Example from a public annual report

Agree the strategy today. The work starts tomorrow.

A board agrees where the company goes. The work happens in departments, and more and more of it is done by AI agents. This page follows one board decision, market expansion, from the annual report to the work AI agents did on it.

One line of the annual report, followed to the work: “Expanding to new markets” (Revolut annual report 2025, p. 19), the initiative RELIABL.IT made of it, and the launch brief an AI agent wrote. The brief is hypothetical.

RELIABL.IT read Revolut’s public 2025 annual report, built the strategy and a plan for the whole company, and turned one decision into work for people and AI agents. The example is a launch, because everyone understands one. The same path works for any decision a bank, insurer or fintech board takes: a new market, a new product, a regulatory change.

Revolut is not a RELIABL.IT customer. The strategy comes from its public 2025 annual report, and its figures and results are its own. Revolut did not see or approve anything on this page. The launch brief in part 3 is hypothetical: the product and every Revolut detail in it are invented and marked as such.

1Strategy first

Start where the value is decided: where the company is going

RELIABL.IT read the annual report and wrote one executive summary and 12 strategy sections, from the core platform and products to risk, compliance and governance. Every gap, action and task after this links back to one of them.

  • 12strategy sections under one executive summary
  • 32questions the report leaves open, for management
  • 10company initiatives with dates and priorities
In the app: the executive summary, in the report’s own words. Where the report gives two figures for one thing, both stay, marked.

RELIABL.IT also marks what the report leaves open

All 12 areas were checked for targets without a baseline or a date, owners not named and decision rights not written down. Each gap comes with the question that closes it. For example: which measures decide whether a new product scales up or stops?

In the app: the open questions, area by area.

Not a verdict. Management knows most of the answers. The list shows what the plan needs on paper before work starts.

10 company initiatives. This page follows one: market expansion

Market by market: licences, mobilisation, launch. High priority, due October 2027. Every bank, insurer and fintech grows into new markets or new products, so this is the one we follow.

In the app: the 10 initiatives. The board shows which ones have work behind them, and the one that has none yet.

2Everyone aligned

Every department gets its part of the strategy, in its own terms

Nobody re-explains the strategy in each unit. Each department and each team gets its own version, tied to the company’s, so every team knows where to go and how it applies to its work.

In the app: company, then Marketing, then the Demand Generation team. Every action shows what it supports, up to the company.

One board decision, market expansion, becomes 15 actions in four departments

CompanySequence market expansion and readiness

Marketing

Define campaign intake and readiness standards

  • Demand Generation
  • Brand and Corporate Comms
  • Events and Field
  • Product Marketing (2 actions)

Product and Engineering

Embed market readiness into product delivery

  • UI/UX Design
  • Product Management
  • R&D
  • DevOps / SRE
  • Quality Assurance

Operations

Design operational readiness gates

  • Procurement

Sales

Implement sales readiness and information controls

83 tasks in all, every one aligned to the strategy the board agrees. In the three highlighted teams, AI agents did the first 16.

3Into work

Each decision becomes tasks for people and AI agents, and agents’ work is checked before it counts

Each action has tasks with owners, dates and measures. Some are for people, some for AI agents. An agent’s work counts as done only when it passes the task’s own checklist.

In the app: one agent task, from first attempt to done.
  1. First attempt: failThe agent reported its own draft as not done: fields could not yet be traced to sources and owners.
  2. Checked line by lineRequired fields, every field traceable to a source and an owner, scenarios tested. Each line passes or fails with evidence.
  3. Second attempt: passDone, and the result says what it rests on: labelled working assumptions, to be confirmed by the process owner.
  4. Some tasks stay with peopleThe system does not let an agent close work that a person must decide.

In three teams, agents drafted the standards a market launch needs, each ready for the owner’s review

Demand Generation

  • Who provides each of 13 launch inputs, and how fresh it must be
  • A campaign brief template
  • A launch gate with four conditions

Brand and Comms

  • Input sources and validation rules
  • A campaign brief template and workflow
  • A publication readiness checklist and a pilot plan

UI/UX Design

  • Local requirements per market in every design handoff
  • A register of unresolved issues
  • A handoff format for engineering

Where the company’s own inputs were missing, the agents wrote 35 working assumptions and labelled every one.

One output in full: the brief every launch campaign must pass before it reaches customers

The agent’s output: the campaign intake and brief template, first section.
  • 12 sections, from the growth priority to investment, brand, controls and approvals
  • Every field names its source, its owner and how fresh it must be
  • Financial-product claims, customer data and AI-assisted content each go to a named reviewer
  • An approved brief is not yet launch-ready. Only the launch gate says READY
  • Tested on four launch scenarios in a desk simulation: four of four as expected

Then marketing used it: a launch brief for a new Swiss business account, against a local challenger

A hypothetical example. The product and every Revolut detail in it are invented and marked as such.

The launch brief: what it is and where it stands.
The challenger’s public offer, line by line, with the claim we may make for each.
  • A Swiss challenger’s offer, read from its public website on 16 November 2026, line by line against ours, with the claim we may make for each
  • Goal, channels, investment and the payback rule it falls under
  • Control flags: product claims, customer data, AI-assisted copy
  • Every approval the brief still needs, and who gives it

And the brief says when not to launch: four conditions, checked per market, before anything goes public

Condition 1

Product available

Pending

Account in beta, capital deposit account in build.

Condition 2

Permissions

Unconfirmed

Licence not granted yet, so no licence or IBAN claims.

Condition 3

Operations ready

Unconfirmed

Support staffing and onboarding capacity not confirmed.

Condition 4

Investment approved

Partial

The budget part approved, the exception still open.

Gate: NOT READY. No publication in Switzerland, and no teasers that imply the product is available.

Also in the brief: public creative does not name the competitor, and a price comparison needs a dated, verified price on launch day.

The system writes the brief, and it also stops the brief from going out before the licence, the product and the money are confirmed by named people.

4The harness

Every agent works to written instructions from the strategy. When the strategy changes, so do they

Every agent gets written instructions taken from the strategy: what to do, what to check and where to stop. Agents get their instructions from the RELIABL.IT MCP server and run on the platform you choose. RELIABL.IT does not require integration with your core systems.

The agent instructions of one action.
What the agent must check, and where it stops.

What it must never decide

“Do not invent control requirements, reviewer identities, decision rights … or approval outcomes.”

Kept current by RELIABL.IT

When the plan changes, RELIABL.IT proposes the change, your team approves it, and the instructions for the affected agents change at once.

If someone else builds your agents, someone else has to change them every time your business changes.

Your company

The same run, on the strategy your company is executing now

What this example used

  • One public annual report
  • No owners, no internal targets
  • Assumptions where inputs were missing

What a run for you would use

  • The strategy your board agreed
  • One decision you choose, then the next
  • Your owners, your approval rules

What you get

  • The plan, aligned across every department
  • Agent tasks with written instructions, kept current
  • You decide what runs, and when

Start with one decision, measured against a baseline. You see the result before you decide on the next one.

What you get from your own strategy

  • 1The strategy in one clear structure, with the questions it leaves open
  • 2Every department’s part of it, in its own terms
  • 3Tasks for people and AI agents, checked before they count
  • 4Agent instructions that change when the strategy changes

Name one decision your board took this year.

A new market, a new product, a regulatory change. We run it through RELIABL.IT and bring the plan and one agent’s finished work.

In one meeting with the owner of that decision, you correct the picture. Then you decide what runs. Timing by agreement.

See it in action · Explore the demo account · darko.butina@reliabl.it

RELIABL.IT is the Management Operating System for the AI powered company. It turns your goals into an action plan for your people and AI agents within minutes.

How this example was made
  • Input: Revolut’s public 2025 annual report. No internal documents, owners or targets.
  • Timing: the strategy, gaps and plan were built on 14 November 2026. The agents did their work on 16 November 2026.
  • Counts: 12 strategy sections, 32 open questions in all 12 areas and 10 company initiatives, as the app showed them on 14 November. Market expansion had 15 linked actions and 83 open tasks on 16 November; six of those actions carry agent instructions.
  • The agents’ work: 16 tasks under three team actions (Demand Generation 5, Brand and Corporate Communications 6, UI/UX Design 5). On 16 November, 8 were done, 1 in progress, 3 left to people by design and 4 waiting for a person to approve a strategy first.
  • Assumptions: the company’s own inputs could not be collected for a public example, so the agents wrote 35 working assumptions, each labelled. In a real run, the owners supply them.
  • The template test: four launch scenarios run as a desk simulation by the agent, not validated with stakeholders.
  • The launch brief: hypothetical. An agent filled in the template for an invented product, a Revolut business account in Switzerland. Everything about Revolut’s Swiss plans, figures, prices and approvals is invented and tagged as such in the document. The competitor’s facts come from its public website, read on 16 November 2026; its name is left out.
  • Screens: captures of the app on this example. The demo user name and plan usage are hidden.
  • Results: Revolut’s figures and results are its own. This example shows how RELIABL.IT turns a strategy into work; it makes no claim about the company.