Paid Pilots with Pre-Agreed Exit Conditions

Sierra

L3

Sierra's design partner program was not a validation exercise — it was the commercial launch in disguise. Between November 2023 and February 2024, Logan Randolph ran a 4-stage intake process for each of six design partners. Stage 1 was a 30-minute discovery call with no selling. Stage 2 was a 30-minute live demo. Stage 3 was security and technical deep dives. Stage 4 was a verbal commitment and a paid contract requiring 10–20% of total contract value upfront. Companies that could not produce this payment were screened out before any engineering work began. The kickoff structure enforced the six components of a well-structured pilot. Duration was 2–6 months, defined at kickoff. Scope was limited to 3–4 specific business problems, with swimlanes established so that success was attributable to specific work. The primary success metric was agreed in the kickoff session — not a list of hoped-for outcomes but a concrete measurement (containment rate, CSAT score, cost per contact). The baseline was the current state before Sierra agents went live. Pricing was paid before the pilot started. And throughout the pilot period, human agents remained available for every escalation — removing adoption risk before the full commitment. The result was 100% conversion from design partner to paying customer across all six. WeightWatchers achieved 70% containment and 4.5/5 CSAT in the first week. SiriusXM used the partnership to build Harmony, a subscription-retention agent. Randolph's explanation of why conversion was total: the upfront payment pre-selected motivated buyers who had already cleared procurement; the co-build process produced a product shaped to each partner's specific problems; and the implementation was happening in production throughout, so conversion merely formalized an ongoing commercial relationship. The design partner structure became the template for Sierra's post-launch commercial motion. Reggie Marable's "Paid PoC" is the scaled version of the same program — paid from day one, time-boxed, scoped to real-world scenarios. The principle Randolph articulated remains the operational center: "We told partners upfront: 'We'll give you dedicated engineers, direct access to our founders, and our cell phone numbers. But in return, we need real investment from you — payment, access to your systems, and weekly meetings to get candid feedback.'" (Logan Randolph, First Round Capital interview.) The structure is a mutual commitment mechanism, not a risk-reduction tool for the buyer alone.

Gong

L2

Gong's alpha program of 12 companies is the earliest recorded instance of the paid pilot structure working as a conversion engine in this cohort. Eilon Reshef recruited the 12 companies from the founders' personal networks. The product recorded and analyzed sales calls from the first day of use — there was no implementation lag, no integration complexity, no training period before value appeared. The value was tangible immediately. Eilon's PMF signal during the trial was precise: "9 out of 10 complaints were how come you didn't even record this call?" Users were angry when a call was not captured, meaning the product had already become essential before the trial ended. This is the structural ideal for a paid pilot: the pilot generates dependency, not just evaluation data. When Amit executed the trial close — directly asking each company to pay — the product had already done the selling. The AE manages a commitment that the product has already earned. 11 of 12 converted. The 12th eventually paid as well. The conversion rate is a PMF signal as much as a sales signal: the product was self-evidently valuable to anyone in the ICP. A conventional free trial would have generated the same PMF data. The paid structure meant the commercial pathway was already open when the trial ended — no renegotiation, no procurement restart, no delay. Jameson Yung's subsequent decision to restore the pilot motion — after Gong had moved away from pilots as the company scaled — confirms how structurally important it was. Yung's observation was simple: users who touched the product "went crazy for it." The pilot was not a tactic for handling a skeptical prospect. It was the mechanism for delivering product conviction. Without the pilot, the company was asking buyers to commit based on a demo; with the pilot, buyers committed based on their own direct experience of the product's value in their actual sales environment.

Glean

L2

Glean's proof-of-concept structure evolved from early free trials into a paid engagement that the company now runs with explicitly defined success criteria anchored before kickoff. The evolution was deliberate: Arvind Jain observed that creating an amazing demo and selling something you cannot yet deliver destroys credibility with the exact buyers you most need. The paid POC structure disciplines both sides — the customer has skin in the game, and Glean has a defined success threshold to hit before conversion is expected. The mechanics of the 90-day paid POC are designed to generate irrefutable adoption data. The primary success metric is an 80% adoption rate within 90 days — not user satisfaction scores, not executive sponsor enthusiasm, but measurable daily active usage as a fraction of licensed users. Search generates this data automatically: query frequency, DAU/MAU ratios, and search success rate (queries where the user clicked a result) are tracked in real time. The CSM brings this data to the executive sponsor meeting, converting the expansion conversation from a vendor pitch into a review of the customer's own usage evidence. The KPI anchoring at POC kickoff removes the most common conversion failure: the post-pilot "we need to think about this" response. When success criteria are defined before the pilot begins, the conversion decision is already made at kickoff — you either hit the metric or you don't, and if you hit it, conversion is the logical next step. Glean's deployment of the permissions model from day one of the POC neutralizes the CISO objection — the security review happens inside the pilot, not as a gate after it. The scale outcomes confirm the structure works. Glean ran 40 design partners before public launch, all converting. The standard expansion trajectory — $60K departmental pilot to $300–500K+ company-wide within 9 months — became repeatable enough to become the primary revenue engine. T-Mobile deployed 100,000 seats; Booking.com deployed company-wide across 14,000 employees. The $1M+ contract segment grew 3x in FY2025, driven by accounts that had completed the POC cycle and expanded through the land-and- expand motion. None of these outcomes are achievable through a free pilot that generates evaluation data without commercial gravity.

Harvey

L2

Harvey's pilot structure is distinctive in this cohort because the primary success mechanism is not a quantitative metric but a qualitative one: does a managing partner at a Vault 50 firm trust the output on their own actual work? The Allen & Overy pilot of 3,500 lawyers running 40,000 queries was not structured around a deflection rate — it was structured around building institutional trust in a buyer community that evaluates technology through the lens of professional liability. The months-long duration and the scale of the pilot were the signal: Harvey was willing to invest before the contract was signed. The PACER demo tactic illustrates the hyper-personalization principle at the individual level. Before any partner meeting, the team downloaded the firm's most recent public federal court filings and rebuilt the demo around those specific documents. Weinberg ran prompts designed to critique the partner's own work: "I would basically download the last thing that they submitted to court. And then I would try to come up with prompts that were like, 'This is bad.'" When the model was right, the partner stopped evaluating and started planning which workflows to migrate. This is the pilot logic applied to a demo: the prospect's own work is the test case, not sample data. Enterprise pilots ran multi-week (8–16+ weeks) with Legal Engineers — former Big Law attorneys from White & Case, Latham, Skadden — embedded with the firm's team throughout. The human approval gate was preserved throughout the pilot: every Harvey output required attorney review before being used in a filing or client communication. This removed adoption risk at the most critical point — a managing partner signing off on a pilot is not betting their malpractice exposure on an untested system; they are evaluating a tool where every output is reviewed before it matters. The conversion evidence is consistent with near-total pilot success. Harvey's first 50 enterprise customers were all referrals — a structure that only works if pilots consistently convert. Gross revenue retention of 98% confirms that customers who committed based on pilot experience did not change their minds at renewal. Seat utilization grew from 40% to 70% in 2024, meaning existing customers were finding more use cases within Harvey's workflow surface during the post-pilot deployment period.

Decagon

L3

Decagon's 4-week pilot is the most precisely structured commercial mechanism in this cohort. Every component is defined before the pilot begins. Duration is fixed at four weeks — not open-ended, not subject to extension by default. Pricing is agreed before the pilot starts; there is no post-pilot renegotiation regardless of outcomes. Success metrics — deflection rate and CSAT — are defined at kickoff, not surfaced after the pilot to retroactively justify a decision already made. A baseline is established (current deflection rate, current CSAT, current tickets-per-agent) so that the improvement is measurable against something real. Human fallback is always available throughout the pilot; the AI runs a portion of traffic, not all of it. And bespoke API integrations are mock-built before the pilot starts, so the prospect experiences a tailored workflow from day one rather than a generic demo. The structural logic of this design is compression. The 4-week limit creates urgency for both parties: Decagon must deploy fast, the customer must engage. Pre-agreed pricing eliminates the post-pilot game where the enterprise buyer evaluates success and then argues the value was lower than expected. Enterprise support metrics are measurable in near-real time — deflection rate and CSAT update daily — so a buy/no-buy decision can genuinely be made on objective evidence at week four. The human fallback means the enterprise CX team does not need to bet the entire support function on an untested system; they need to run a portion of traffic through it and observe results. The case evidence is the most specific in the cohort. Duolingo English Test: a prior vendor had worked for a full year and could not get chat live; Decagon went live in one month and achieved 80% chat deflection immediately. Notion ran a formal five-vendor RFP with explicit criteria; Decagon won and delivered 2x deflection improvement and 34% faster resolution. Rippling deployed with 75+ custom routing tags across 12+ product lines — a complexity level that required Decagon engineering on-site, but produced deflection improvement from 38% to 50%+. In each case, the pre-agreed structure meant that when the pilot data came in, the contract conversation was about confirming an already-made decision, not starting a new negotiation. This is the operational meaning of "conversion as the expected default."