Measuring What Matters: A Mid-Year AI Value Assessment

If your board asked you tomorrow which AI initiatives are genuinely changing the business, how confidently could you answer? 

That question has come up repeatedly in Griffiths Waite’s conversations with enterprise leaders over the last six months, from the AI Summit to our roundtables and workshops. After months of pilots and experimentation, the focus is changing, from “what are we doing with AI” to “what is AI delivering?” 

As we reach the mid-point of this year, it’s an ideal opportunity to take stock of where you are in that transition. Are you still deep in the experimentation stage, creating pilots but not really seeing a return? Or are you focusing your efforts and investment on tools that are ready to deliver tangible value?

Ahead of our next AI breakfast roundtable, our Chief AI Officer Mark Simpson, looks back on the last six months of in-depth AI conversations, and gives his insight into the lessons learned by the most successful AI innovators: lessons that every business leader should prioritise between now and the end of this year. 

Five things to review from the first half of the year 

1. Do you know where your AI investment is going? 

AI experimentation has spread quickly, often faster than leadership visibility. Teams are onboarding tools and building agents around their own challenges, creating useful learning but also ‘tool sprawl’: lots of fairly basic apps with limited potential to grow.  

Start by auditing what already exists. Where are teams seeing results, where is effort being duplicated, and could something that already exists solve a wider problem? The bigger risk is not just license cost, but valuable learning staying isolated. 

2. Are you mistaking productivity for business value? 

Saving time does not automatically create value. We regularly see business cases that multiply small productivity gains across thousands of employees to produce a significant theoretical return, but that value only exists if the saved capacity improves something the business cares about.  

More AI usage is activity; completing a task faster is productivity. Productivity creates capacity, but you must convert that capacity into value by filling it with something that the business really cares about, whether that is conversion, customer experience, quality, risk or operational performance.

The question shouldn’t be, “where can AI save us time?” It should be, “If AI succeeds, which business metrics should improve?” 

3. Do your pilots prove what you think they prove? 

An impressive demo can prove that an idea is technically possible without proving that it can deliver business value. Prototypes built around synthetic data or controlled environments may not test live data, integration or production conditions.  

There is nothing wrong with a technical proof of concept, but be clear about what it has actually proved. Once it becomes a business pilot, it needs a measurable outcome, an owner and a credible path from concept to a live application.

4. Are you trying to make all your data AI-ready at once? 

Data readiness does not have to mean fixing the entire data estate before you move forward. Start with the application and ask what it needs to know, which information must be authoritative and where gaps can be handled safely.  

A transaction needs reliable source data; an assistant providing context may be able to work with less complete information. This allows you to work in thin slices, strengthening the data and knowledge needed for valuable applications rather than waiting for every data problem to be resolved.

5. Are you scaling AI, or just scaling access to it? 

Giving 1,000 people access to AI does not make the organisation intelligent. The value comes when those learnings start to accumulate, for instance when an agent can be extended by another team, new knowledge can be added, and successful ideas can solve more than one problem. To do this, you need to establish who owns that knowledge, how its quality will be measured and how it will remain current. 

Change the way you think about scale. More users is one measure, but so is increasing the usefulness of an application across the organisation. If successful agents and knowledge remain isolated, you have scaled access rather than capability. 

To summarise: you scale access by buying licenses; you scale intelligence by accumulating knowledge, and you scale value by applying that intelligence with guidance, integrations and reuse.

The next six months: Five priorities that can't wait 

By conducting the mid-year review above, you should have a stronger idea of where AI is producing results, where activity has stalled and which applications have proven their value and can be scaled up.  

The next six months should be about concentrating effort on those opportunities. Here are five priorities to guide you to the end of this year.

1. Back the pilots that are worth scaling 

Not every successful pilot deserves further investment. Look beyond the initial demonstration and ask whether it solves a real business problem, can work securely with live systems, still makes financial sense at scale and can maintain its quality over time. These should become explicit investment gates applied consistently across the AI portfolio, with a clear decision to scale, strengthen or stop each initiative. 

We have seen applications perform well initially and then deteriorate because testing stopped with the pilot. The strongest candidates, that drive the best business performance, are the ones you can continue to evaluate and improve once they leave the controlled environment of a trial. 

2. Move AI into the process, not just alongside it 

AI can do more than make individual tasks faster. We are seeing document-processing applications move from extracting information to interpreting it, routing it and guiding what happens next. Agent-based assistants, meanwhile, are moving from answering questions to planning and coordinating multi-step work. These agents are using context, enterprise knowledge, business rules and tools to take action across systems, then involving people when judgement or approval is required. 

That is particularly important for customer-facing AI, where a natural interface can hide a very limited capability. Look at where AI could genuinely improve the process, rather than simply making one existing step faster. 

3. Get control of the costs before usage outruns value 

As AI usage grows, you need to know what it costs and what return it creates. More prompts, tokens or agent use may show adoption, but without the right measures they can simply increase spend.  

Success should still be judged by the outcome you are trying to improve. For a software team, better code quality, faster feedback loops or fewer defects tell you more about value than increased use of coding agents. The aim is to understand the relationship between cost and value well enough to know where greater investment is justified, and where tools need to be retired.

4. Build for change, not today’s model

Some early AI adopters may already be falling behind because they built too tightly around particular tools, models or platforms. AI capabilities have changed rapidly, and they will keep changing, so assume you will eventually upgrade or switch them out.  

The enduring investment is in the foundations beneath them: your enterprise knowledge, security, guardrails, integrations, evaluation capacity and business processes. Build those so you can take advantage of better models as they emerge without rebuilding the entire solution. The aim is not to predict which technology will win tomorrow, it is to create an architecture that allows you to change direction without losing the value you have already built.

5. Use AI to amplify your people’s expertise

If AI creates capacity, think about how people can use that time to apply more of their knowledge, judgement and experience, but do not stop there. Their expertise should also help shape how the AI works: the knowledge it draws on, the decisions it supports, the boundaries it operates within and the situations where human judgement must take over. 

GenAI gives business and technology teams much more common ground, and we have seen commercial, finance, product, marketing and engineering teams develop ideas together in our workshops. Bring those perspectives together earlier, so applications are shaped by the people who understand both the business problem and the technology. The goal is not simply to give experts better tools. It is to create AI capabilities that capture, extend and continuously benefit from their expertise. 

From activity to enterprise value 

Not every promising pilot is ready to scale. Before you commit more time and budget, use our seven-point infographic to test whether yours has the right foundations to deliver enterprise value by the end of this year. 

Download: Is your AI pilot ready to scale - 7 attributes to look out for.

If you’re looking for an opportunity to have open, honest conversations with other enterprise leaders about AI to help steer your priorities over the next six months, join us for one of our popular AI breakfast roundtables. You can find out more about upcoming events here.