Last year, conversations with business owners about AI opened with "what can you do". This year the opening line is almost always one of three questions. We tallied the most frequent questions across more than a hundred client conversations in the first half of 2026, and they come down to these three.
Question one: once we spend the money, when do we see a return
Clients used to ask "can it be done"; now they ask "how long until payback". The change happened because more buyers have already been burned. Nobody believes the standard "results in three months" pitch anymore.
Our habit is to do the arithmetic in detail: for a customer service scenario, start from how many inquiries the current team handles per month and what that labor costs, estimate the share AI can absorb, then give a range instead of a single figure. "Six to nine months conservatively, four months if things go well" earns more trust than "results quickly".
Question two: does the data stay with us or with you
Data sovereignty was the most raised concern this year. Clients usually put it plainly: "if my customer records and business data go into your system, how is security guaranteed?"
No script gets around this one. It takes governance and product: encrypted storage, tiered permissions, a private deployment option, and a contract clause stating that data ownership stays with the client. Our contract template devotes a page to data ownership, and once clients read that page the rest of the negotiation goes smoothly.
Question three: after you leave, who maintains it
Two years ago, delivery closed the project. This year clients immediately ask what happens afterwards. Models age, knowledge bases need updating, prompts must follow business changes — an AI project with no operations is abandoned within three months.
So our quotes now include an operations package: post-launch knowledge base maintenance, model performance inspection and monthly review reports. Some clients ask whether we are simply selling a long-term revenue stream. We admit that we are — but the bigger point is that an AI project without operations wastes the client's money and damages our reputation.
Rational procurement rewards teams that do the work
A rational market squeezes out vendors selling concepts and gives more orders to teams that actually deliver. ZClaw's strategy over the past two years has been plain: make one scenario work end to end, then expand. If you are evaluating AI vendors, put these three questions to them exactly as they are — the quality of the answers alone will filter out half the candidates.