Five questions most founder can’t answer about their own company
AUG 12, 2026
AI doesn’t fix broken processes. It makes them faster.
In January 2026, artificial intelligence was cited as the reason for about 7% of announced job cuts across the United States, according to outplacement firm Challenger, Gray & Christmas.
Five months later that share reached 31% and it was the leading stated reason for job cuts for the fourth month running.
That jump does not necessarily mean the technology can now replace entire departments. However, it does mean the rollout was accelerated.
Needless to say, “AI induced layoffs” sounds significantly better than “we hired too many people and demand softened”.
Andy Challenger of the firm has said that naming AI in a layoff announcement can win over investors while pushing employees away, and that this is why the messaging has shifted from hedging toward citing AI directly.
AI is genuinely absorbing work in support, testing, content and routine coding across industries, but it is difficult to know today how many of these layoffs actually came from efficiency gains.
The gap between what is announced and what shows up
In a survey of 1,993 respondents across 105 countries, McKinsey found that some 88% now use AI in at least one part of the business. Only 39% report any impact on profit at the company level (and most of those put it below 5%), and around 6% report that AI accounts for more than 5% of their operating profit.
Adoption is close to universal. Measurable impact on the bottom line is not.
If you are a founder somewhere in the middle of this, the useful question is not whether AI works. The question is whether the way you use it is beneficial in your company — and founders might find this difficult to answer.
Here are five questions you can start with.
1. Does your team know what the AI tools they are using cannot do?
Ask your leadership team what the AI they use cannot handle. If the answers are vague, or if someone believes a subscription could absorb an entire function without any limitations or human intervention, they have not examined the tool or its output in depth.
The gap between what AI tools are marketed as and what they actually do can be significant, and you won’t know until you have tried implementing them properly. Nobody outside the engineering teams that built AI tools has more than a couple of years’ experience with the current generation. Everyone is still learning, which means you have time to catch up, but your organisation needs to understand the limits of AI. Until it does, you cannot tell whether a tool is poorly deployed or is simply being asked to do something it never could.
2. Who decided what to automate — and had they ever actually done the work being automated?
Do you know who wrote your automation list? If it came from a leadership brainstorming session, it likely reflects what leadership imagines the work looks like. However, only the people who actually do a task every day know which parts are repetitive, which are judgement, and which only seem repetitive. Almost every employee has at least one workstream that eats hours per week and would lose nothing in quality if it ran faster.
If the employees were never asked, your company is automating a guess.
3. Count your AI subscriptions. Now count the people who could demonstrate competent use of each.
If the second number is smaller than the first, you are paying for tools rather than building capability.
The follow-on question is harder: name the person accountable for each tool. Not the person who found the tool, onboarded the vendor, or approves the invoice — name the person responsible for tracking successful implementation and letting you know about the results. If you can’t name them, and nobody is in charge of supervising the usage, then nobody will raise their hand when the tool doesn’t yield the results you expected.
This is one of the four areas a Durable Ops audit looks at, Ownership. It is the area founders are most confident is covered, and the one that most often is not.
4. What would prove the deployment worked?
Speed and cost are short term objectives and they are easy to measure. They are also the objectives most likely to be met, while other metrics are quietly deteriorating.
Take the example of Klarna. In February 2024 the company announced that its AI assistant had handled 2.3 million customer conversations in its first month. Spectacular results. Klarna reported that two-thirds of its customer service chats were handled by AI, which it described as the equivalent workload of 700 full-time agents. Their resolution time fell from 11 minutes to under 2 minutes. The company projected a $40 million profit improvement for the year.
However, in May 2025, Klarna’s CEO Sebastian Siemiatkowski told Bloomberg that cost had been “a too predominant evaluation factor” and that the result was lower quality. Klarna began recruiting human agents again for complex cases.
Klarna itself disputes the “reversal” framing — the company has said it remains committed to being AI-first and that the hiring pilot is not a change of strategy. This is not a critique of Klarna. They are taking the AI revolution seriously, as they should. They also stayed flexible on the metrics, adding response quality alongside speed and cost. Their deployment was successful, but they originally didn’t measure everything they needed.
5. If you’ve already cut roles, what date have you set to check?
If you have reduced headcount assuming that AI will absorb the work, there should be a date set in your calendar to verify that assumption.
If there is no date, you have made a bet and not a decision. Rehiring is expensive and it takes time. While you are rebuilding a team, a competitor who did not make that mistake is galloping forward.
What these five have in common
Every question above is a symptom of the same underlying condition — the tool arrived on top of work that nobody had redesigned first.
McKinsey’s finding on this is the most useful number in the whole survey. Among the organisational changes linked to AI success, fundamentally redesigning workflows had one of the strongest links to meaningful business impact — and among the small group of high performers, 55% had redesigned workflows, against roughly 20% of everyone else.
When you drop a tool onto a process that was already unclear about who owns what, the AI will not fix the process; it will run it faster. That is why the number of subscriptions doesn’t convert into capability, and why the headcount decisions are sometimes premature.
Where to start
You may have found that you couldn’t answer one or more questions above about your own company. We can help you find the bottlenecks and redesign your workflows before implementing AI. We start with an audit.
Durable Ops runs one, free, across four areas — Clarity, Ownership, Traceability and Incentives — and tells you which one is actually costing you.
Leave a Reply