Our work across very different organisations suggests leaders should first decide what capability they want AI to make stronger, then engineer the intelligent business layer that can scale it.
There is a question we keep hearing in executive conversations, usually somewhere between the AI vision slide and the implementation slide, “Can we get an agent on this by Q1?”
It is an understandable question. Today, Agents are becoming tangible tools, demonstrated, named, funded, and placed on a transformation roadmap. For instance, a finance agent, a workforce agent, a procurement agent, an executive assistant. We’ve observed they give leadership teams something visible to point to when boards want evidence that the organisation is moving on workforce and AI solutions.
But after working with organisations across government, financial services, professional services, healthcare, education, and traditional operational businesses in the Middle East, USA and UK we have come to think the question starts in the wrong place.
The most consequential AI decision is often not where to deploy an agent, it is deciding what the organisation wants to become better at because AI exists. This distinction sounds subtle, but it changes almost everything that follows.
When leaders start with an AI use case, they are often linked with a task or workflow that technology can automate. When they start with an organisational capability, they are forced to think instead about what creates disproportionate value, such as, judgement, customer relationships, institutional knowledge, commercial intelligence, cultural interpretation, decision quality, or the ability to respond to uncertainty.
This approach leads to very different systems being built. We have started calling this amplification before automation. It emerged as a pattern we noticed after seeing the same mistake appear in several very different organisational settings.
The pattern is simple:
Amplify, Engineer, and Simplify.
First decide which organisational capability should become materially amplified. Then engineer what methods-layers, workflows, memory, data, permissions, economics, and human-AI boundaries are required to make the capability repeatable. Finally, ask what the organisation can now do that was previously too expensive, too slow, or simply impossible.
This is more demanding than identifying an AI use case. It is also much closer to the real management problem underpinning strategies investors have been sold.
Start With What You Want More Of
One engagement during the US-Iran conflict involved a GCC government-related programme exploring how could AI preserve the future of culture and heritage. At first the challenge was framed as, “How should we use AI in a cultural experience?”
That question immediately pointed towards the technology though, including conversational interfaces, digital guides, personalisation, generated content, and other visible front-end applications.
But when we got under the skin of root cause, this was not the important question. Culture and heritage contain meaning precisely because people interpret them. Experts, communities, institutions, and visitors bring context that is difficult to reduce to a transaction. The management challenge therefore shifted towards identifying what should be amplified. Greater access? Better interpretation? Participation? Preservation? The ability to transmit expert knowledge to more people?
Only after answering those questions did, we ask where does AI belong.
The same pattern appeared in financial services, but in a very different form. Senior leaders in a GCC bank were trying to understand what AI and agents would mean for jobs, skills, performance, and employability. The obvious conversation quickly becomes one about how many roles AI might replace and what employees should be trained to do.
But is a job the wrong unit of analysis? When thinking about a workforce role conducted by a human, it is a bundle of activities consisting of routine processing, judgement, relationship management, exception handling, learning, and accountability. Some of those activities can move to AI, some should remain human, and others become more valuable precisely because AI absorbs the surrounding work.
For a relationship manager, for example, the opportunity may not be to automate the relationship. It may be to remove enough research, administration, and routine preparation that the human has substantially more capacity to exercise judgement and develop important clients.
This is why “learn AI” is an inadequate workforce strategy. The real question we find brings strategic outcomes is “What do we want our workforce to become better at once AI performs more of the surrounding work?”
That is what we refer to as an amplification question.
Hours Saved Are Not the Same as Value Created
One sales function engagement in a UK SaaS organisation made this distinction unusually visible.
This B2B organisation had 76 salespeople who were collectively spending roughly 800 hours every month researching accounts and markets, on 25% of the businesses ARR equating to approximately five full-time employees operating an eight-hour working day. The obvious automation opportunity the leadership felt they needed from an agent was to reduce the research cost in employee time.
An agentic capability did exactly this, where the cost of the routine research task fell dramatically.
The interesting result was not the efficiency statistic though, and more importantly the CEO did not want salespeople removed. They wanted released capacity to redirect toward larger, higher-touch strategic accounts i.e., ARR of 75% and above. Within three months the organisation attributed approximately £395,000 in additional annual revenue to the change.
This is a very different way to think about productivity.
If AI saves an employee ten hours a week and those ten hours disappear into email, meetings, and additional low-value work, the organisation has technically improved productivity but may not have increased enterprise value due to the time-on-task savings.
The more useful questions which emerged during the usage of ‘research sales agents’ was “what inefficiencies towards net profits has AI released?”
We call this latent capacity unlock. The amount of scarce human capacity freed from repeatable activity and successfully redirected toward work that creates greater value.
When we raised this question with the CxO it changed the executive measure from “How many hours did the agent save?” to “What did our people do with the capacity we released?”
For a COO, those are not equivalent questions.
Why Organisational Knowledge Management Systems gather dust
The second recurring problem appears once companies begin moving beyond individual productivity tools.
Experienced organisations work because people know things the systems do not.
The norm we feel and see without AI in procurement operations generally sees a senior logistics operator who recognises an exception before it becomes visible in a dashboard, a relationship manager who knows that an apparently routine customer request carries a larger commercial signal, a project manager knows which supplier is likely to create problems even though the spreadsheet says the bid is attractive. An experienced consultant knows which piece of evidence should change the direction of an engagement, and which is merely noise.
This knowledge is valuable but fragile, and much of it sits in workforce heads, inboxes, meetings, personal documents, and habits developed over years. We find AI exposes that fragility because an agent can only work with what the organisation has made available to it.
While developing an AI-supported strategic-foresight solution for a USA tier 1 consulting firm, the underlying expertise had been developed by a highly regarded consultant over more than two decades of working with fortune 500 organisations and global governments. The asset was not the ability to write a 300+ slide deck. It was the method underneath the report, how a question is framed, which signals matter, how uncertainty is interpreted, where human challenge enters the process, and how multiple futures are translated into decisions.
Simply asking a general-purpose model to “do strategic foresight” did not reproduce that expertise. We decomposed the work into processes, skills, guardrails, decision points, and reusable knowledge objects. Once that happened, an engagement that would historically have taken roughly eight weeks was compressed to approximately two days while keeping experienced human judgement at the points where it mattered most.
The important outcome was not merely speed. It was an expert method had now begun moving from an individual dependency toward a repeatable organisational capability.
This suggests another measure executives may need. Repeatable Expertise.
Instead of asking how much documentation the company has produced, leaders should ask a harder question, “If the person who knows how this works left tomorrow, how much of the organisation’s ability to make the decision would survive?”
That is a fundamental question when it comes to revenue lose reasons, and a better test of whether AI is strengthening institutional capability alongside its profitability.
The New Business Moat. Document Automation to Commercial Intelligence
The third pattern becomes visible when leaders look beyond the first workflow.
In one hotel project-based business, the original opportunity appeared to be automation of bills of quantities, specifications, tenders, revisions, and cost comparisons. These were time-consuming processes and therefore obvious candidates for AI.
The documents were not the real source of value, they were containers for a much larger network of knowledge involving projects, suppliers, products, brand standards, commercial history, cost comparisons, specification changes, and experience accumulated across previous jobs.
Automating a document would make the current process faster. However, connecting the information underneath those documents could create something different, a persistent commercial-intelligence capability.
Once supplier history, specifications, revision data, project decisions, and commercial comparisons can be connected, the organisation can potentially benchmark across projects, recognise commercial risk earlier, make faster sourcing decisions, reuse supplier intelligence, and create new information products for customers.
At this point we found management questions had moved a long way from “Can AI write the BOQ faster?” to “What new capability do we possess because this intelligence can now operate across the business?”
That is the move from amplification to simplication.
The Intelligent Business Layer
These cases also explain why companies cannot achieve this by simply buying the better model.
Once leaders know what they want to amplify, they can engineer the layer that allows the capability to operate reliably.
In our own work at Synaptix Solutions, we originally treated these as separate technology problems. We built frameworks for determining whether an AI opportunity was viable, modelling the economics of agents, creating persistent agent memory, assigning identity and accountability to agents, governing AI-generated software, and codifying experienced judgement.
Over time it became clear that these were not independent problems. They were different parts of the same intelligent business layer. This layer includes at least six things:
- A method-layer: the rules, judgement, and operating logic that define what good looks like.
- A memory layer: the organisational context and previous decisions that should not have to be rediscovered every time a workflow runs.
- Trusted data and clear permissions.
- Explicit boundaries: between what the machine can do independently and where human judgement must enter.
- Governance and traceability so that the organisation knows what happened and why.
- Increasingly it needs economics: the cost of inference, tools, human review, model changes, and operating the capability at real production volume.
This is why we have seen successful prototypes can still become a failed transformation. A demo proves that the LLM can do something, it does not prove that the organisation can operate that capability repeatedly.
Three Questions Before Your Next AI Investment
The framework we now use is for executives asking where to start when navigating AI.
Amplify: What should become materially stronger?
In an age of software, organisations naturally start here. Identifying the capability that creates disproportionate value feels uncomfortable but naturally delivers. It might be judgement, relationship depth, commercial intelligence, institutional knowledge, resilience, cultural interpretation, or something else. Then ask whether AI can genuinely increase its reach or quality. We find leaders identify whether they need AI or whether it belongs in a roadmap at a point of business maturity.
Engineer: What must exist underneath it?
Map the actual work, including what your experienced workforce knows but formal systems do not capture. Determine what needs to become data, method, memory, permissions, human oversight, and measurement. Modelling the economics before scale, we find is the hardest challenges for CxOs. The aim is not to automate every part of the workflow; it is to make the capability repeatable without stripping away what made it valuable and in a non-deterministic way.
Simplify: What is now possible that was not possible before?
This is where leaders should resist the temptation to stop at productivity. If scarce expertise can suddenly operate across many more situations, if fragmented knowledge can be reused across projects, or if expensive human activity can be concentrated on the problems where it creates the greatest value, the organisation may be able to create a different service, workforce model, customer experience, information product, or source of revenue.
That is where AI moves from efficiency to what we call fast strategy.
Why AI Capabilities Needs New KPIs
When we grounded AI in business process Synaptix Solutions we found existing KPIs of productivity did not float. The shift with agents exposes a measurement problem, and we found it raised questions around the norms of an organisation. We know this, because we have seen it.
Most organisations are measuring how you perform, individual contributor. Let’s consider an example of current day AI rollout, AI through adoption, licences by hyperscalers, hours saved, task completion, and perhaps short-term ROI. Those measures are useful, but they say little about whether the organisation itself is becoming more capable.
Our engagements with client work suggest three additional measures worth developing.
- Latent Capacity Unlock measures how much scarce human capacity has been released from repeatable work — and, critically, how much of it has been reinvested into higher-value activity.
- Repeatable Expertise measures how much strategically important knowledge and judgement can survive the departure or unavailability of the person who currently holds it.
- Compounding Intelligence measures whether codified organisational knowledge is being reused across workflows, teams, and decisions to create new strategic or commercial value.
These measures are still developing, and they will need different operational definitions in different organisations, but they point toward a more useful executive question.
Instead of asking how much AI the organisation has deployed, ask whether AI is making the organisation’s valuable capabilities stronger, more reusable, and more difficult for competitors to replicate.
Start With the Capability, Not the Agent
Agents will become easier to build and easier to buy. Every major software platform will offer them. This means access to an agent will increasingly be the least interesting part of an AI strategy.
The differentiating capability will sit underneath: what the organisation knows, how it exercises judgement, how intelligently it allocates human capacity, and how effectively it can convert those things into reusable operating infrastructure.
This is why we would change the question at the beginning of an AI investment, not use case, “Where can we deploy an agent?”, but “What do we want this organisation to become significantly better at?”
Once leadership can answer that, the technology conversation becomes considerably easier, and considerably more valuable for investors.
Book an AI analysis call →
A straight, practitioner-led read on your AI programme. No pitch deck, no facilitator layer.
