For most companies, workflows and processes are the centre of their business performance. These deep rooted and critical workflows determines success. Therefore workplace technology must serve the workflow, not the other way round.
My advice is always: don’t start with AI, start with frustration and pain point instead. The key to smooth AI integration is process innovation, not just product innovation. There is a temptation to focus on the technology, rather than the business need or the end solution, but I would recommend that businesses first conduct a thorough review of all processes and operations.
How can businesses review their processes to best incorporate AI agents?
No matter what size the business, start with these three questions:
- Where do things regularly get stuck?
- Where are people doing repetitive, routine tasks?
- Where do customers or staff regularly say: “this should be quicker”?
The answer to those three questions is where you will find that AI agents tend to fit best.
Start small, reviewing processes sounds overwhelming, especially in an increasingly challenging business environment where staff are faced with many different, competing priorities all at once.
Just pick one workflow, which could be onboarding a customer, raising an invoice, handling enquiries; anything that stands out as a pain point.
Write down the steps as they actually happen. Highlight where the manual handovers are, where information is repeated, and where there are delays waiting for approvals or responses. If a process needs lots of copying, chasing, or checking, it’s a strong candidate for an AI agent.
What are the three key ingredients for a strong agentic AI foundation?
The big mistake I see is trying to “AI‑wash” broken processes.
I recommend fixing the workflow first, as you may find parts are no longer needed, or a new version works better. Once ‘fixed’, then automating that new workflow is more effective, which is how we work in our Business Productivity team.
The three key ingredients to build a strong foundation within workflows for agentic AI are: data, ownership and mindset.
First, and most importantly, is structured data. AI agents don’t need big data; they need clean and consistent data. This can vary from customer details in one, connected database to documents stored in a searchable record system, whatever suits your business needs. You don’t need perfection. You just need enough structure for an AI to reason sensibly.
Second, clear guardrails on ownership and accountability. Agents work best when someone is accountable for the outcome and there is clarity on what the agent can and cannot do. This can be as simple as the agent prepares, suggests and drafts content, but the human still approves. We find what works best is human-enabled AI, not replacing human workflows completing with agentic AI.
That balance is key. You don’t lose control, instead, you become more efficient.
Third, but by no means least, is a mindset of incremental improvement. The businesses that win will not do ‘Big AI Projects’. They will launch small, test quickly, keep what saves time and discard what doesn’t.
Agentic AI requires a complete change in mindset, as it is more a system of continuous trial and error, learning and improvement than a traditional IT system.
What small, incremental changes can SMEs make now?
This is the most important bit.
You do not need agents tomorrow to prepare for agents later.
Some very practical steps that can be actioned now are standardising templates (quotes, proposals, onboarding docs), reducing email‑only processes and movingkey workflows into systems. Documenting simple processes, even in bullet points, and cleaning up file storage so information can actually be found. It sounds simple but these steps are often overlooked in the rush towards new technology and can be pivotal in the success of AI implementation.
How can AI agents excel within teams?
Our advice is to work with teams to shift their mindset towards agentic AI co-pilots assisting, not replacing them.
A useful concept for conveying this message can be ‘foundational rails’, or train tracks. When you’re ready to run AI agents, the train moves much faster.
Using real-life examples tailored to your business sector where AI agents have already had success in other businesses can help teams get used to the idea. Today, AI agents are best at multi‑step, rules‑based workflows where humans add judgement at key points.
Some great use cases include triaging and drafting responses to customer enquiries, with humans reviewing the response and making the final decision. Similarly, in sales, agents can help chase quotes, follow-up and update CRMs, all in the background to enable the sales team to have more face-to-face time with customers. For finance and IT, AI can help match information, update statuses, manage tickets and perform checks.
One pattern stands out: the ROI comes from reducing interruption, not replacing people. While agents quietly handle the background noise, humans focus on decisions and customers. Agents don’t replace valued employees; they protect their time.
Start with personal productivity to build confidence, then move into business productivity to unlock real value. And remember: the future won’t belong to the biggest businesses, but to the best‑prepared ones.
