The Instrument Before the Answer

We set out to rebuild a website. What actually mattered was the order in which we let evidence, not opinion, make each decision.

We set out to redesign a website. Before we had chosen a single colour, we had scored dozens of competitors on seven dimensions and found that our proposition was not the problem — our routing was. The redesign got smaller, cheaper and sharper the moment the evidence arrived, and that reversal is the most useful thing I have learned about buying expertise you do not hold.

Every regulated firm in Jersey eventually has to commission work it cannot personally grade — a custody migration, a CRM replacement, an outsourcing arrangement, an AI deployment. In each case you are buying a specialism you do not own, from a supplier who owns it completely, and you remain accountable for the outcome. The uncomfortable truth is that when you cannot judge the craft, you fall back on taste — “make it feel more premium,” “that does not look like us” — and taste is precisely the thing a capable supplier can out-argue in any meeting.

I have written throughout this collection that AI in a regulated business is a leadership and operating-model decision first and a technology decision second. This instalment is the same argument told through a single, ordinary project — a website rebuild — because the sequence of decisions we happened to take turned out to be a repeatable method for governing any specialist supplier. The website is incidental. The order of thinking is the point.

PwC has found that regulatory and data-protection concerns are the primary blocker to outsourcing in financial services — which means the instinct to hand a specialism entirely to a supplier is exactly the instinct a regulated firm cannot indulge. The question is never whether to rely on a supplier; it is how to govern one you cannot personally grade.

What follows is that project in the order it actually happened. At each stage I have tried to name the one thing that moved the project forward — the differentiator — because in every case it was not effort, budget or a better opinion. It was a small, deliberately built instrument that converted an argument into evidence and let the evidence decide.

Stage one: the assumption we started with

The project began the way these projects always begin: “our website is tired, we should redevelop it.” That sentence sounds like a decision, but it is really a default. It frames the work as visual — a nicer-looking version of the thing we already had — and it quietly hands the most important judgements to whoever holds the design skill, because visual quality is the one axis a non-specialist client feels able to comment on.

The first move that changed the trajectory cost nothing. We refused the visual framing and rewrote the question. Not “how should the site look,” but “what does a prospective client actually learn, infer and conclude about us before they ever pick up the phone?” That reframing did more than any budget line that followed, because it moved the centre of gravity away from taste and towards evidence.

A redesign brief is not a design decision. It is a positioning decision wearing a design costume — and until you undress it, you are letting aesthetics answer a commercial question.

Stage two: the instrument that narrowed the field

Having changed the question, we needed something better than internal opinion to answer it. So before commissioning any creative work, we built a market-analysis instrument. We took the relevant population of comparable Jersey firms and reviewed every one of them the same way — the same public sources in the same order, the same fixed set of prompts, the same seven weighted dimensions covering how clearly each firm positions itself, how it signals credibility and trust, how coherent it is across its channels, its visible thought leadership, and how ready it looks the moment before a first meeting.

The discipline mattered more than the scoring. Every firm was assessed against identical criteria, run in tranches so quality did not drift, with observed fact, inference and simple absence of evidence kept strictly apart. What we ended up with was not a list of opinions about competitors. It was a coordinate system — a map on which we could locate ourselves objectively rather than anxiously.

The finding reversed the project. Our proposition was already strong, particularly for the audiences that matter most to us. What was weak was not the substance but the routing: how a visitor is guided to the right service, how proof is made visible, and how the highest-value journeys convert. That single result removed an enormous amount of scope. We were no longer buying a rebuild; we were buying sharper navigation and stronger proof on a proposition that already worked — a smaller, cheaper, lower-risk piece of work.

This is the heart of the whole story. The instrument did not decorate the decision. It made the decision — and it made it by taking scope away, which is the opposite of what a supplier pitch is designed to do.

Stage three: from competitors to audiences

A competitive map tells you where you stand against other firms. It does not tell you how you land with the specific people you are trying to reach. So the next instrument turned the lens ninety degrees: we assessed how our current public presentation was likely to be received by each of the audience groups we actually care about — affluent private clients, the professional gatekeepers who refer them, and planning-led prospects.

That analysis surfaced something a competitor benchmark never could: places where our tone or structure was unintentionally narrowing our appeal, closing a door on an audience we wanted open. The differentiator at this stage was the shift from “us versus them” to “us versus the audience we want.” The first tells you whether you are winning. The second tells you what to change.

Stage four: handing over a decision, not an opinion

Only now did the supplier receive a brief — and because of the two stages before it, the brief was different in kind from the usual wish-list. It set out an audience hierarchy, page priorities, a proof architecture, delivery standards and post-launch content expectations. It told the agency what commercial job each part of the site had to do, and left the craft of doing it to them.

This is the quiet governance move most supplier relationships miss. We did not hand the specialist our analysis to interpret, nor our taste to indulge. We handed them a decision, expressed as objectives, against which their expert work could be measured. A brief built from evidence is not a request for a supplier to impress you. It is a standard the supplier has agreed to be held to.

Stage five: the standing reviewer, and the one line that mattered

As deliverables started to arrive, we needed to review them consistently over months, not in a single burst of enthusiasm. So we built a standing review instrument: a fixed hierarchy of controlling documents so that when sources disagreed we knew which one governed, a consistent scoring scale, and — crucially — a rule that separated a genuine shortfall by the supplier from a limit we had imposed ourselves through budget, time or our own lack of expertise. You do not penalise an agency for a constraint you created; but you do record its effect on the outcome.

The single most important instruction in the entire project was one line in that reviewer: if a promised artefact is not actually present, do not infer it, do not simulate it, and do not score it — say plainly that it is missing. That line exists because the failure mode of any confident reviewer, human or machine, is to fill a gap with plausible invention. In a compliance-sensitive review, plausible invention is the worst possible outcome, because it reads exactly like diligence.

This is where the AI lesson of this collection lands in practice. A review instrument earns trust not by how much it produces, but by its willingness to refuse. An automated reviewer that will not manufacture a finding it cannot evidence is doing the first job of the first line of defence — and it is the only kind you can safely put in front of a regulated sign-off.

This is not a private preference. The JFSC’s 2026 guidance on the use of AI in Jersey’s financial services sector asks firms to keep staff who can verify outputs against source data and spot fabrication, and to keep a competent person responsible for unusual cases — the exact discipline that one line enforces. The Jersey AI Playbook says the same thing in operating-model terms under its human-AI partnership principle: for anything beyond low risk, AI output is untrusted until a named human validates it and records that validation. An instrument built to flag rather than invent is how you make that principle real instead of aspirational.

Stage six: the checklist as a gate, not a list

When the test site arrived, the temptation was to circulate it and collect comments. Instead we converted the review into a gated sign-off. Every item had to be cleared against evidence, with a severity scale so that a stray full stop and a mis-stated regulatory permission were never treated as equivalent. A short set of issues were designated hard blockers — launch could not proceed while any remained open — and sign-off required two independent keys: the business owner and, separately, Compliance.

The difference between a checklist and a gate is accountability. A checklist is a list of things someone looked at. A gate is a controlled event with named owners, an evidence trail and a defined set of conditions that must be true before the firm commits. A sign-off is not a group nod. It is a decision the firm can later show it took deliberately, with reasons — which is the same standard we would apply to any material decision under scrutiny.

The shape of that gate is not arbitrary either. The JFSC guidance sets out a materiality ladder, and for a higher-impact use case — anything that could materially affect customer outcomes, regulated activity or market-facing presentation, which a public-facing website plainly does — it expects senior approval, independent validation and challenge, and evidence-ready governance. Two independent keys, a severity scale and hard blockers are simply that expectation expressed as a working instrument. The Playbook’s managed-transformation principle adds the other half: real stage gates that are genuine stop/go decisions, not milestones you glide through.

Stage seven: what survived contact with reality

A control that has never met the real artefact is a theory. Applied to the live test site, the gate immediately caught the things a round of casual comments would have waved through: small copy errors in legal pages, technical settings that were correct for the finished site but wrong for a test environment, and placeholder content standing in for people and proof that did not yet exist. None of these would have been visible to taste. All of them were visible to an instrument that required evidence rather than an impression.

That is the final differentiator, and it is the one people skip: a control is not real until it holds on the messy, half-finished, live version of the thing. Everything before that point is preparation. The test is contact.

The pattern, in one view

Read the stages back to back and the same move repeats every time: a default — an opinion, a preference, a supplier’s confident recommendation — replaced by a small instrument that put evidence in its place. This is the reference card I now keep for any specialist engagement, not just websites.

Why a website story is really a migration story

Nothing in that table is about web design. Replace “website” with “custody platform,” “client system” or “AI model” and the method is unchanged. The firms that struggle with major change are rarely short of effort or budget; they are short of instruments — the small, disciplined artefacts that force a decision onto evidence rather than negotiate it on confidence. Gartner’s finding that AI value stays trapped until organisational change catches up points at the same gap: the return does not come from the tool, it comes from the method around it.

That is the quiet case for AI in a regulated firm. Not that it designs or predicts — but that it makes these instruments cheap to build at every decision point, and a well-governed one can be held to the standard that matters most: it will tell you what it does not know. The Jersey AI Playbook calls disciplined, responsible adoption a source of ethical differentiation — competitive advantage in a jurisdiction built on trust. A firm that cannot out-expert its suppliers can still out-govern them. The advantage is not the specialism. It is the sequence.

The next time you commission expertise you cannot personally grade — a migration, a platform, a model — will you bring a better opinion to the table, or will you bring the instrument that lets the evidence decide before a single colour is chosen?

A companion case study to the series, showing the collection’s method at work on a single real project; future instalments will return to specific governance instruments in more depth.

Sources: practitioner experience of a Jersey financial-services website redevelopment, described in anonymised and generalised form; JFSC, Guidance on the use of AI in Jersey’s financial services sector (2026), including its materiality ladder; Jersey AI Playbook (Government of Jersey et al., 2026); PwC on regulatory and data-protection concerns as the primary blocker to outsourcing; Gartner, Value Is Trapped: Unlocking AI ROI Through Organizational Change (2026).

Professional disclaimer: The views expressed are my own and are shared for general informational purposes. This content is intentionally non-confidential and does not include proprietary, client-specific, or employer-sensitive information. It is not legal, regulatory, investment, or professional advice; readers should seek appropriate advice for their circumstances.

Financial services leader (20+ years) | Lean Six Sigma Green Belt | London School of Economics (LSE) AI Leadership Accelerator

A version of this article was first published on LinkedIn as part of the Sequence body of work. This is the canonical version.

The views expressed are the author's own and shared for general information. This content is intentionally non-confidential and does not include proprietary, client-specific or employer-sensitive information. It is not legal, regulatory, investment or professional advice. © 2026 Paul Soares. Sequence is the trading name of Sequence Advisory Limited (in formation) · hello@sequenceadvisory.com