2025-03-06 08:00:00
Imagine if every time you edited a document, the word processor forced you to retype everything that had been written before that edit.
How expensive would that be for a company?
This is exactly how data transformation works today. Each time a data engineer modifies some part of the data stack, the Cloud Data Warehouse & its transformation layer recalculates everything.
What if the system were designed so that it only recalculated the metrics needed? How much less expensive would it be?
Databricks partnered with Tobiko to quantify the impact. Selective recalculation delivers a 9x cost savings.
Selective recalculation - only calculating what needs to be - delivers the 9x cost savings. This efficiency becomes critical as data transforms from a business asset into the business foundation itself with AI.
Migrating to a new system is often expensive, but with SQLMesh’s dbt adapter, no code changes are required to the existing schema to support this cost savings.
In a world where every cloud dollar counts, it’s time to stop forcing your data warehouse to rewrite “War and Peace” when all you need is the Cliff’s Notes — your CFO will thank you.
2025-03-05 08:00:00
As startups scale, effective management becomes the difference between chaotic growth and sustainable success. After analyzing hundreds of posts on startup management, I’ve distilled the key pieces of advice that founders and leaders should keep in mind.
Management, like engineering, has repeatable design patterns that can be learned and applied
Communication complexity increases exponentially with team size, creating natural breaking points in management structures
Evaluate the decision-making process rather than just the outcomes
Great leaders balance execution (laying bricks), tactics (building walls), and vision (creating cathedrals)
Adapt your delegation approach based on the person’s experience level with the specific task
Culture becomes a critical management tool as your company scales
Effective management isn’t a natural gift but a discipline that can be learned, practiced, and refined. The best startup leaders recognize that scaling their management capabilities is just as important as scaling their technology or customer base.
2025-03-02 08:00:00
Which is the best business in AI at the moment?
I analyzed Q4 revenue data from publicly traded companies across multiple sectors—software companies, consulting firms, and hardware manufacturers to determine which segment dominates the AI market.
NVIDIA’s data center business dominated the field, generating $31b in Q4 revenue with impressive margins exceeding 70%.
In second place, Microsoft’s AI business, including Azure, is at a $3.25b Q4 revenue.
IBM reported $2b, which was a big surprise to me. Clearly, IBM’s role as the third largest generator of AI revenue is a combination of consulting and software revenue. In their most recent quarterly earnings, the company clarified the breakdown in their book of business : 20% software & 80% consulting.
For OpenAI and Anthropic, I estimated Q4 revenues by taking 40% of their reported 2024 annual projections. Data for Accenture and Salesforce came directly from their published earnings transcripts.
Just how large is the GPU business relative to others? Well, let’s stack them all up.
The GPU business is approximately four times larger than all of the software, infrastructure, and consulting sectors combined, based on public disclosures & pressreports.
While these figures are illuminating, they don’t capture the complete AI market landscape.
This analysis doesn’t capture several significant segments: hyperscalers like Amazon & Google who haven’t broken out their AI segments’ revenue, AI-focused data companies like Snowflake and Databricks, inference-as-a-service startups, other hardware manufacturers like Broadcom, and security companies such as Palo Alto Networks and CrowdStrike—all of which have burgeoning AI businesses.
Maybe this group doubles the revenue.
While precise total market figures remain elusive, the trend is clear: hardware investments still outpace software and services revenue in today’s AI landscape.
With hyperscalers investing at least $250b in datacenters this year, software revenue has a long way to go and years of exceptional growth to catch up to equal it.
2025-02-28 08:00:00
During a recent Theory Office Hours with Kady Srinivasan CMO at Lightspeed Commerce, Dropbox, and Klaviyo, we discussed several powerful insights emerged on how early-stage companies should approach marketing.
Here are the highlights :
Define Your ICP Before Anything Else
The most fundamental decision for any startup is determining your Ideal Customer Profile (ICP). Without clarity on exactly who you’re targeting, GTM efforts become diluted and ineffective.
The earlier the company, the narrower it should be. With scale & capital, it can broaden.
“I think it’s super important to know which layer of this you are tackling”. What we were selling was not just email marketing. We were actually selling email as a growth lever, meaning we could go in and say, ‘we’re going to help you drive 40% of your revenue from email,’ which was completely not done in the industry before. We could only get to that level of insight because we knew our ICP really well. We changed the game ie we changed the problem people bought us for."
When launching a startup, narrowly defining your ideal customer profile (ICP) is not just helpful—it’s essential for survival. Focus is “the most important thing” for startups, with “trying to straddle too many things” being “the biggest killer of startup success.”
A precise ICP forces you to concentrate your limited resources on a specific audience where you can perfect your sales motion, messaging & value proposition.
Counter-Position Against Industry Norms
Early-stage startups should challenge the status-quo. This isn’t about attacking competitors directly, but rather challenging how customers think about solving their problems.
“Counter-positioning is essential for startups because there’s a way that the world has moved, and you are saying, ’no, I think it will move differently,’” “You have to surprise the buyer. You have to tell them, ‘you’re thinking about the world incorrectly.’”
Klaviyo did this by positioning itself not as an email tool but as a revenue driver. We told customers, “Google is killing all your third-party cookies. You will only have first-party data to rely on, so if this is your time to switch, you have to do it now.” This created urgency around the problem and showed customers what was at stake.
Align your GTM strategy & Product to your ICP & TAM
Your GTM strategy should always result from considering your ICP’s buying behavior, needs, & the TAM, not the other way around.
PLG motions with 2-4% conversion rates require millions of potential users to achieve $100m+ ARR business. In markets with smaller customer counts, each lead is so valuable it should be managed by a salesperson, which increases conversion rates to 15-30%.
If you are a PLG business, you can migrate upwards over time. If you are an SLG business, becoming a PLG business is quite challenging to impossible. At Dropbox, we were primarily a PLG business that became more SLG over time as customers pulled us market.
Build Credibility Through Brand Value, Not Fear
While fear can drive short-term action, it creates negative brand associations. The most successful companies build marketing around positive outcomes and aspirational values.
“Creating fear never works, because in the immediate, you can probably prompt people to take action because they’re like, ‘Oh my! I must do something,’ but it leaves a negative perception in their mind.” “You don’t become a beloved brand over a period of time.”
Instead, focus on how your solution helps customers achieve positive outcomes. Klaviyo positioned itself as a premium brand that took its craft as seriously as its customers did.
“We don’t want to be like every other B2B email company with blue fonts. We want to be completely different.”
As you craft your marketing strategy, remember this advice:
“Your marketing should answer: I’m the best at… I’m the only… I’m the fastest ever… That automatically gets you into the blue ocean space. You’re playing a different game at that point.”
These four principles—narrow ICP, counter-positioning, aligned GTM & product, & positive brand building—don’t just help startups capture market share; they redefine markets entirely.
2025-02-27 08:00:00
Most startups play defense when discussing pricing with customers. They dance between asking for too little, leaving money on the table, and asking for too much, only to lose the customer’s interest. The very best companies lead their customers in that dance. They use pricing as an offensive tool to reinforce their product’s value and underscore the company’s core marketing message.
For many founding teams, pricing is one of the most difficult and complex decisions for the business. Startups operate in newer markets where pricing standards haven’t been set. In addition, these new markets evolve very quickly, and consequently, so must pricing. But throughout this turmoil, startups must adopt a process to craft a good pricing strategy, and re-evaluate prices periodically, at least once per year.
There are only three pricing strategies startups should pursue: Maximization, Penetration and Skimming. They prioritize revenue growth, market share and profit maximization differently.
Maximization (Revenue Growth) - maximize revenue growth in the short term. Startups should pursue maximization when there are no clear differences in customer segments’ willingness to pay, and when the optimal short term and long term prices are equal. Many mid-market software companies price with the goal of revenue maximization, negotiating for the highest possible price in each sale.
Penetration (Market Share) - price the product at a low price to win dominant market share. A bottoms-up strategy lends itself to penetration pricing. Price low to minimize adoption friction, grow quickly, and then move up-market after developing broad adoption. Penetration pricing leads to land-and-expand sales tactics. Expensify, Netsuite, New Relic, Slack follow this model. Penetration prioritizes market share.
Skimming (Profit Maximization) - start with a high price and systematically broaden the product offering to address more of the customer base at lower prices. Skimming is widespread in consumer hardware. Apple sells the latest iPhones at the highest prices, and repackages older models at lower prices to address different customer segments. As Madhavan Ramanujam tells it, Steve Jobs was both a product genius and pricing genius. By pairing the two skills, he led Apple to record-breaking profits quarter after quarter.
Skimming is less common in the software world because few startups develop a product at launch that will be accepted by the most sophisticated customers (and those willing to pay prices that generate the greatest margin). There are exceptions: Oracle’s database, Tanium’s security product, Workday’s human capital management software.
There are three ways to justify a pricing plan:
Value-based pricing charges customers a fraction of the incremental value created by the product or a fraction of the costs saved by the product. This is often seen in ad tech or any type of optimization technology. A startup increases conversions by 50% and they take 10% of the gain as their fee. Value based pricing is also employed in slightly less rigorous ways. Salesforce sells CRM seats based on an aggregate ROI of increased sales productivity for example. So does Expensify, which decreases the time to file expenses.
Cost-based pricing is when startups mark up the product they sell by some margin. Many infrastructure as a service companies do this. AWS, Twilio, Heroku, etc. It’s very common in commodity or nearly-commodity industries, where customers know the prices of the components used to provide the service.
Competition based pricing works well in markets where the price and value of a particular type of product are well established. Startups adopt the pricing model well known in the industry.
Positioning is the most frequently forgotten of the 4 Ps in Marketing. Salesforce exemplifies exceptional positioning. Used strategically, pricing can be a weapon, a source of competitive advantage in the market. Your company can employ pricing to communicate to the market whether your product is a premium, mid-market or low cost alternative.
Startups can choose to price below the market, to gain share and grow quickly (Zendesk, AirWatch); they can choose to price at the market price and differentiate based on product features (Dropbox and Box); or they can charge a premium for their product, which reinforces their positioning as the gold-standard in the sector (Palantir and Workday).
To be effective, a startup’s pricing strategy must align with its marketing case studies, website messaging, PR releases and sales pitches. If all the arrows point in the same direction, then pricing becomes an asset to reinforce the company’s position in the market.
The number of total potential customers multiplied by the selling price of the product equals the total addressable market (TAM) for a startup. Generally speaking, bigger TAMs are better. If there are a small number of relevant customers, as in Veeva’s case where the entire market is about 200 pharmaceutical companies, the average revenue per customer must be very high. At IPO, Veeva’s average customer paid the company about $750k per year. On the other hand, if there are millions of potential customers, as in Expensify’s case, the average revenue per company can be much smaller and still justify a billion-dollar-plus TAM.
Pricing impacts the structure of a sales team and their day to day performance for two reasons. Higher price points decrease sales velocity, the number of deals closed per sales rep per unit time, and increase sales volatility, the chances a deal closes.
Inside sales teams selling $5-30K products can sustain a deal velocity of 3-8 transactions per month, depending on quota. This keeps morale high and creates a very predictable revenue forecast. No individual customer signing or balking will materially alter the company’s ability to achieve plan.
On the other hand, higher price points require more skilled, more expensive salespeople. Called field sales or outside sales people, their compensation starts at about $250k per year for on-target earnings (OTE - combination of salary and sales commission). Outside sales teams chase larger accounts, and may close 1-3 per year. But if all of them go sideways, the company’s revenues for the year will suffer materially.
Many SaaS startups launch with monthly pricing which encourages customers to try the product and engenders demand. At some point, most SaaS startups switch to annual contracts for three reasons. First, revenue becomes much more predictable. Second, annual contracts often include terms that require pre-payment up-front which rewards the startup with lots of cash to grow faster. Third, contracts mitigate churn rates because the customer is only making a renewal decision once per year, instead of 12x per year. Employing contracts can materially improve a startup’s cash position, unit economics and predictability.
Your pricing plan has to enable the company to become profitable at some point. The value of your business is the discounted sum of all its future profits. Adopting a lower price point may increase sales velocity, create lots of demand, and keep sales teams happy, but if the price point doesn’t generate enough gross margin to achieve reasonably quick payback periods, and the business suffers from an increase in churn, the company is in trouble.
Just a quick reminder:
Payback Period = Cost of Customer Acquisition/Gross Margin
The gross margin is the revenue per customer minus the costs to provide the service. A decrease in price reduces gross margin and will consequently increase payback period.
Compared to software companies, grocery stores are terrible customers because grocery stores have single digit margins. The margin structure of your startup’s customers matters a great deal when setting pricing. All of your product’s cost must be paid from your customers margin. The more margin your customer has, the more they can pay you for your product.
If your customers demand predictable bills, then per seat pricing is the way to go. The question is do they prefer it or do they demand it? Most customers will prefer predictability, but won’t necessarily demand it. Would they switch if the pricing weren’t predictable? That’s a question worth examining in your pricing research.
If you would like to create switching costs, per seat pricing with annual contracts establishes some lock-in. Usage pricing provides more flexibility to customers to try alternatives.
Usage-based pricing (UBP) or activity-based pricing (ABP) has emerged as a dominant model for many of today’s most successful software companies. As Lee Kirkpatrick, former CFO of Twilio, noted during Office Hours, “Twilio was one of the pioneers of usage-based pricing.” The company grew from $15M in ARR to more than $1B with this model, consistently achieving better than 130% net dollar retention.
Aligns vendor success with customer success: When a customer like Uber grows, their usage naturally expands, benefiting both parties. This creates genuine mutual interest in the customer’s growth.
Reduces adoption friction: Individual developers or small teams can begin using the product with minimal financial commitment, making it easier to start a relationship that can expand over time.
Manages cost structure: For companies with significant COGS (cost of goods sold), UBP allows better management of gross profit by passing through costs when appropriate.
Enables natural expansion: MongoDB and Ethereum, two database companies with nearly identical revenue trajectories through 2020, both employ usage-based models that have supported their explosive growth.
Lubricates the conversion funnel: Prospects can sign up and grow their accounts seamlessly. Usage data feeds product-led growth (PLG) lead scores, enabling account executives to outbound to the most promising users. As customers’ needs evolve, they can expand naturally without friction.
While UBP offers many advantages, it does come with tradeoffs:
Complicates churn measurement: If a customer uses your product intermittently (every third month, for example), standard monthly churn calculations will show the account churning and reactivating, skewing your metrics.
Challenges sales compensation: Building effective compensation plans for sales teams is more difficult because the value of an account can’t be fully measured at the point of sale. Teams must reinvent their GTM strategy with new quota structures, sales materials, and margin calculations.
Makes capacity planning harder: With less visibility into maximum usage requirements, engineering teams may struggle to provision infrastructure appropriately.
Requires ongoing purchase decisions: Customers must implicitly or explicitly decide to continue using the product each billing period, rather than making a one-time annual commitment.
Customer frustration with estimation: Customers may struggle to estimate how much of a product they’ll use and experience surprise from overage charges. This creates anxiety in the purchasing process that doesn’t exist with more predictable seat-based models.
Longer sales cycles: Recent data shows usage-based pricing models experienced 29% longer sales cycles in 2023 compared to 21% for seat-based companies. Enterprise-focused companies with usage-based pricing bore the greatest increase at 44%.
Many successful companies begin with pure UBP and then evolve their pricing models over time. As Twilio demonstrated, even usage-based companies can create predictable revenue streams by implementing annual contracts with committed usage levels, with overages billed at higher rates (a two-part tariff).
Amazon Web Services exemplifies this hybrid approach with a “spot market” for instances charged via usage alongside a “reserved instance” market where capacity can be pre-purchased for discounts of about 50%. Similarly, Salesforce began with a usage-based approach before shifting to annual seat contracts when churn rates became significant and revenue predictability faltered.
One powerful strategy for usage-based pricing is deliberate underselling. As Lee Kirkpatrick shared, Twilio account executives would intentionally undersize initial contract commitments to:
While this approach trades smaller initial deals for long-term growth, it creates healthier customer relationships and more efficient expansion opportunities. With this model, Twilio maintained contracted revenue at less than 50% of ARR while achieving industry-leading retention metrics.
When selecting a usage-based pricing model, ask these three questions:
Application software companies typically sell seats. Infrastructure companies sell API calls, licenses per core or host, SMSs, bandwidth, storage by the GB. Switching from the norm in your category introduces friction.
Most application software companies don’t sell via UBP. Slack is a notable exception. Selling constant seat counts stems from the perception that the number of people using software shouldn’t change much from one month to the next. For most application software, the predictability of fixed costs outweighs the benefits of flexibility.
Infrastructure usage, however, can vary widely depending on:
Selling UBP to a buyer accustomed to buying a flat seat count introduces friction into the sales process. This effort may not be worth it unless your company’s strategy is specifically to differentiate on price structure.
| Company | Product | Unit | Pricing |
|---|---|---|---|
| AppDynamics | APM | CPU Core | $6 / core / month |
| ScoutAPM | APM | API Call | $1 / API call / month |
| Lightstep | APM | Service | $85 / service / month |
| Instana | APM | Host | $75 / host / month |
| Splunk | APM | Host | $55 / host / month |
| DataDog | APM | Host | $31 / host / month |
Even within the same category (Application Performance Monitoring), companies use different units for their usage-based pricing. This diversity can be an advantage: it makes it harder for customers to directly compare prices. How many API calls per host or services per host equal $31 per host per month? The difficulty in comparison potentially reduces price competition.
However, it might also confuse customers who are accustomed to buying the service in a different way. Consider whether your startup is differentiating on pricing to compete with an incumbent, or if you’re selling a superior product at a premium, in which case using the same pricing model with higher fees reinforces your brand positioning.
The goal of UBP is to align the cost of software with the value. The unit of pricing is crucial to unlocking that alignment.
The unit must be:
How much should a Fortune 500 bank pay for your startup? How about a 50 person SaaS company? The pricing scheme needs to satisfy these boundary conditions.
Often, a straight UBP pricing model doesn’t scale into the enterprise. A Fortune 500 company may not consume enough units to justify a $250k or $2M deal. To remedy this challenge, consider introducing pricing layers:
Another approach is adding a platform fee to create a two-part tariff. The platform fee instantly boosts the annual contract value and can be tailored per customer segment.
Some customers fear the sticker shock of dramatic usage in the first billing period. To offset this risk, many sales teams cap the charge in the first billing period to ensure customers who sign up and use substantially more of a service don’t experience bill shock. This approach builds trust and gives customers time to adjust to the usage-based model.
There are many companies who employ a two-part tariff: a base platform fee and an ongoing usage fee to capture positive aspects of both types of pricing strategies. Segment is a good example of this. The platform fee establishes a stable relationship and the usage pricing enables the customer to scale up or down as a function of their traffic which might vary throughout the year.
Modern behavioral economics points toward three-part tariffs as potentially the optimal structure, especially when the number of vendors in a category is small. In a three-part tariff, the software has a base platform fee, but the fee includes a certain amount of usage for free, and each additional unit of usage costs extra.
For example, the software might have a base platform fee of $25,000 because it includes the first 150k events for free. Each marginal event costs $0.15.
Research suggests 3PTs capture more value because customers tend to buy larger plans than they might need. Customers who switch to a three-part tariff increased their usage by 15.1% on average, while those who remained on a two-part tariff increased usage by only 0.9%.
Startups struggle to set the right price for their products because pricing dynamics in the field don’t obey the laws taught in the classroom. The standard supply and demand curves imply that as price increases demand decreases, but this isn’t always the case.
Veblen goods defy traditional pricing theory. Demand for Veblen goods increases as prices rise. This behavior is commonly observed with luxury goods, but it also manifests in SaaS sales processes, particularly for enterprise customers.
Bill Macaitis, the former CMO of Zendesk, described Veblen goods behavior when Zendesk began to address enterprise customers. The product marketing team initially charged a modest premium for the enterprise product, but demand was immaterial. As they experimented with other price points, the team discovered demand surged as the price ballooned. Today, the Zendesk enterprise plans cost 10x as much as standard plans.
Enterprise buyers often equate price with quality. At a very small price point, they ask: Since the product is so inexpensive, is it a toy or true enterprise solution?
Performance pricing means explicitly pricing a product in terms of the customers’ revenue gained or cost reduced from its use. Conceptually, performance pricing is very rational. The buyer should be willing to pay between 10 to 15% of the revenue or cost savings for the use of the product.
But performance pricing has three significant challenges:
It cedes pricing power to the customer. Each year when the contract comes up for renewal, the customer will ask, “What have you done for me this year?” If the SaaS startup cannot continuously improve the performance for the customer, the customer is bound to churn.
It commodifies the category by reinforcing a single dominant purchasing parameter: performance. Vendors will all compete on percent improvement of the key metric, leading to discounting and price erosion.
Sales teams lose leverage. If the only metric that matters is performance, then great account executives won’t be able to shine. Building a relationship won’t be valued in the category, or at least it’s not enough to overcome sub-par performance.
Economic conditions significantly impact the effectiveness of different pricing models. During economic contractions, sales cycles tend to lengthen, particularly for usage-based pricing models.
Recent data from 2023 shows:
These statistics suggest that during economic uncertainty, predictable pricing models become more attractive to buyers who need to carefully manage budgets. Companies may need to adapt their pricing strategies during these periods, potentially by:
In a world where AI agents are 2.5-3x as productive as humans, software pricing will need to evolve. The traditional SaaS business model of annual prepaid contracts based on seats faces challenges when a human is no longer operating the software.
There are a few alternatives for AI-driven products:
Triple the per seat price: If the AI agent is 3x as productive as a human, the software company could charge 3x as much per seat. This would be a significant increase in price, but the value of the software would be much higher.
Move to usage-based pricing: AI software might be priced like databases, charging for compute. This aligns value well but may inject unpredictability into the pricing model.
Pay for performance: Some AI companies are exploring charging for outcomes. If an AI agent replaces a role that is compensated for specific outcomes, then pricing could align with those outcomes.
Looking at current AI pricing trends in SaaS, there’s significant variance. Google charges more for their AI features than the base seat, while Loom charges about a 33% premium. The ratio of AI price to base price ranges from 0.32 to 1.11 across major SaaS vendors. This variance indicates that the market is still determining the optimal pricing approach for AI capabilities.
Complex or unintuitive pricing model. A good pricing model appears simple and logical to the customer. It may be complex behind the scenes, but the tax should align itself to the customer’s perception of ROI clearly.
Moving to annual prepay too late. Annual prepay generates cash flow to accelerate growth. Customers effectively lend the startup money to grow. Push for it as early as possible.
Employing static pricing. Price demand curves aren’t static. They change with time as your marketing team builds a brand, develops reference customers, and creates ROI case studies.
Failing to embed concessions in the proposal. When selling to mid-market companies, structure proposals with the expectation that procurement teams will negotiate lower prices.
Using the wrong price discovery questions. When asking customers about pricing, focus on relative price rather than absolute price. How much are they willing to pay compared to another product?
Like product development, developing pricing is a never-ending exercise. A startup’s environment, its product and its positioning change with time, and price must evolve in tandem. The best way for a startup to ensure its price is reasonably optimal is to create a framework for evaluating price and revisit the data a few times per year.
To end this conversation on pricing, I’ll quote Lawrence Steinmetz who wrote a book on sales called How to Sell at Margins Higher Than Your Competitors: “The first thing you have to understand is the selling price is a function of your ability to sell and nothing else. What’s the difference between an $8,000 Rolex and a $40 Seiko watch? The Seiko is a better time piece. It’s far more accurate. The difference is your ability to sell.”
Sales, marketing, product and pricing, when aligned, create powerful branding and margin-expansion effects.
2025-02-25 08:00:00
Chegg filed suit against Google for changes in their algorithm forcing the company to consider a sale.
They allege the Google AI Overviews feature displays Chegg’s AI-enabled Q&A homework helper. This suit stands as the first of its kind challenging Google for changing search patterns, but it won’t be the last.
The data tells a stark story. Looking at Chegg’s traffic using SEMRush analytics, their organic traffic has dropped from 5.6 million to 3.7 million visitors in months—a 34% decline.
Organic keyword volume has shrunk from 11.1 million keywords to 3.5 million—a 68% collapse. This means for millions of student queries where Chegg once appeared, Google’s AI intercepts that traffic.
Is this an isolated case or part of a broader shift?
To find out, I sampled a few other companies. First: HubSpot. The pattern is similar,even though it’s a very different product. B2B SaaS vs B2C content.
Stack Overflow —the internet’s premier knowledge base for programmers is seeing the same downward trajectory. When AI can directly answer coding questions where developers work, why visit a webpage?
Not all content publishers are suffering. The New York Times shows a completely opposite pattern. Teir traffic continues to surge upward and to the right in this same timeframe.
What explains this stark contrast? Perhaps a strategic arrangement? The Times signed a $100 million licensing deal with Google in 2023.
As AI agents crawl and digest the internet, we witness the birth of a new economic hierarchy.
Content publishers face an existential choice: forge partnerships with AI platforms, pivot their business models, or watch their traffic disappear into agentic answers.
