The Profit Elevator
Posts
AI is About Proof of Value

AI is About Proof of Value

The easier it is to generate a proof of concept, the quicker the pivot to demonstrating value.

Simon Maughan
July 29, 2025

Velocity to Value

There are two ways to attract attention writing about AI. The first is to treat it as a cure-all and retell grand stories of ground breaking successes. The second is to dismiss it as toy for tech-nerds and imply it’s dangerous for normal people. Neither approach fits with my experience working with businesses.

There is a big difference between running a business backed by your own capital, or that of close contacts, and a startup seeking outside capital almost straightaway. A lot of the advice to startups is about gaining traction. This revolves around experimentation, failing fast, being slow to hire and quick to fire, and other cultural norms that have become part of Silicon Valley folklore.

AI’s superpower is speed, which it makes it ideal for all of the things the advice says a startup should do. This does not work for the majority of businesses that are making it without endless cash to burn.

The number of people using AI is exploding. The number using it effectively is not. This does not mean that AI will go away. There is far too much invested in its success and there is genuine improvement in its capabilities month-by-month. It does mean, however, that the right metric to measure AI is not the volume of experimentation, but what we might call the velocity to value.

A Simple Way to Prioritise Ideas

The value of a process to your business depends on the time saved or revenue earned, and the number of times the process happens. An idea that saves ten minutes matters if it occurs 500 times a week.

Then you must consider whether staff will adopt the process. An idea that transforms how they work may save a fortune on paper, but deliver little because people are overwhelmed. The final consideration is complexity. A brilliant new process that is complicated to introduce is a significant risk to a business.

Ask these questions before embarking on a new project:

What’s the measurable outcome? Hours, margin, error rate, conversion, NPS?
Who owns the process today and tomorrow? Name them.
Which data do we need and is it accessible/clean/legal to use?
How will frontline staff be trained and supported?
How will we evaluate performance? Precision, task success rate, cost per task?
What’s the risk profile? Regulatory, reputational or operational and how do we mitigate?
What does putting into production mean? Uptime, monitoring, incident response?

If you can’t answer these questions on a single page, you’re not ready.

Augment or Automate

Not every task should be automated. Sometimes the best move is to augment. This means giving people leverage rather than replacing them. Ask:

Is the task high stakes? (legal filings vs. internal memos)
Is the output easily verifiable? (spreadsheet formulas vs. strategic advice)
Is context fluid or fixed? (policies change weekly, stock keeping units do not)
What’s the cost of a wrong answer?

Aim for “human in the loop” by default, “human on the loop” for stable, low-risk tasks, and “no human” only when automation is safer and cheaper.

The Importance of Culture

At MSBC, we spend a lot of time thinking about the right model to deploy for a project. Nonetheless, the biggest predictor of AI success is the culture of the client. These are the signs we look for when deciding whether to work with a business.

The most important element is top down support. Leadership must back the plan with a budget for tools and training, otherwise a project is a non-starter.

Then the culture must allow teams to fail. If an idea must to work to justify the expense then don’t try it. Avoid the alternative of spending on projects that don’t make a meaningful difference.

Thirdly, ensure that operations and compliance are aligned. Involve everyone from day one.

When you get these three things right you’ll see AI ideas bubble up from the frontline. Get them wrong and you’ll burn cash with nothing to show for it. There are several coping strategies that we watch out for.

“We’ll get an intern to try GPT.” This is the summertime special. It can be great for learning and useless for sustainable value. The enthusiasm dies when the intern leaves.

“Let’s integrate every tool we use.” There will be a lot of AI use that can stay at the personal level. Training people to prompt LLMs does not mean they must automate their work. Just make sure that access to data is controlled behind the scenes.

“Legal will look at it later.” If you are not planning for success then you won’t deliver it.

“We need a custom model.” 99% of small businesses use cases don’t. Spend that budget on the right tool and proper training.

Tailored For Any Budget

The majority of small businesses we work with start out with a significant fear of failure. They are not used to dashing off experiments and seeing which ones work. They have too much respect for cashflow and are a long way from Silicon Valley.

What these businesses do share is a desire to improve. This might be due to fear of missing out, or a desire to leapfrog the competition. Most often though it’s the result of the same curiosity that lead them to start the business in the first place.

These businesses are led by people who make the time for change. Once we identify that then the solution can be tailored for any budget.