AI transformation programme for companies: from where you are today to your team’s full potential

The AI transformation programme takes a company from informal AI use to redesigned processes, where AI carries out part of the work and the team checks it. There are five stages, from mapping to roll-out across every department. Each stage starts only when the previous one has shown measured results.

For owners and directors of SMEs and groups in Mozambique who want the same team to do more, starting with finance.

Why is using AI piecemeal not enough?

In many companies each person uses AI in their own way: an email here, a summary there. That brings isolated gains but does not change how the team works. The step change happens when the process is redesigned so that AI carries out a part of it, with prepared data, approved rules and a person checking. Without that, AI depends on each individual’s initiative.

How much finance work can AI already do?

In March 2026, Anthropic, the company that develops Claude, published a study comparing two measures across 22 occupational categories: the share of tasks AI can do in theory, and the share where use of Claude at work has already been observed.

94%of business and finance tasks are within reach of AI, in theory.

29%is the share where real use is already observed, in Claude usage data.

  • Theoretical AI capability
  • Observed use (Claude data)
20%40%60%80%100% ManagementBusiness andfinance94% · 29%Computer andmathsArchitecture andengineeringLife and socialsciencesSocial servicesLegalEducation andlibraryArts and mediaHealthcarepractitionersHealthcaresupportProtectiveserviceFood and servingGroundsmaintenancePersonal careSalesOffice and adminAgricultureConstructionInstallation andrepairProductionTransportation
Theoretical AI capability and observed use in Claude data, by occupational category, as a percentage of each category’s tasks. Source: Anthropic, “Labor market impacts of AI: A new measure and early evidence”, Maxim Massenkoff and Peter McCrory, 5 March 2026. Chart redrawn by TRBusiness. Values were read from the original chart and are approximate.

In our experience, the gap closes one routine at a time: the company chooses where to start, redesigns the process around AI and trains the team to check the result. That is the work we do.

See the values by occupational category
Theoretical AI capability and observed use in Claude data, by occupational category, as a percentage of each category’s tasks.
Occupational categoryTheoretical capabilityObserved use
Management92%13%
Business and finance94%29%
Computer and maths94%33%
Architecture and engineering85%5%
Life and social sciences77%10%
Social services51%4%
Legal89%21%
Education and library62%18%
Arts and media84%19%
Healthcare practitioners60%4%
Healthcare support28%2%
Protective service31%2%
Food and serving16%1%
Grounds maintenance4%1%
Personal care18%1%
Sales62%26%
Office and admin90%34%
Agriculture15%1%
Construction17%1%
Installation and repair18%1%
Production19%1%
Transportation12%1%

Read the Anthropic study (opens in a new tab)

What are the five stages of the programme?

  1. Mapping and knowledge

    We start with meetings with the manager of each area, to understand the processes, the time they take and what the board wants to solve first. In parallel, an introductory course shows the team how AI works and what it can do in practice. This stage works as the AI diagnostic: the financial proposal comes only after it.

  2. Initial structuring and pilot plan

    Together we choose one routine for the pilot and work through it in four steps.

    1. Pilot plan definition and approval. Which routine, with which data, and how success is measured. The scope is approved before we start.
    2. Mapping the new processes. We redesign the routine so that AI carries out a part of it, with rules and checkpoints.
    3. Real-world application. We run the new process on real data from a month already closed, side by side with the current process.
    4. Measurement and handover. We measure hours before and after, and the checked quality. The routine is documented and handed to the team.
  3. Assisted operation

    The team starts running the routine day to day, with our support. We fix what comes up, refine the rules and make sure the process works without us.

  4. Automation

    What repeats and is already stable starts to run automatically, connected to the company’s systems. The team keeps the checking and the decisions.

  5. Roll-out to every department

    With the method tested and the numbers in hand, we take the same work to the company’s other areas, one at a time.

We move to the next stage only when the numbers are in hand.

Where we start: eight finance routines

These eight routines are known uses of AI in finance. In each one, AI prepares the work and the team checks and decides.

  • Month-end closeGather schedules, check them and close in days, not weeks.
  • ConsolidationSeveral businesses and formats, one comparable view.
  • Variance analysisActual vs budget, with explanations and questions.
  • ReconciliationsBank, intercompany and suppliers, with exceptions flagged.
  • Treasury13-week cash forecast, updated at each close.
  • Pay and collectSupplier invoice triage, priorities and collections follow-up.
  • Tax and auditFiles, document requests and checks prepared.
  • Board reportingManagement commentary and dashboard ready for the meeting.

Where does human oversight sit?

Where does human oversight sit?
TaskHow AI helpsWhere human oversight sits
Month-end closeCollects schedules, checks totals, lists what is missingThe team approves every adjusting entry
ConsolidationConverts formats and aligns charts of accountsThe controller validates the account mapping
Variance analysisHighlights variances and proposes questionsOwners explain; the board decides
ReconciliationsMakes the obvious matchesThe team handles the exceptions
TreasuryUpdates the forecast with known dataThe treasurer reviews the assumptions
Pay and collectClassifies invoices and prepares follow-upsPayments and messages go out only once approved
Tax and auditOrganises files and checklistsThe team and tax advisers sign off
ReportingWrites the draft commentaryThe finance director reviews and signs

What do you receive during the programme?

  • Process mapA picture of the current work, with timings, owners and priorities.
  • Approved pilot planScope, data, success criteria and timetable, approved before we start.
  • Redesigned, documented routinesEach new routine is written down, with its rules and checkpoints.
  • Before-and-after measurementHours and quality measured on the company’s real data, not estimates.
  • A trained teamTraining applied to the real tasks of each area.

Who does the work, and where has it been done?

Igor Mosquini Bastos, founder of TRBusiness

TRBusiness was founded by Igor Mosquini Bastos, an economist with an MBA in Finance. He personally leads the AI diagnostics and the transformation programmes.

  • 20+ years in financeAcross Africa and Brazil.
  • More than a decade in MozambiqueFinance leadership in multi-entity operations.
  • Today · TRBusinessBusiness-transformation consulting and AI applied to processes.

How it starts: the AI diagnostic

The programme always starts with the AI diagnostic, carried out with the board and the manager of each area. At the end you receive three things:

  • Process mapHow the work is done today, who does it and where time is lost.
  • Three priority routinesThe routines worth starting with, and why.
  • Pilot planThe chosen routine, the data it needs and how the result is measured.
  • Each stage of the programme is contracted separately and moves forward only with measured results.
  • The first conversation has no cost.

What happens to the company’s data?

We start with fictitious or sample data. Real data enters only with business accounts of the AI tools, which do not train models on your data, with minimal access to the folders needed and a signed confidentiality agreement. The access rules are written into the pilot plan before the work begins.

And when the process needs tools?

Sometimes a redesigned process needs more than AI. In those cases we build and connect what is missing, within the programme and only when the process calls for it.

  • Mobile paymentsM-Pesa and e-Mola connected to the website or the system, with automatic confirmation.
  • WhatsApp customer serviceAutomatic replies and hand-over to the team.
  • Sales process and CRMEvery customer, proposal and follow-up visible to the board, including what goes through WhatsApp.
  • Management indicatorsThe numbers the board uses to decide, updated without assembling Excel files.

Frequently asked questions

Where does an AI programme start?

It starts with the AI diagnostic, in stage 1. We meet the manager of each area to understand what they do, how long it takes and what the board wants to solve first. In parallel, the team takes an introductory course. Only after that do we choose the pilot routine and present the financial proposal.

How long does it take?

It depends on the number of areas and on the routine chosen for the pilot. The first stage is short, so it does not tie up the team. The pilot uses data from a month already closed, so results arrive quickly. The timetable for each stage is in the plan approved before we start.

Do I need my data to be organised?

You do not need to start with everything in order. Organising the data and the rules is part of the work in stage 2. What we need is access to the files and to the people who know the process. If the data is very scattered, that shows up in the mapping and goes into the plan.

Does AI replace the team?

No. AI carries out the repetitive part of the work, and the team checks, corrects and decides. The aim is for the same team to do more and spend more time analysing than gathering data. Every routine has a human checkpoint before the result goes out.

What happens to internal control?

Internal control stays in place and becomes more visible. Each new routine has defined checkpoints, an owner for every approval and a record of what was done. AI does not approve entries or payments: it prepares the work for whoever approves. Segregation of duties stays exactly as your company defines it.

Does it suit groups with several companies?

Yes, and that is usually where the gain is largest, because consolidation and reporting across several entities take a lot of time. Each company keeps its own indicators; AI combines, compares and explains, without flattening what is different. The group finance team still decides how entities are presented and which differences matter.

Do we have to change our accounting software?

No. AI works with the files and exports your accounting system already produces, so the team keeps its current software, chart of accounts and habits. If it later makes sense to connect a routine directly to the system, that belongs in the automation stage of the programme, after the routine has proved itself.

Which AI do you use?

We work mainly with Claude, from Anthropic, because it is the model we know best in financial work. We have no affiliation with Anthropic. The method works with other models, and the final choice depends on your company’s tools and rules.

The first step is the AI diagnostic.

In a few meetings we understand your processes and the board’s priorities. The proposal comes afterwards, with your own company’s numbers.

Last updated: 18 November 2026