How do you ensure ROI on an AI Rollout?

AI is absolutely the biggest buzzword right now, arguably eclipsing earlier hype cycles like “blockchain”, “metaverse”, “cloud” or even “dot-com”. It is being woven into everything from search engines to productivity tools, art, writing, customer service, chip design, food delivery, you name it. You can’t open any news feed or a tech blog without seeing it (how ironic!). Unlike some previous buzzwords, AI isn’t just hype. Tools like ChatGPT and Claude show tangible, everyday utility. That mix of genuine innovation and marketing spin turbocharges the buzz. But every company now claims to be “AI-powered” whether it’s a to-do list app or a toothbrush, and upon detailed reading many a consumer will be left wondering why on earth most of these companies bothered! 

Do you really need a water bottle with an AI hydration coach or a toaster that uses AI to detect when toast is done? Was the spend on creating such products reflected in happier consumers or increased sales? What was the ROI on each of these products? If you’re in a company looking to roll out some form of AI in your products or projects, how can you ensure that it’s worth your time and effort, and genuinely is a positive for your customers and clients?

Clear vision, plainly stated and reiterated frequently

All too often, executives will confidently declare that they want to become an “AI powered organisation”. To much fanfare, they then go and hire a team of suitably technical people to make this a reality. Cue a one to two-year gap with little or no tangible result, before the technical team silently shuffles off in the night, never to be seen again. Sadly, without a real vision on what they are meant to achieve, this team was set up for failure. It is the job of the people in charge to set the vision of the firm, and this should clearly state the purpose of what every project using AI should be setting out to achieve. This cannot simply be “To become AI powered”, it must specify what is going to happen and why. This becomes especially compelling when all employees in the firm can help assist, with their own expertise. Ensure that this vision is stated plainly, and make sure that it is reiterated frequently. 

Vision needs strategy

If you’ve given a clear vision, which should be the north star which your AI projects will aim for, it is essential to outline your strategy, i.e. how do you get to your vision? This then allows for your technical teams to implement the strategy. We updated the key elements needed in your strategy here: Key Elements to an AI Strategy.  Every company update is an opportunity to show that your AI strategy is grounded in reality.

Measure in key areas

Pick the key areas where you would like AI to make a difference, for instance revenue growth, employee productivity, cost savings. This should align precisely with your stated vision, and each measure should build towards your vision. Each of these key areas should be measurable in a straightforward manner and mean that the teams looking to deliver can focus fully on making a difference, rather than trying to measure how much of a difference they’ve made. Remember that the value doesn’t only have to be intrinsic value, anything which clearly aids a build towards the vision should be recognised. For instance, building out data assets for a small project which aids productivity might be difficult to trace to the bottom line. Understanding that a happier employee whose life has been made much easier, as well as bringing data assets online for future projects, clearly has a benefit to the company. Work out a way to recognise this, and highlight it. See the end of the article for some ideas

Stakeholder management

Each different stakeholder for the AI vision will likely have their own focus for which measurement they want achieved. Stakeholder management can be tricky, so there is no one correct way to do this. One thing that can be helpful is displaying, clearly and succinctly, all the benefits as measured above. Whilst it can be difficult to quell the particular murmurings for one person’s bugbear, showing the entire strategy is progressing should ensure continuing buy in.

Plan for Experimentation, Iteration and Failures

There is so much which is possible with AI, sometimes it can feel like you want it to do everything at once. You can see self-driving cars, you get recommended all the movies and content you really connect with, you can see AI imaging predicting all sorts of medical anomalies a human couldn’t possibly spot. Why shouldn’t you want something as large and impactful as one of these use cases? 

What often gets missed in these examples are the countless iterations of models, data and previous product versions which have happened behind the scenes. For instance, self-driving cars alone rely on all sorts of previous technology changes which had their own lifetime to evolve and adapt: 

We are only just now starting to see self-driving cars hit the streets, and even then most of them are still curtailed by needing a human at the wheel. Whilst your vision should look towards an end goal, any project which contributes towards that goal should be celebrated. Ensure each project clearly shows either intrinsic value, or value towards future projects.

Harness the exponential potential of AI

It is tempting to look for a return on your AI spend so that it is cost neutral immediately. Whilst this would be great, bear in mind that this normally leads to an outsized focus on giving evidence that a project has made a difference. It is highly likely that projects will take the majority of a year to come online, so there has to be an understanding there will not be immediate payoff from starting an AI project. Further, in order to show an immediate return, cost savings are often the focus as they are the easiest to evidence on the bottom line. Too often, ROI gets reduced to cost to implement vs cost saved, and companies are taking the same view of AI ROI. Cost savings, however, are the smallest part of the equation. Larger ROI comes from productivity gains that compound, from employees who are happier and more empowered using tools that are more efficient. Cutting costs gives you a one-time, capped boost. Empowering people fuels your growth year after year.

Given AI’s potential to give exponential growth, would you want to focus on an area where the upside is limited? Ensure that you are giving the right amount of focus to growth areas i.e. revenue growth or employee productivity gains. They are harder to evidence, but ultimately have the potential to exponentially grow your business.

MVP value is hugely greater than a POC

Too often a project is presented as “get a POC (proof of concept) running” or similar wording. This is a very dangerous precedent, as by its very nature a POC offers no intrinsic value, and very little future project value. An MVP (minimum viable product), running in a production environment, with production data and bona fide users holds infinitely more value than a POC. Not only are you testing any AI methodologies or model choices, you are ensuring that production grade data flows correctly, the support model is correct, and you are able to adjust any workflows to fit with the user’s needs. Therefore, if you are looking at the different stages of going live with a project, reward the benefits appropriately. Seeing a model work one time with one set of data on a data scientist’s screen is certainly a huge step in the right direction, but seeing the outcome of a model work every time on a user’s screen, with all data feeds taken care of automatically is an exponentially bigger step.

Start small, grow bigger

Technology takes time. Just because you’re not able to show immediate results, it doesn’t mean that you’re on the wrong track. When at the start of your journey, make clear to all stakeholders that results can take time. Ensure that you have intermediate milestones which gain upfront agreement and can be celebrated in their own right. Being asked to give financial proof that the project is successful when you’re in an initial build phase is downright value destruction. Google was loss making for several years before growing into profit, Netflix had the right idea for years but the technology hadn’t caught up yet. When imagining the workforce of the future, don’t expect the future to be tomorrow. It can take time, but the results will be worth it.

In Summary

It might be helpful to think of the ROI on your AI vision to be:

ROI (AI) = ((Cost Savings + Productivity Gains + Employee Empowerment Value + Rev Growth
             Opportunities) − Risk Events)/(Cost of AI Projects)

Bring all of the benefits of your AI vision to light and ensure they get the correct recognition. The path might be longer than some may wish it to be, however if this gets highlighted up front it’s easier to stomach. Remember to try and harness the growth potential of AI, not just the cost savings. Where would you rather be in a decade’s time, the lowest cost provider of the same service, or the pioneers that reimagined your industry?

Examples of measurable impact in key areas

1. Cost Savings (upside is mostly linear and capped)

2. Productivity Gains (compounding benefit)

3. Employee Empowerment Value (cultural multiplier)

4. Growth (strategic upside)

5. Risk (which should be measured and managed)