Envitia AI Blog – Beyond Hype and Headlines – How to get Started (Ch.2)

In Envitia CTO Richard Griffith’s second blog in the series focusing on AI, he gives some practical advice on the considerations and questions that organisations and individuals should address as they embark on their AI adoption journey.

“AI is developing at an incredible pace and the possibilities it brings are so broad that it can be hard to understand how you can get started. Less than a year ago technologies like ChatGPT, Bard, Gemini and Copilot simply weren’t in the public consciousness, roll forward to today and there are daily headlines hyperbolising the effects that AI is going to have. At Envitia, we speak to a lot of business leaders who know that they should be using AI but are not really sure either how or where to begin. In this post I will try and simplify how you can use AI today.

Understand the business outcome

A key aspect of any technology implementation is to start with the business outcome(s) you are trying to deliver. Understanding the “why” and “what for” is vital. It is easy to fixate on the power of new technology without fully understanding the business outcomes it is able to drive. As the great Douglas Adams wrote, “We are stuck with technology when what we really want is stuff that just works”. This has never been truer, we must understand what we want to work, before technology has any chance of helping.

Business outcomes are the “what” you want to achieve and should be measurable, whether that be in terms of revenue delivered, costs saved, employee retention, reduction in carbon emissions or whatever else you seek to improve within your business.

Enterprise Architecture: why-what-how

When you consider business outcomes and start to establish the rationale behind investing in a change you need a blueprint for aligning business strategy with technology and processes, driving efficiency and agility. This is where Enterprise Architecture comes in, the art of joining the “Why” to the “What” to the “How” for business change. We’ll come back to this later……

Strong foundations support lasting change

Whilst it is tempting to dive straight in, for successful business change you have to start with the foundations, the often-overlooked keystones that enable successful delivery. For business change there are four key foundations to consider: People, Process, Technology and Data. They tend to have different maturity levels in every organisation, and these maturity levels are what differentiate successful change from disaster/vanity projects. There are many recent examples of people jumping into AI technology and spending vast amounts of money for little or no result. And there will be many more!

What has AI got to do with data?

Everything. Data is what powers the chatbots, the copilots, your digital assistant. All of these underlying models and algorithms are using your data to power responses. Moreover, AI is like a student and data is its textbook. The quality and quantity of data it ingests determine how well it “learns” to perform tasks, make predictions and solve problems. Get that textbook wrong and AI can become biased, inaccurate, and even harmful.

Here’s an example of how AI and data can empower a sales organisation and where it can go wrong…

Scenario: Imagine a clothing retailer with a large online store. Their sales team struggles to personalise recommendations for each customer, leading to missed upsell and cross-sell opportunities. As a result, the retailer implements an AI-powered sales tool that analyses customer data, including purchase history, browsing behaviour and demographics. This tool can:

·         Identify patterns and predict preferences

·         Generate personalised recommendations

·         Optimise targeting and messaging

·         Constantly learn about the retailer’s customers and refine its recommendations accordingly

Successfully deployed the business impact would be significant: sustainably increasing sales, improving customer experience and increasing the efficiency of the sales team. Obviously, this all sounds great. But there is risk, feeding bad or incomplete data into such an AI-powered sales tool could lead to a range of negative outcomes:

1. Biased recommendations: If the data used to train the AI contains biases, like focusing on past purchases only from a specific brand or demographic, the recommendations will be skewed and miss opportunities for other relevant products. This can alienate customers and limit sales potential. How many times have you been recommended products online that you have already purchased? If you have recently purchased a sink plunger it is highly unlikely you want another one. (Ed – unless you collect them, which is your own business).

2. Inaccurate predictions: Inaccurate data, like outdated purchase history or faulty product categorisation, can lead AI to make wrong predictions about customer preferences. This can result in irrelevant recommendations, frustrating customers and potentially damaging brand perception. For example, recommending winter jackets in summer based on previous purchasing history so the individual is treated incorrectly because of the inaccurate data.

Similarly 3. Ineffective targeting: If the data is incomplete and lacks useful information on factors like customer location or seasonal trends, the tool might suggest products irrelevant to the context, leading to wasted marketing efforts and lost sales. For example, suggesting only high-end items to all customers, ignoring preferences or affordability. In this example the tool’s ability to determine the individual uniqueness is degraded because of the incomplete data.

4. Reduced trust and adoption: If customers receive irrelevant or unhelpful recommendations, they might lose trust in the AI tool and the brand itself. This can lead to decreased engagement and resistance to future AI-powered initiatives.

5. Security and compliance issues: AI Tools are designed to consume data like wildfire, anything that the tool has access to it will be used in many new and novel ways. This power is also a risk, if the tool has access to information it shouldn’t. For example, if you got a recommendation from an online store that told you about your neighbour’s buying habits, you would naturally be anxious about how your own data was being shared.

Data Foundations: setting up for success

The lesson is clear, any AI implementation must be based on reliable, well maintained and complete datasets. Regular data cleaning, validation, and inclusion of diverse customer segments are essential to allow AI tools to work to maximum effect, avoid biases and mitigate the risk of misfire.

Data Security and Data Governance have major consequences when AI consumes data at vast scale, as any weaknesses that may have gone unnoticed in your data security model will be magnified and exploited (essentially deliberately!) by a modern AI algorithm.

To enable AI exploitation in any organisation Envitia believes that the data foundations are the most important pillar to get right. As you can see from the example above, without the right data foundations, AI implementations will fail to provide the outcomes expected and can be a costly mistake.

What are these Data Foundations?

Data Architecture, Data Modelling, Data Quality and Data Security are key tenets that need to be considered for a sound data foundation.

Data Architecture is the blueprint for how an organisation manages its data across the entire landscape of the company. It defines the structure, organisation, and flow of data within an organisation, ensuring it’s accessible, reliable and secure. Think of it as the foundation upon which all your data-driven activities – storage, access, analytics, reporting, and importantly AI – are built.

Data Modelling serves as a blueprint for storing, organising, and accessing data efficiently at an individual dataset level. If Data architecture is the house foundations then Data modelling is the blueprint for individual rooms within the house. It specifies the placement of walls, doors, windows, and furniture within each room. Data modelling ensures data accuracy, consistency, and usability within specific systems. It facilitates easier data retrieval, analysis, and reporting. It’s worth stating that this is not a one-time activity – it’s an ongoing process.

Data Quality refers to the practice of understanding the condition of data based on specific criteria that make it fit, or unfit, for its intended purpose. There are lots of considerations (dimensions) of what makes data fit for purpose such as accuracy, consistency and timeliness, and there is very rarely a bit of data that is fit for all purposes. Once measured and understood, practices such as data cleaning, data governance and change management all contribute to increasing data quality across an organisation.

Data Security and Governance are the pillars that mean the right people have access to the right data at the right time and data that needs to be protected is handled appropriately. The new copilot applications bring Data Security and Governance into renewed focus. Whilst most organisations will have taken time (and expert advice) to store and protect data appropriately, how rigorously or regularly do they check? The power of modern AI tools to relentlessly seek and consume data means that they will, inevitably, find data that they are not supposed to if it is available. A classic example would be an organisation that deploys a copilot tool on its internal SharePoint infrastructure. The tool would be fantastic at, for example, finding past credentials or summarising reports, but could also be used inappropriately. Any faults in permissions on SharePoint could lead to the tool being able to find information that it shouldn’t. What if a mischievous employee asked copilot “What was the last HR incident involving Employee X” and had confidential results returned. Ouch.

How to get started…

So your CTO / CEO has said you “need to implement AI” and you are unsure where to start. I would suggest you start with modern Enterprise Architecture. At Envitia we don’t mean the “old” EA where you get a bunch of management consultants to write a series of UML diagrams that don’t mean anything to non-architects and end up on a shelf. Instead, we base our Enterprise Architecture on a considered design of change: a way to guide an organisation through change to focus on business value, delivering early benefit and iterative improvement.

Start with answering these questions:

·         First things first – “why” – what business outcome are you trying to influence or change?

·         How is that business outcome influenced by technology or data?

·         What quality does the data need to be to make a decision on it?

·         How will the data be secured and what secure information needs to be used?

Once you have answered these questions you have gone a long way to establishing the business pillars upon which a successful pilot implementation can start. A pilot rollout of, for example, an AI copilot would be to one department or group of people who have been trained on what AI (especially generative AI) can and can’t do. You then need to monitor the outputs and compare with the original business outcomes you were expecting. Is the combination of the new tool and your data delivering as expected? Evaluate the results, decide on any iterative improvements (or simply fail fast) and then decide what to tackle next. An agile approach with a clear ‘north star’ but a flexible road to get there will deliver value quickly whilst allowing for flex as your business, the environment or the technology changes. And let’s be honest, all three will change!

At Envitia we’ve been solving complex data problems for 30 years. Come and talk to us about how our approach to Enterprise Architecture and our focus on getting the Data Foundations right can start your AI journey today. Reach out to me if you would like any further information.”

Richard Griffith

Join our team of talented innovators

View all job opportunities

Related articles