish.
Lifesight’s geo-test concentrated advertising in seven states and used fourteen comparable states as controls, with minimal or no campaign spending. The reported result was a 5% incremental lift in Walmart in-store sales at 95% confidence. Lifesight published the analysis; Walmart Connect did not validate or endorse it.
That experiment addresses something a click trail cannot directly settle. A television viewer may buy days later, in a store, without clicking anything. Geography becomes the instrument for comparing what happened with an estimate of what would otherwise have happened.
What the buyer is actually buying
Lifesight sells annual subscriptions. Its current plans are Performance, Precision, and Enterprise. Performance covers testing and adjusted attribution; Precision adds MMM, forecasting, and budget planning. Enterprise extends the program across brands and countries. Precision and Enterprise include managed measurement, bringing strategists and marketing scientists into the work.
The live pricing page asks buyers to request a proposal. A documentation FAQ lists a $20,000 annual starting fee for measuring up to $5 million in annual marketing spend. Treat that as a published starting reference: the current plan structure, integrations, and support scope need a specific quote.
Time is another cost. Current guidance puts a first attribution read at roughly one to two weeks and a full MMM read at four to six weeks once suitable data is available and validated. Geographic experiments commonly run four weeks or longer. Evidence has a production schedule.
Give the agent a better question
The October 16, 2026 platform announcement puts AI agents inside this workflow. Teams can investigate changes, compare plans, and prepare recommendations. The website says budget actions require approval unless customers have set rules authorizing them. Connecting analysis to action makes the quality of the underlying evidence consequential.
A second October development invites scrutiny. Lifesight released the data, saved forecasts, and scoring code behind ensemble forecasting research. Analysts can rerun the calculations and benchmark alternatives. Publishing the machinery offers something a polished product demonstration cannot: an opportunity to check the work.
Start with the sale that wouldn’t happen
Thomas acknowledges that small samples, weak signals, and low budgets can make calibration uncertain. Teams also need credible comparison markets and usable business data. A model cannot manufacture an informative experiment from an unsuitable setup.
There is nevertheless a question any company can borrow. Before asking which advertisement deserves credit, ask what outcome the business expects to change, and how it will recognize that change. Agree the test before seeing the result. Then be prepared to spend differently. Otherwise, even an excellent answer becomes another number in a meeting.