Choose Reference Customers: Turn Down Orders That Do Not Build Product Capabilities

Choosing Reference Customers The first lesson in customer acquisition is selection: 100 users who genuinely believe in you are worth more than 10,000 who find you merely acceptable. In B2B, their equivalent is the reference customer. Beyond revenue, these customers send a signal to the entire industry. Their strategic value can multiply for three main reasons. 1. Reference customers are the strongest sales asset. Enterprise decision chains are long and risk aversion high, making peer endorsement the shortest route to persuasion. Legal AI company Harvey illustrates this. In February 2023, its first announced major customer was Allen & Overy, or A&O, a leading global law firm with 3,500 lawyers and 43 offices. PwC and Cleary Gottlieb Steen & Hamilton followed. Legal services place great weight on peer endorsement: winning A&O opened the door to top global firms. Palantir followed a similar early path. The CIA was demanding but extraordinarily valuable as an endorsement; trust built in intelligence later opened doors on Wall Street, in manufacturing, and in government projects. Harvey moved from a reference customer toward industry-wide adoption. In November 2022, A&O created a Markets Innovation Group led by partner David Wakeling to quietly test the then-obscure Harvey. Rather than watch a demonstration, it tested the product across the firm. By the pilot’s end, 3,500 lawyers had submitted around 40,000 real business requests across 250 practice areas and 50 languages. Wakeling’s assessment was widely quoted in legal circles: ‘In fifteen years at the forefront of legal technology, I have never seen an innovator like Harvey. It will unlock AI’s enormous potential and transform the legal industry.’ 2. Reference customers determine your product’s DNA. In the forward-deployed engineering (FDE) model, learning in the field feeds back into the product. Your first ten customers effectively help shape it. Choose the wrong ones and the product evolves in a direction with little general applicability. a16z’s first advice to startups, ‘sell smart,’ addresses this. A startup cannot meet every demand. Select customers with common business scenarios and system environments, forming an ideal customer profile (ICP), so experience from each delivery accumulates and can be reused rather than canceling out. 3. Reference-customer quality matters an order of magnitude more than quantity. FDE delivery capacity has a natural ceiling. An elite team can serve only a single-digit number of customers deeply at once. Choosing wrongly costs more than one lost order: an inefficient project can consume the best team for half a year. Check for a clear business owner, access to real business data, and demands that are not grandiose and detached from reality. Add a reference-customer criterion: agree early whether a successful customer will publicly endorse the project through a joint case study, an industry-conference presentation, or hosting prospective customers. The value of a reference customer unwilling to endorse you publicly should be discounted by at least half. Beware of ‘demand locusts.’ Product developers should beware of ‘product locusts’: early users who swarm in, leave after using the product, and misdirect iteration. Their FDE equivalent is the ‘demand locust’: a customer with ample budget and abundant demands that consumes all team resources without producing reusable experience. Demand locusts have three characteristics: • Their requirements clearly diverge from the company’s overall strategy; completing the project builds no reusable capabilities. • They treat the FDE team as cheap outsourced labor, allocating tasks by headcount instead of aligning on business outcomes. • Internal organizational conflicts consume energy; your main job becomes helping one department prove another has problems. Saying no is difficult because these orders are often financially tempting. Yet a misguided pilot costs more than immediate expenses: it consumes time, morale, and the compounding product benefits a good customer could have generated. #Forward-deployed engineering (FDE) ---- Robin: ‘Knowing which projects to pursue and which to decline is equally important. So is choosing which customers to serve. Learn to turn down attractive opportunities that are not yours, say no to customer requests, and select customers. Not every dollar can—or should—be earned.’