Ok. We’ve got some superpowers. Now what? (Part 2)

We need to take a look at the ‘jobs to be done’ within the business.

To keep things focused, we will ignore the use of AI within products or services for now – e.g. hyper-personalised customer journeys, AI avatars, etc. Instead, we look at AI in the operating model – how it can help run the business better.

Every business runs on a set of core functions: sales, service, operations, research, IT, finance, HR, etc., that jointly deliver customer outcomes, and if things go well, hit financial goals. Each function can be broken into processes, tasks, and activities, stitched together with hand-offs and interdependencies. On top, sit cross-cutting layers: governance, risk management, decision making, plus strategy and transformation to continuously evolve and reposition the business.

The three horizons of AI applications

To explore how AI capabilities can be used across these functions, it helps to consider three horizons of application.

1. Retooling – new tools, same jobs. You give people copilots / coworker type tools that make them a lot more efficient. They draft faster, analyse quicker, and spend less time on tasks. The process stays the same, but the speed and efficiency improves.

2. Reengineering – embedding AI into the process. You integrate the AI capabilities into the process, redesigning workflows from the ground up. AI sits inside the flow now, doing a part of the job. Think customer service, where the AI bot answers all calls, handles triage, and potentially some customers through to resolution.

3. Reimagining – replacing end-to-end processes with AI. Here AI agents handle complete processes autonomously, with humans supervising, or stepping in for edge cases only.

The reality across functions: uneven progress

We recently took stock of current AI offerings from specialist vendors across a few industry-agnostic support functions. This clearly shows progress, but unevenly across the board:

• Customer service is furthest along: lots of retooling, and increasingly serious reengineering, with early reimagining in voice agents and automated ticket flows.

• Sales and marketing are heavy on retooling (content, copy, analysis), lighter on deep process reengineering.

• Finance and HR are slowly moving from simple automation (invoices, payroll, basic analytics) toward workflows where AI is more central, but concerns about risk and compliance keep the brakes on.

• Legal, project management and strategy are using AI for drafting and analysis, but wholesale reimagining is still rare outside a few leading firms.

Source: Rygur research, company websites and news releases

It is clear that we do not yet have many applications that fundamentally change the way work is done – for now, the majority of solutions apply to specific processes or parts of processes, rather than true end-to-end AI. Think augmentation, rather than reinvention for now.

How to decide where to apply AI

Given all this, how do you actually decide where to start, or where to double down? Two big questions matter more than anything else:

1. What are your priority problems or opportunities?

Not: “Where can we use AI?”But: “What do we most need to fix or improve?”

Do you need to improve your support service levels, where hiring more people is non-economical? Or do you need to increase your sales team’s effectiveness? Or is there far too much overhead or non-productive work being done in the business?

Start by listing the pain points and opportunities by function: customer, revenue, cost, risk, employee experience. AI should be pointed at problems that already matter, not invented use cases just to justify a model subscription.

2. What is your current capability to build, deploy and adopt AI?

This is the unglamorous but critical bit. Look across four dimensions:

• People – Do you have folks who understand both the domain and how to work with AI tools?

• Process – Are your workflows documented, measured and stable enough to change deliberately?

• Technology – Do you have the basic plumbing (APIs, integration tools, data access, security) to embed AI without duct tape?

• Data – Is the data you need accessible, reasonably clean, and governed, or is it scattered in spreadsheets and inboxes?

Our research shows that this capability piece is where scaling often stalls: many firms have dozens of pilots, but only a small fraction ever make it into robust production because they don’t have the capabilities or capacity required to scale.

The practical play here is to find applications of AI that lie in the intersection of both answers. The solution should be one that you really need and will be supported by your current capabilities and maturity.

Build vs buy

There is one more big question: how should you go about doing this?

On one side are the orchestration tools and low‑code platforms — Zapier, Make, n8n and a growing list of agent builders and workflow engines. Claude Cowork can itself be used effectively within workflows as an agent. One of the most exciting facets of this AI rush is the ability to ‘build it yourself’ using little more than plain English instructions. The leading model providers will have you believe that you can wire up your own AI‑powered operating model by the end of the week.

On the other side are specialist AI vendors who do one thing (support automation, AP, HR, legal, etc.) extremely well, with their own tooling, playbooks and support teams.

The temptation, especially for technically minded teams, is to think: “We have Zapier/n8n/Make and access to great models. We could just build this ourselves.”

And often, you could. The question is whether you should.

Three practical risks to keep in mind when building yourself:

1. AI slop: A little bit of AI everywhere, but no clear ownership, no standards and no way to tell what is working. Lots of automations; very little accountability.

2. Tool sprawl: Each team picks its own tools and orchestrations, resulting in a zoo of bots, workflows and agents that nobody fully understands or governs. Managing this is not just annoying; it can become a security and compliance problem.

3. Black box behaviour: As workflows get more complex, it becomes harder to understand why an AI-driven process did what it did — especially when it is assembled from several tools and models strung together with minimal documentation.

This does not mean “never build.” It means:

• Build when the workflow is core to your differentiation, and you’re prepared to own the complexity.

• Prefer specialist vendors when the job is important but standardised, and somebody else already does it for a living.

• Use general orchestration tools as a platform for experimentation and glue, not as an excuse to rebuild an entire ecosystem you could just buy.

Your approach should be tailored to your specific circumstance. Do you have the capability, capacity, and appetite to build solutions yourself? Are you building a core-differentiator, or an application that is not really specific to your business? The important thing is whether value actually shows up: in customer metrics, revenue, cost, speed, risk reduction, or employee experience.

Starting is not the hard part. Scaling is.

Starting with AI has never been easier. You can spin up an application in an afternoon, plug an AI node into your workflow tool, or test an agent in a sandbox with no engineers in sight.

What the data consistently shows, though, is that scaling AI is much harder, and bankable ROI remains elusive. Most organisations are using AI somewhere; far fewer have turned it into a repeatable, well‑governed part of how work gets done.

So, the real leadership challenge is not “How do we get into AI?”It is:

• Where do we deliberately deploy it across functions?

• At what depth: retooling, reengineering, or reimagining?

• What capabilities do we actually need from the superpower cabinet?

• And how do we avoid turning our operating model into an AI‑flavoured spaghetti bowl?

Get those decisions right, and AI stops being a novelty act and starts becoming something far more interesting: a practical, compounding advantage in the messy reality of running a business.

A leader's checklist: before you deploy the superpower

To get going, here are seven questions worth sitting with honestly.

1. What problem are we actually solving?Not "where can we use AI?" but "what is genuinely broken, slow or expensive right now?" If you cannot name the problem clearly in one sentence, the AI project probably should not start yet.

2. Which capability do we need, and how mature is it?Writing? Research? Customer interaction automation? Process orchestration? Some of these are already reliable at near-professional level. Others still need significant human oversight. Match the tool to its current actual maturity, not its marketing maturity.

3. What horizon are we building towards?Retooling (better tools, same jobs), reengineering (different processes, new flows) or reimagining (fundamentally different operating model)? Each requires a different level of investment, change management and risk appetite. Be honest about which one you are actually ready for.

4. Do we have the capability to do this well?People, process, technology, data. Which of these is the weakest link in the specific area you want to change? Rushing to deploy AI on top of messy data, undocumented processes or a team that has not been brought along is a reliable way to produce expensive disappointment.

5. Build, buy or partner?Just because you can wire something together with n8n and an API key does not mean you should. Think about: who will own this when it breaks? Who updates it when the model changes? Who handles the governance? Bespoke builds are notoriously expensive in the long run.

6. How will we know if it is working?Define what success looks like before you start, not after. Time saved, cost reduced, resolution rate improved, error rate down; pick measures that connect to something that matters, and review them regularly. AI that "feels useful" but cannot be measured tends to fade.

7. What are we not going to do?This might be the hardest one. AI makes it tempting to say yes to everything. Every function will have ideas. Every vendor will have a demo. The organisations that scale AI well are almost always the ones that said no to most things early on and executed a small number of applications properly.

A final thought

The biggest shift in thinking about AI is not technological – it is managerial.

The technology is genuinely remarkable. It can already write, analyse, code, summarise, plan, converse and automate at a level that would have seemed extraordinary just a couple of years ago. And it is improving, quarter by quarter, in ways that are not slowing down.

But remarkable technology, poorly deployed, creates costs and confusion. The graveyard of enterprise tech is full of systems that were impressive in the demo and ignored in the office.

The leaders who will get the most from this moment are not the ones who move fastest. They are the ones who think most clearly: about which problems matter, which capabilities are ready, and how to build something that actually sticks.

The superpower is real. The question is whether you are ready to use it well.