Sam Shar on

AI with an accountability check

Proving AI's return, managing its risks, and keeping consultants and vendors honest about what it delivers.

2 newsletters. Updated 30 Oct 2026

Sam's approach

What I do

I run TSCi, an enterprise technology consultancy, and I start AI work with the commercial outcome rather than the tool. Before anything is built, I agree with the CEO or CTO what result counts, what it is worth, and how it will be measured. Every deployment is then tied to named KPIs with a baseline taken before go-live, and TSCi keeps tracking those same KPIs afterward. The people who build it stay to see it used and to train the people using it. The best returns I see come from mundane work such as process automation, routing and reporting.

How I work

An AI ROI expert has to be free to say that a deployment is not paying, which is why I hold TSCi clear of vendor alliances. A consultancy that has signed with the platform it recommends cannot give neutral advice on that platform, and boards should price that in when they read the recommendation. Pilots stall when nobody owns the number after launch, so ownership is assigned before the build starts. Security is reviewed before rollout, because rushed deployments open new holes. I use AI to reduce complex programming tasks and keep it out of the proofing process. Adoption is not the win. Delivery is.

When to bring me in

Bring in an AI ROI expert when spend has gone out and nobody can show what came back, when pilots are not reaching production, or when the firm advising you also sells the technology. This is the time to take stock, not to stop learning. Questions a board should ask before approving an AI program.

  • What commercial outcome does this produce, and what is the baseline today?
  • Who owns that number after go-live, and when is it reported back to us?
  • Does the adviser hold a commercial alliance with the vendor?
  • What new security exposure does this create, and who has reviewed it?
  • Who trains and supports the people using it once the project team leaves?

I am happy to talk any of these through.

Questions about AI with an accountability check

How do you measure the ROI of AI?
Decide what the project should change before you start, such as hours saved, costs cut or revenue won, and measure that from day one. Track it for months, not weeks, because results often fade once the launch team leaves. Count the full cost, including training, integration and security. If nobody can say what an AI project will do for the business, it is an experiment, not an investment.
Why do so many AI projects fail to deliver results?
Most start with the technology instead of the problem. A company buys a tool because everyone else is buying one, then goes looking for somewhere to use it. Nobody defines what success looks like, so the project can't show a return even when it works. Add messy data, thin training and a delivery team that leaves at launch, and the result is a pilot that impresses in a demo and never changes how people work. Tie each project to a measurable result from day one, and keep the builders around until the result shows up.
What security risks come with rushing AI projects?
Speed is when the basics get skipped. Staff paste sensitive data into public tools, systems are connected to company data without clear access rules, and vendors are chosen before anyone has asked where the data goes or who can see it. A rushed project can also create new ways in for attackers, and the team may not know how to spot them. Treat security as part of the design, not a final check. Decide what data the system can touch, who is responsible for it, and what happens when something goes wrong before anything goes live.
Can you trust a consultancy that partners with an AI vendor?
Sometimes, but you should ask direct questions first. A partnership isn't automatically a problem, and many good firms have them. The risk is that a firm that earns more from one vendor will lean toward that vendor, whether or not it fits your business. Ask how the firm is paid, whether it earns anything from the products it recommends, and whether it would tell you to use a different tool or not to use AI at all. An adviser you can trust is happy to answer, and to put the answer in writing.
What should a board ask before approving an AI program?
The board should focus the thinking towards the result, and ask 'what business problem is this solving?' and 'how will we measure it, and when will we know whether it is working?'. Then ask who is accountable, what it will really cost over time, and what data it will use and how that data is protected. Ask what happens if the results don't arrive, and who stays involved after launch to make sure people actually use it. If the answers are vague, the board is being asked to approve a hope rather than a plan.

What Sam is saying

On AI with an accountability check

"The difference between a useful AI programme and an expensive science experiment is not technical brilliance. It is whether it delivers measurable commercial results in a clearly defined period."
Sam SharSam Shar, CEO, Trend-Setters Consulting (TSCi)30 Oct 2026
1 of 6

Get in touch

For press, speaking or working with Sam.