Valenta Told Kingston to Optimise Before You Automate. I've Watched Caribbean Companies Skip That Step for Years

Anonymous, French, 18th century, Design for a Machine, 18th century. The Metropolitan Museum of Art, public domain.

This month a Toronto-headquartered automation firm told a room of Kingston executives not to buy its product yet. Valenta co-founder and CEO Jayesh Kasim was in Jamaica for sessions attended by people from GraceKennedy Limited's Foods, Financial Group and IT Operations divisions. "For us, we solve business problems using technology," he said. "Our methodology is to always optimise before we automate." Vendors rarely lead with that.

Valenta's Caribbean franchise is run by Aurora Technologies Limited, founded by Erica Anderson. The firm reports that finance processes taking about eight hours a day usually fall to one or two once the workflow is fixed and automation is added afterwards. About 80 percent of its first Caribbean engagements start in finance: accounts receivable, accounts payable and payroll, the repetitive rules-based work Kasim recommends automating first, with decisions left to people. Simpler projects go live in three to five weeks. Roger Grant, Valenta's managing partner for Jamaica and the Caribbean, says each digital assistant is tested with the client's subject matter experts before it reaches production.

Automation repeats whatever process it is given

My training is in physics, which teaches you to ask what a system does to its input rather than what you hope it does. An AI model cannot tell a well-run finance process from a badly run one. It learns the pattern in front of it and runs that pattern faster, with fewer people in a position to notice when it goes wrong.

Give it eight hours of manual reconciliation built around a workaround nobody has questioned in five years, and you get that workaround running at machine speed. It is also harder to remove, because it now lives in a system instead of in someone's habits.

I have seen the same ordering mistake at board level through StarApple AI's training work across the region. In the study we published on our board-level AI programme, organisations that completed the training cut the time to set up AI and data governance from 11 to 15 months to 6. The main reason it had taken so long before was that boards approved AI initiatives before anyone had worked out what governance each one needed.

Charles Thurston Thompson, French Machinery, 1855. The Metropolitan Museum of Art, public domain.

What optimise-then-automate involves in practice

1. Start with high-volume, rules-based work where the right answer is not in dispute.
2. Fix and document the process before building anything.
3. Test the digital assistant with the people who do the work, not only the engineers, before it touches live data.
4. Keep judgement calls under a named person's authority.

Anderson described the benefit in human terms: "You're giving them time to go home. You're giving them time to be with their kids, and reducing the stress and panic at the end of each reporting cycle." Valenta holds ISO 9001 and ISO 27001 certification and says it does not use client data to train its models, which matters in a region where data protection law is still new.

None of this is new to a competent systems integrator. What stands out is a vendor saying it in Kingston, where the usual pitch is to buy the platform and let the process problems sort themselves out once data flows.

What PwC's 2026 barometer adds

PwC published its 2026 AI Jobs Barometer the same week, and the Jamaica Gleaner carried the local reading on 8 July. The figures are global:

• jobs requiring AI skills are growing nearly eight times faster than the job market as a whole;
• workers with AI expertise earn a 62 percent wage premium;
• among organisations most exposed to AI, those leading in its use recorded 163 percent labour productivity growth;
• entry-level roles in AI-exposed occupations have grown 35 percent since 2019, which runs against the fear that AI simply removes junior jobs.

PwC Jamaica's Hugh Thompson said AI is raising the value of human capabilities, and the workers who gain combine AI literacy with critical thinking, communication and leadership. His colleague Jossett Francis Wint said the bigger story is how jobs are changing. The wage premium goes to people who understand a process well enough to know what the tool should do and to spot when it is doing something else.

Jamaica's readiness assessment found the same gap

On 1 April 2026 the government launched its UNESCO AI Readiness Assessment Methodology report at Jamaica House. It drew on close to 200 people, including researchers, entrepreneurs, officials, young people and community representatives, with funding from the European Union, and scored the country on five dimensions: legal and regulatory, technological and infrastructural, economic, social and cultural, and scientific and educational.

It found foundations in place: the Data Protection Act, the Cybercrimes Act, a National AI Task Force set up in 2023 and National AI Policy recommendations delivered in 2024. Dr Andrew Wheatley, the minister responsible for Science, Technology and Special Projects in the Office of the Prime Minister, named what is missing: a National AI Oversight and Implementation Council, AI education from early childhood through tertiary and vocational levels, and a closed gender gap in STEM. He has directed the National AI Task Force, of which I am a member, to prioritise AI literacy across government, saying: "Jamaica does not need AI hype; we need AI governance."

That is Valenta's advice at the scale of a country. If the institutions and literacy are not built first, AI deployed by government will sit on top of the gaps the assessment has now documented.

Andrew Joseph Russell, Lifting Engine "Vibbard" from Draw at Long Bridge, March 1864, March 1864. The Metropolitan Museum of Art, public domain.

Model pricing and open standards reward the same discipline

OpenAI released its GPT-5.6 lineup on 9 July 2026 as three priced tiers rather than one flagship model:

A finance team that has mapped its process can pick the cheapest tier that handles each step. One that has not will default to the most expensive and pay five times more per token than it may need to.

In December 2025 the Linux Foundation formed the Agentic AI Foundation. Anthropic contributed the Model Context Protocol (MCP), alongside Block's goose framework and OpenAI's AGENTS.md, with platinum members including AWS, Google, Microsoft, Bloomberg and Cloudflare, and Snowflake at Gold level. MCP had passed 97 million monthly SDK downloads by then. A shared standard for connecting AI tools to business systems means a small Caribbean firm no longer has to pay for a custom integration with every vendor. It still has to know which systems and processes the integration should serve.

"Mi tell client dem the same thing every time: no bother buy the shiny AI tool if the process underneath it a mess. The AI don't fix the mess. It just run the mess faster, and now it cost more to unwind." - Adrian Dunkley, AI Boss

Cheap automation has made the ordering mistake cheaper to make and more expensive to undo once it runs at machine speed. The part I cannot yet show is the counterfactual: Valenta's eight-to-two-hour figures are the vendor's own, and I have not seen independent before-and-after data from a Jamaican client. A firm that runs the pilot below will produce it.

What a board should do before signing an automation contract

1. Process owner: map the target process end to end and time it for two weeks before the request for proposals goes out, recording volume, handling time, error rate and how often cases need an exception.
2. Procurement lead: require every shortlisted vendor to run a paid pilot of 4 to 8 weeks measured against that baseline, with a written success threshold, instead of a demonstration.
3. Data protection officer: complete the data governance review, including the processor contract and where data is stored, before the contract is signed.
4. Head of the business unit: list the decisions in the process that need judgement and name the person who keeps each one after automation.
5. Board or audit committee: review the vendor's price line by line against the cost of an internal build, and plan in weeks for simple automations and months for anything touching governance.