A CRM’s first job is to represent the business faithfully. Every team needs that picture to anticipate what comes next. Better AI emails matter less than knowing customers are losing interest before you hire another ten salespeople.
AI is changing what customers need, what competitors can offer, and what anyone can charge. Product-market fit can erode while a company is still executing last quarter’s plan. In an unstable market, companies need to see those changes before they commit more money and people.
Better forecasting requires a more complete record of the business and its environment. Business systems have evolved around that requirement before.
Airlines, 1968. A sold-out flight could still depart with empty seats. American Airlines needed to anticipate cancellations and passengers who never arrived. Its reservation offices had operated independently, leaving the airline without the shared passenger data needed for sophisticated forecasting. SABRE centralized bookings and cancellations. In 1968, American introduced an automated flight-load predictor to determine how far to overbook. Knowing who had reserved a seat was the starting point. The valuable prediction was how many would actually fly.
Retail, 1987. P&G forecast consumer demand from orders placed by retailers and distributors. Those orders reflected promotions and inventory decisions, producing large swings even when consumer demand was steady. Its partnership with Walmart supplied a better measure: checkout data showing what shoppers actually bought. Consumer purchases became the basis for replenishment. Better forecasting required seeing through the retailer to the end customer.
CRM, 1999. Enterprise CRM had an adoption problem. In Behind the Cloud, Marc Benioff cites a Gartner estimate that 65% of Siebel licenses went unused. Salesforce reduced the burden with a simple interface delivered through a browser. The cloud removed installation work; the interface reduced the effort of entering customer information. The forecast depended on a record that salespeople would actually maintain.
Each advance changed the information available to the business. Better forecasts followed a more complete record.
Why now
A company can keep hitting its targets while the reasons customers choose it are disappearing. AI changes what it costs to serve a customer, what competitors can offer, and what anyone can charge.
In March 2025, Epoch AI reported that the price of matching GPT-4 on PhD-level science questions had fallen by 40x per year.
Cheaper inference can improve your margins while letting a competitor include your core feature for free. Better models can let customers do the work themselves. The technology that makes a product cheaper to deliver can also weaken the reason to buy it.
Chegg saw the consequences in 2023. In May, it reported that ChatGPT was hurting new customer growth and declined to forecast revenue for the second half of the year. Its forecast depended on how students would use an alternative it did not control.
At Tome, we reached a million users and raised a $43 million Series B in February 2023. Within a year, we concluded the product would not carry a company. We cut the team from 70 to 7 and started over. Rapid adoption had not established that the product could sustain a business.
Past growth is evidence of demand under past conditions. When AI changes the alternatives customers have and what they will pay, last quarter’s growth becomes a weaker guide to next quarter’s revenue. Better forecasting gives a company time to revise its pricing, product, and hiring before those shifts become a revenue shortfall.
What has been tried
Automating repetitious work isn’t the true endgame of AI CRMs. The early ones promised automatic field updates, account research, follow ups, and outbound prospecting. These are useful capabilities; but, woefully unimaginative for a system that holds the company's understanding of its business.
Automation inherits assumptions about what is worth doing. Tell an agent to prospect into a market and you are betting that those companies need your product, will pay for it, and are worth the cost of acquiring. Get those bets wrong and the agent can write excellent emails and book worthless meetings. Choosing whom to pursue is itself a forecast. The value of the work depends on the quality of that forecast.
Execution can improve while the underlying bet gets worse. A better system must connect what customers say, how they use the product, and what they buy to detect when those bets need to change. It must give every team the evidence to anticipate what will work, forecast the consequences, and decide where to act.
The business world model
Our ambition is to make the future of the business radically more predictable. That requires the fullest record of business reality that technology can preserve. No company knows in advance every question it will need to answer. Its current forecasts cannot define the limits of what it records.
APIs bring together sales conversations, campaign responses, support issues, and product usage. LLMs connect them to customers, people, and commitments. Each team's observations become evidence for the others. Fields and summaries are beliefs of that history. The original evidence and chronology must survive them.
Consider a competitor that begins including your core feature for free. Exposure to that competitor becomes a new forecasting problem. The model must revisit past conversations and usage, assess which customers depended on the feature, and estimate the renewal revenue at risk. The evidence has to exist before the threat is recognized. A new field cannot recover evidence that was never retained.
A business world model represents the business, how it changes, and the actions available to it. Forecasting happens at every level. A BDR predicts which accounts are likely to buy. An AE predicts which features will resonate in a demo. A marketer predicts how a campaign will land. A PM predicts which roadmap investment will improve adoption or retention. Each commits time or resources against an expectation of customer behavior. Each needs access to what the other teams have observed.
A more complete representation would also connect customer history to budgets, team capacity, and conditions outside the company. Doubling the sales team and holding hiring steady to address churn imply different futures for revenue and cash. Evaluating those choices requires forecasts under each, with the evidence, assumptions, and uncertainty exposed.
The next CRM must give every team a clearer view of reality and what comes next. It should reveal weakening demand before the company expands its sales team, and emerging opportunities while there is still time to pursue them.
We are building Lightfield to make a seemingly unpredictable business radically more predictable.
The value of seeing the future is having time to change it.
