“Death of a Salesforce”: Why AI Will Transform the Next Generation of Sales Tech

The battle between every startup and incumbent comes down to whether the startup gets distribution before the incumbent gets innovation. In sales tech, it’s easy to assume incumbents like Salesforce and Hubspot have the edge. First, they are embedded as “systems of record,” so sales leaders are loath to rip them out and replace them. Second, these incumbents (and their well-entrenched peers) are not sitting out the AI revolution; conscious of protecting their competitive moats, they are rapidly adding AI features to stay relevant.  

We believe AI will so fundamentally reimagine the core system of record and the sales workflows that no incumbent is safe. 

Instead of a text-based database, the core of the next sales platform will be multi-modal (text, image, voice, video), containing every customer insight from across the company. An AI-native platform will be able to extract more insight from a customer and their mindset than we could ever piece together with the tools we have today.  

Sales workflows will fundamentally change. With AI, sales teams will no longer need to spend endless hours researching new leads or prepping for calls — AI will be able to do it in seconds. Reps won’t have to suss out the readiness of potential customers because AI will have automatically compiled a ranked list of primed buyers, and will keep it constantly updated. Need personalized marketing collateral for a deal? Your AI wingman will produce whatever assets you need and feed you live tips while you’re on a call to help you close. 

In this post, we dive into how AI will change selling as we know it, which early companies are leading the way, and the implications on the broader industry.  

A new system of record

While incumbents often adapt to new platform shifts, they are rarely able to completely rethink their architecture. Salesforce (founded in 1999) and Hubspot (founded in 2006) were enabled first by the arrival of the relational database, and later, of course, the cloud. The core of these companies’ foundation is a structured representation of sales opportunities, in rows and columns, with their related criteria in text.    

Over time, with the proliferation of point solutions, data became siloed in discrete activities and tools along the sales funnel. There’s no complete view of what’s happening, end-to-end, across the sales process. A solution that adds personalization to increase conversion at the top-of-funnel has no data on whether that personalized touch ultimately increases the close rate. In this 2019 podcast with People.ai, we discussed the opportunity to aggregate this sprawl of data, and the potential for a unified go-to-market data model to streamline workflows.

With LLMs, the core of the next sales platform could be entirely unstructured and multimodal, including text, image, voice, and video. A company’s sales platform could include data about existing and prospective customers from countless sources: recordings and transcripts from any conversation with someone at the company, emails and Slack messages, sales enablement materials, product usage, customer support activity, public news, financial reports…the list is endless. Furthermore, the LLM powering the platform would be constantly ingesting data to create the most up-to-date context. 

Redefining workflows

With this data infrastructure, common sales activities may be redefined, or even disappear completely. Simultaneously, we’ll likely see seller workflows emerge that simply aren’t possible today.

In essence, the way sellers and buyers interact will be fundamentally different.

Early days of the future

Emerging AI-native sales solutions are not simply AI-powered versions of existing categories. Instead, they are enabling new proactive sales motions and evolving to serve multiple use cases. As a result, point solutions in seemingly adjacent spaces are overlapping with each other more so than ever. Traditionally, for example, converting inbound website leads and automating outbound campaigns would be considered separate tasks. With AI agents, a tool initially designed for one such task could seamlessly expand to handle both. It won’t be long before an AI agent is capable of growing an organization’s sales pipeline across all channels.

When mapping out the market for AI applications in sales, it’s helpful to define broader categories for types activity:

• Intelligent Pipeline: This is the lifeblood of any sales force, and perhaps the most natural insertion point for AI. There will be many AI-driven approaches to automate pipeline building in the near-term. For example, teams use Clay’s enrichments and AI research agent to prepare high quality lead lists for their sellers to outbound to. Sellers can even use messages personalized by AI instead of drafting their own.
• Digital Workers: The goal for many AI solutions will be to own as much of the end-to-end sales process as possible, taking a lead from qualification to close. For example, teams use 11x today to automate the SDR role end-to-end, which means 11x goes as far as booking meetings with prospects. In the future, the scope may expand to closing the deal entirely. 
• Sales enablement + insights: When working a deal, AI can provide valuable context and information beyond the scope of sales activities. For example, Naro will automatically go through sellers’ emails and surface company documentation that is relevant for responding to questions from buyers.
• CRM + automations: Ultimately, there must still be a way to capture and organize unstructured and multimodal data in a way that is useful to both humans and AI workers. For example, Day joins seller conversations with customers to capture context and constantly refresh its knowledge base. It then summarizes the insights on customer “pages” that are easily digestible by not only the sales team, but also anyone in the company. In the enterprise segment, People.ai’s SalesAI product leverages their unified data model — which can span multiple CRMs if needed — to automate sales activities such as accounting planning and content generation. 

Market impact

Sales, marketing, and customer success will blend together

Today, sales, marketing, and customer success teams often feel siloed, with poor knowledge sharing and rough handoff processes between them. With more comprehensive, shared context and shared insights, go-to-market teams will be more in sync and better able to collaborate with each other. In fact, it’s possible that as all important customer context is reflected in the same source of truth, and as activities are guided by the AI, job functions could start to blend together. Sales and account management and customer success may simply be seen as different ways of adding a human touch to go-to-market. No more fighting over who gets credit for which part of the upsell — you could even imagine a world in which quotas are redesigned to be more team-based than individual rep-based to more accurately reflect the opportunity for fluid collaboration throughout the sales cycle.

Fluid GTM strategies

Another interesting outcome would be more dynamic and flexible approaches to go-to-market within the same company. Today, companies typically decide where to focus resources based on target segments and annual contract value ranges — for example, a top-down sales motion or an inside sales-assist motion. They often hire and build teams around a prescribed strategy. The assumptions around these economics will look very different in an AI-first world. Companies may be able to reorient their resource allocation around what’s best for the customer — to close this account, what’s the best go-to-market approach? This has implications on brand, as well. Today, many companies choose to deliberately characterize themselves as either enterprise-grade or developer-first; in the future, companies should be able to cater to both of these buyer personas with highly customized sales journeys, meaning the parent-level brand can be broader and more all-encompassing. (Said differently, the sales motion of the B2B company of the future may in fact be Everything, Everywhere, All at Once.)

Per seat pricing → outcome-based pricing 

The prevalence of AI-native software companies may be the kiss of death for seat-based pricing, as there is a clear opportunity to more tightly align pricing with value delivered. Our partner Alex Rampell considers Zendesk to be the canonical example here: consider a company that pays 1,000 support agents $75k per year each, with each agent answering 2,000 tickets annually. As is typical today, each agent is armed with a Zendesk license at $115 per seat per month, bringing total annual software spend on customer support to nearly $1.4M. In this scenario, the human cost per ticket — $75M in agent salaries / 2M total tickets — is $37.50, while the software cost per ticket — $1.38M in Zendesk spend / 2M total tickets — is only $0.69. Under the new AI-first paradigm, with everything shifting towards selling outcomes, Zendesk will have a predicament on its hands — what’s the best way to price a successfully resolved ticket?

The question for AI founders to wrestle with, as Gokul Rajaram points out, is which metrics or outcomes are the right ones to serve as the atomic unit for billing. In sales, the spectrum of outcomes from least to most valuable is the generation of unqualified leads (e.g. top of funnel prospect list) to the completely automated closing of a deal (e.g. AI software sells your product with no humans involved). Unqualified leads tend to be cheap as they are not particularly valuable – it’s very hard to assign any probability to whether they will buy your product. To draw parallels to other online business models, charging for unqualified leads would be most akin to cost-per-click advertising. 

On the other end of the spectrum, AI sales companies could charge for the juiciest outcome possible – closing a deal. The monetization model here could look more like that of many online lending marketplaces, which typically charge some take-rate (usually 3-5% of principal) on an originated loan. This model is of course lower-volume, higher bounty – the probability of making it all the way through the funnel to a funded loan (or closed deal) is relatively low, which means the take must be substantial. The interesting consideration in the context of AI sales software is comparing a potential take-rate to that of an AE. Whereas AEs usually earn 10-15% of the annual contract value (ACV) of a deal in the way of a commission (in addition to their salary), an AI sales agent that closes deals entirely on its own could do so for much less, providing an opportunity for immediately obvious ROI. 

There’s of course no “right” answer here — we’re looking forward to seeing which models become most popular with founders (and perhaps more importantly, their customers) in these early days of selling AI-assisted outcomes. As with the status quo today, relative price differentials between selling leads and outcomes will always be a function of how efficient the software is at converting prospects to closed deals.

Conclusion

AI’s potential will not be limited to streamlining the sales activities we have today; instead AI will compel us to reimagine sales processes and workflows completely. The relationship between sellers and buyers will evolve, as will GTM strategies. Hence, the sales software stack in the future will look fundamentally different.

If you’re building toward this future, we’d love to talk to you. Reach out to zyang@a16z.com, mandrusko@a16z.com, and angela@a16z.com.