Drawing the Bow: Why Your AI Journey Must Feel Worse Before It Feels Better

When you draw a bow, the arrow moves backwards before it flies forwards. That movement is not failure. It is the mechanism. Draw too little and the arrow drops at your feet. Rush the release and it goes wide. The draw is the investment that makes the shot possible.

The discomfort of the transition phase gets read as evidence that the approach is wrong, the technology is overhyped, or the organisation is not ready. That conclusion arrives too early, and it costs them the shot.

What genuine AI adoption reveals first

The first thing genuine AI adoption surfaces is not capability. It is unreadiness. Data that is inconsistent, incomplete, or siloed. Processes that work because a person compensates for them, not because they are actually sound. Governance frameworks written for a different era. Access patterns nobody documented because nobody needed to.

None of this was hidden before AI arrived. It was papered over by human judgement, workarounds, and institutional knowledge carried quietly in people’s heads. AI surfaces it.

This is the draw. It is uncomfortable. It is supposed to be.

The avoidance pattern

The most common response to this discomfort is bolt-on augmentation. Find a tool that fits over the existing process without disturbing it. Add a chatbot to the customer portal. Put an AI assistant in front of the inbox. Run a pilot in one team and call it a proof of concept.

These are not wrong in themselves. The problem is when they become the programme: when an organisation mistakes AI veneer for AI capability and declares success without ever doing the uncomfortable work underneath.

The draw gets skipped. The arrow goes nowhere.

Boards and leadership teams are particularly susceptible to this because the J-curve looks like failure on a dashboard. Productivity dips during transition. New issues emerge that were previously invisible. Costs increase before they fall. The instinct is to intervene: to stop the experiment, revert to what is known, or replace genuine transformation with something that looks better in a slide deck.

Trimming headcount ahead of proven AI capability is the most damaging version of this instinct. It assumes productivity gains have already arrived. They have not. They come after the investment, not before. And counter-intuitively the increased productivity is likely to create opportunities to expand into areas previously left on the back-burner.

The minority path

The organisations that get through the draw phase are not the ones with the best technology. They are the ones that named the dip before it arrived, gave people permission to be in it, and measured the right things during the transition.

Not output volume, which will likely drop. But learning indicators: how many processes have been documented and tested; how many error patterns have been identified and fed back; how much unspoken institutional knowledge has been made explicit.

This is what leaning into the void actually looks like. It is not comfortable. It is also not optional if genuine capability uplift is the goal rather than AI veneer.

The dark factory question

Here is a useful frame for where that capability should ultimately land.

The term “dark factory” calls to mind a robotic manufacturing line or an automated fulfilment centre, dark because robots do not need lights. But a dark factory does not have to be physical. It can be software. It can be information flows. Entire classes of work (scheduling, reporting, compliance checking, data entry, first-pass analysis) that run without human hands on them.

The question worth asking is not “how do we add AI to what we already do?” It is: which parts of this business could operate in the dark? Which workflows, once redesigned, could run without requiring human attention at every step? What guardrails could be implemented deterministically, rather than human-powered?

And if all of those parts were dark, what would that free your people to do?

That is the real question underneath AI adoption. Not efficiency. Not cost reduction. But: what is uniquely human about this organisation, and how do we amplify it? Dark factories make human frontlines possible. The draw is what gets you there.

What the draw phase looks like in practice

For an individual: confusion about which tool to use; time on configuration that does not immediately pay back; outputs that need significant editing; a period of learning a new way of working rather than doing the actual work. Expect several months before the system returns on its investment.

For a team: process documentation that surfaces how inconsistent actual practices are; disagreements about the right way to do things that AI now makes the tension visible; temporarily higher administrative load as operating agreements are built. And time tutoring the AI capability from novice intern through stages of competence and trust development to an autonomous specialist.

For an organisation: change management work that precedes any efficiency gain; governance questions that were previously hypothetical now requiring real answers; a temporary reduction in throughput as capability is built.

Three things to carry through the draw

Name the dip before it arrives. Tell your board, your team, and yourself that things will feel like they are going backwards before they go forwards. Pre-frame the J-curve as expected rather than surprising. Shift conversations from avoiding the dip, to minimising depth and duration. For example, provide employees with personal access to some AI tools to create familiarity and understanding ahead of business deployment.

Measure learning, not output. During the draw phase, output volume is the wrong signal. What matters is whether the organisation is accumulating genuine capability: documented processes, clean data, defined control gates, working feedback loops, explicit operating agreements.

Do not bank productivity gains ahead of capability. The productivity gains are real, but they come after the investment is made, capability is deployed, and operation is tuned.

The draw is not evidence that the approach is wrong. It is evidence that real change is underway.