Faster doesn’t mean richer

Artificial Intelligence (AI) is only valuable if it positively impacts the bottom line, and, as of now, many businesses struggle to demonstrate this. In simple terms, the ultimate reason businesses adopt AI, as Simon Crumplin, Acora’s Director of Data Strategy & AI, puts it, is to generate more revenue, either by cutting costs or speeding up processes.

But are businesses really succeeding? Are you really getting richer?

With this in mind, we explore insights from Acora and HANDD’s September Customer Forum. This engaging event brought together customers and experts from Acora Group, Acora One, and HANDD for a full day of practical insights, business updates, and networking.

From the expert-led sessions, we selected three that focused on the main challenges businesses face when implementing AI.

  • Transforming AI into return on investment (ROI). Simon Crumplin shared practical insights on what it takes to maximise the ROI of AI.
  • Unlocking unstructured data. Darren Humphries, Acora’s Chief Information Security Officer (CISO), explored how businesses can safely unlock unstructured data and embrace AI data-driven innovation.
  • How do you become self-sufficient in data? Richard Conway, Acora Group’s Chief Technology Officer (CTO), explored ambitious but feasible strategies to help you become self-sufficient in data, leveraging AI for financial success.

This article highlights key themes they shared and includes the videos of the three sessions.

The doom loop and the false economy of speed

“Why are we doing AI? It fundamentally is to make more money. Whether you’re saving money, whether you’re going faster, it’s all about how we transform our organisations to go.” Simon Crumplin

In the rush to adopt AI, many businesses fall into the trap of focusing on activity rather than actual return. This is what Simon defines as “the doom loop of slop,”. Here is how it often plays out:

  1. A business uses ChatGPT to create a better Request for Proposal (RFP) than they could have produced manually and sends it to the vendor.
  2. As the document is too long to read, the vendors utilise another AI tool (e.g., Claude) to summarise it and generate a response.
  3. Then the business processes these AI-generated responses through yet another AI tool to determine the next questions to ask.

This cycle of passing documents back and forth through AI speeds up processing, but the quality of the outputs quickly deteriorates. Documents grow increasingly long, and buyers become more and more confused than they were when using traditional methods like manually scoring responses in a spreadsheet.

This phenomenon is a “false economy,” as Simon defines it. While AI allows you to complete your tasks faster, speed and volume become meaningless metrics if they are not connected to tangible business outcomes. Moreover, while the time you save typically prevents overtime, it doesn’t significantly contribute to your business growth.

A strong business case includes the ROI of AI

The ultimate goal should be to improve efficiency and profitability in business operations. To achieve it, incorporate ROI into all your AI initiatives from the very beginning.

Describe each use case in a straightforward, one-page business case that highlights the expected return, its real impact and the improvement on existing systems.

AI spaghetti and the cost of complexity

“If we can’t serve for less, we’ve got no ROI. All we’re doing is adding more spaghetti to the spaghetti bowl. Right? It doesn’t get us anywhere.” Simon Crumplin

Businesses mistakenly believe that simply acquiring AI tools equates to improved efficiency and faster operations. This often leads to departments piling on more AI-based applications and add-ons, while citizen developers churn out agents at an astonishing pace. For instance, one of Simon’s customers increased their agent count from zero to 2,000 in just 16 months.

However, merely buying more AI software doesn’t improve efficiency or deliver a better ROI. Instead, it creates more complexity or, as Simon calls it, “a bigger spaghetti junction. Because if you add AI without retiring anything, all you’re doing is “adding more spaghetti to the spaghetti bowl.”

So, as Simon suggested:

  • Stop purchasing more applications. Before acquiring new tools or add-ons, assess their actual value. A surplus of applications often complicates processes rather than improving them.
  • Reduce the number of applications. Aim to cut your application portfolio in half and focus on optimising the tools you already own. This is how you streamline operations.
  • Demonstrate the value of solutions. Validate every new software acquisition with solid evidence rather than relying solely on lengthy AI-generated documentation.
  • Break down silos. AI doesn’t respect outdated organisational boundaries. This creates a unique opportunity to integrate previously disconnected departments such as human resources (HR), finance, and business development (BD).

Businesses face significant challenges related to application management and data quality, which can affect performance and the ROI of AI. Thus, to drive meaningful change and leverage AI effectively, the emphasis must remain on achieving strategic commercial outcomes rather than focusing only on the technology itself.

Permissions: the hidden risk

“You’ve got to sort data out. Because you can’t do AI until you sort your data out.” Simon Crumplin

At the forum, Simon asked how many people believed their data access permissions were correct. Almost no one raised their hand. AI operates at speeds that can easily surpass the safeguards businesses have implemented, leveraging whatever access you grant to it.

Nevertheless, AI doesn’t create this permissions problem. It simply exposes and accelerates the surfacing of those that already existed.

4 permission problem examples

To better understand this risk, let’s consider the practical examples Darren shared in his session:

  • All-or-nothing connectors. AI connectors linked to systems like Jira, Salesforce, or Snowflake can only be toggled on or off for your entire business. There are no user-specific permissions. This means that without robust AI governance policies, your AI connector can access all of your data.
  • Personal accounts. If one of your developers logs into a corporate AI tool and adds a GitHub connector, everything he has built at the office may end up in his personal GitHub because he can use any of his GitHub accounts. This is something that even a CISO cannot control.
  • Disappearing labels. In Snowflake or Microsoft Fabric, you can tag fields and tables with sensitivity labels within the database. However, when AI retrieves and extracts data, these labels can be removed, exposing confidential information such as employee payroll data to anyone.
  • “Safe” environments are not foolproof. A segregated environment doesn’t always protect you from permission bleed-out. When you activate third-party modules, even isolated ecosystems like Microsoft Copilot can expose information as the modules gain access to all of your data.

Due to this “permission creep,” sensitive data categories such as employment contracts or salary information can easily be accessed by individuals outside the relevant teams. These examples reveal significant business risks that you as a business owner cannot afford to overlook.

AI Governance that says yes

Effective governance doesn’t hinder progress. It’s a robust enabler for AI. The real issue lies in the current approach. As Darren Humphries pointed out, “If you actually read your data classification policy internally, it’s not worth the paper it’s written on.” In other words, conventional data classification policies often aren’t effective enough.

His cautionary tale regarding how HR legitimately sends CVs to personal email addresses is a typical example. Blanket policies that label every CV as personally identifiable information (PII) can disrupt valid processes. Employees relabel the document as they see fit, breaking your data loss prevention (DLP) mechanisms.

Consequently, businesses often resort to monitoring as the only means to maintain operations.

Leverage AI to categorise your data

“AI moves quicker than all of us, and it’ll do everything that we can’t even think about. So to me, unlocking this context in unstructured data is a piece of gold. And finally, because of AI, we can do it. And it’s mind-blowing.” Simon Crumplin

Although Simon admitted he has never met anyone who has truly delivered DLP, he noted that if businesses can improve data discovery and labelling, DLP becomes straightforward.

AI can help with this. Data security posture management (DSPM) tools read and categorise data based on context, increasing visibility. Here is how the model works:

  • Appoint data owners. Designate data owners and stewards in each department to manage data classification and usage.
  • Eliminate unnecessary data. Data is a liability. So delete data that you no longer need. For example, one client successfully deleted 38% of its unstructured data.
  • Apply proper labels. Ensure all data is labelled appropriately.
  • Restrict access. Remove excess access rights to prevent permission creep.

Once you have completed these steps, existing security Microsoft tools, proxies, and firewalls can enforce the policy.

Addressing shadow AI

Shadow AI is another significant issue, as many systems now incorporate AI without clear awareness, as Darren points out. Businesses use their own AI tools with third-party AI solutions embedded in different products such as Customer Relationship Management (CRM) and security systems.

Employees often use personal AI tools, such as ChatGPT, at work, which puts customer data and sensitive information at risk of exposure.

To tackle these challenges, implement an AI policy and maintain a register of all AI tools. It will help you manage this problem effectively.

The ideal AI governance allows you to control risks while enabling users

“…the problem is that there’s a mandate without the proper agenda, and the agenda should always be: How are we going to release data products as quickly as possible while controlling our risk and enabling our users? And you can do all of these things if that is your agenda.” Richard Conway

When potential policy violations arise, opt for a supportive approach rather than outright bans. Prompt users with a warning: “You’re about to upload client confidential data against company policy. Do you wish to proceed?”

This approach aligns perfectly with Darren’s vision for a governance forum that enables citizen developers rather than restricting them, as well as Richard’s solution for those AI governance committees lacking strong agendas.

Fundamentally, an effective plan, Richard says, should prioritise rapid delivery of data products while controlling risks and enabling users.

Maximise the ROI of AI by becoming self-sufficient

To become self-sufficient:

1. Ensure you can get rid of AI tools as quickly as you bought them

Think of AI tools as resources that should be easy to replace. This approach lets you decommission or replace any tool as soon as a better or more advanced tool is released.

Darren explained how Acora achieved this. At the beginning of 2026, Acora phased out monday.com, initially empowering six staff members to develop their own internal project management tools and applications.

These tools were then combined into a cohesive suite built 80% by citizen developers who now own the code, while professional developers refined the remaining 20% to production quality.

2. Shift from an operative model to an adoption plan

“Five minutes. 157 words a minute on average, we’re seeing. 20 times richer than the written word. People just drop data, and we get context in reams of unstructured data. AI’s great at extraction and reprocessing.” Simon Crumplin

AI can enhance business efficiency by transforming unstructured data into valuable insights. So, as Simon suggests, to drive adoption, implement AI gradually, demonstrating its value at each step.

Collaborate closely with front-line teams, rotating the development process every two weeks. Focus on rapid, iterative progress. Remove the need to log in to tools. Leverage guided voice capture that empowers your users to extract specific information from pages-long documents effortlessly, without typing.

All these actions make access to critical data seamless, speed up processes, and enrich user experiences, removing friction.

3. Embrace simplicity

Implement what Richard discussed during his Self-sufficient in data session. Appoint a designated owner for each data product. The person will oversee access control, data quality, and associated risks.

Refrain from building missing foundational elements. Wait for solutions to evolve. For instance, many businesses invested heavily in crafting a data mesh, only for Microsoft Fabric to emerge years later with the same embedded mesh capabilities.

Articulate your operations in a language that is accessible to all. It empowers non-technical personnel to engage with data meaningfully. AI has now evolved to help teams familiarise themselves with complex systems fast, leading to enhanced operational efficiency. Follow the same line.

Governed data is at the heart of the ROI of AI

In essence, the ROI of AI hinges upon well-managed data that is labelled, properly permissioned, and actively owned by business units. This is what transforms information into insightful decisions, those decisions into actionable steps, and ultimately, those actions into measurable results.

Acora is your ideal partner in this transformation journey. Our expertise in data and AI, cyber security, intelligent applications, and AI as a service fosters an environment that facilitates this shift. As Simon puts it, “We’re just an enabler.”

But everything we do is aimed at making you succeed.

Discover what we can achieve together. Explore the recorded sessions from our September forum and get in touch.

Transforming AI into ROI: Simon Crumplin, Acora’s Transformation Director of Data & AI

Unlocking Unstrctured Data: Darren Humphries, Acora’s Group CISO

How do you become self-sufficeint in data: Acora’s Group CTO