Thought Leadership

Automation first, AI where it’s actually needed

There is one important distinction that should be addressed before any company begins using AI to optimise its operations. Most of what firms think requires AI can be done through basic automation. ‘Let’s try to get our basics right’, he explained.

Automation handles rule-based tasks with no judgment. AKM Global runs bots that enter invoices - 100,000 can be processed in a day or two, no AI required. Generative AI handles drafting, summarisation, research and reasoning. Agentic AI plans and executes autonomously but can carry significant token costs. The discipline is knowing which tier a task needs.

There have been recent reports of a technology firm that ran up a $500 million bill in a single month by giving unrestricted AI access to thousands of employees. AKM Global uses pre-filtering to control usage. As an example of managing escalations, rather than asking an agent to scan 300 to 400 daily emails, they’re instructed to look only for phrases like ‘termination of engagement’ or patterns of repeated follow-up using automation first. The AI agent then reviews only ten or twelve emails, and the task is completed at a fraction of the cost.

The architecture settled on at AKM Global separates the intelligence layer from the execution layer. Copilot or Claude provide the judgment. Power Automate and UiPath are execution-layer tools rather than AI solutions. They do not incur token-based costs and integrate seamlessly with Microsoft environments. Understanding where each component of a workflow sits within the technology stack is essential for accurately managing and controlling the firm’s operational spend.

Starting with Copilot

Rather than searching for a new product, Maheshwari’s first question was where the fastest return on existing investment lay. The answer was Microsoft Copilot, which was already licensed and part of the company’s working environment.

At a personal level, Maheshwari highlighted how he uses Copilot to analyse his emails, track responses, organise meetings by location, and even visualise them against the London Underground map while travelling. Tasks his personal assistant used to handle can be done without additional cost.

Where AI earned its place

Three use cases from AKM Global’s experience required genuine AI judgment rather than automation.

1. Transfer pricing documentation:

The firm works with around 250 comparable companies for each engagement. Identifying which to accept or reject previously took a qualified chartered accountant twelve to thirteen hours: reviewing each firm’s business profile, applying client-specific criteria, and documenting the rationale. An AI agent now performs these tasks by visiting each company’s website, assessing the functional profile against a defined strategy and producing a structured decision with written justification.

Time saved on that step: 92%.

Overall saving across the engagement: 46%.

Token cost per run: one to two dollars.

Across the firm, the expected annual saving is around 11,500 hours.

This took a few days to build.

2. An HR policy bot:

With a workforce of around 900 employees, HR executives would spend most of their time answering employee questions by looking up internal policies. AKM Global loaded all its policies into a bot and gave staff access, freeing the HR resource to focus on more strategic and value-added activities. It took 1 day to build the solution.

3. Website content:

Facing a full overhaul, AI was used to audit all pages simultaneously — SEO, discoverability, calls to action — applying twelve specialist perspectives at once. A page that had scored 3.5 against a Big Four competitor’s 9 moved to 6.9.

Starting somewhere

The key takeaway: start somewhere, and start yourself.

The AI tools have become powerful enough that a technical background is no longer a prerequisite. You can ask Claude how to write the prompt, how to build the agent, and keep asking until it makes sense. But as a leader, you need to be the one asking. Delegating the learning means losing the ability to judge what’s possible.

AKM Global has since developed AI services into a billable service line supporting clients where it has access to the relevant systems and environments. That came from doing the work internally first and understanding it well enough to take it outward.First-hand experience with AI, rather than relying on a vendor’s pitch, fundamentally changes the quality and credibility of conversations about AI adoption.