Selling AI to a business that has already been disappointed by it

Most established businesses have already tried something with AI that did not deliver. The sales conversation starts from there, not from enthusiasm. What works is a smaller promise. One process, one measurable outcome, a defined period. Not "we will transform your operations" but "we will handle the first pass on these inbound requests and we will show you the accuracy and the time saved after four weeks". Define the success metric before starting, with the buyer, in their terms — hours saved, response time, error rate, cost per item — and agree how it will be measured. A pilot without an agreed metric ends in a discussion about impressions, which the incumbent way of doing things always wins because it is familiar. Charge for the pilot. Free pilots get deprioritised by everyone including the buyer, and a paid one has an owner on their side. Expect the security questionnaire and have the answers ready: where data is stored, whether it trains models, who has access, which subprocessors are involved. Being able to answer in a day rather than three weeks is a competitive advantage at this stage of the market. And name the failure mode out loud. A vendor who explains what their system does badly, and what the human check catches, is more credible than one who claims it does not happen — particularly to a buyer who has watched a previous vendor claim exactly that. Then expand from something that worked. A reference inside the same company is worth more than any deck. This article is general information, not legal, tax or financial advice. Rules change and every deal is different — check your own case with a licensed professional. Alberto Zaltzberg — Adonait · adonait.com