Let’s start with the number that matters: 33.81%.
That is the score Decide Agent 2.0 posted on SpreadsheetBench V2, a global AI benchmark that tests how well AI models handle the kind of spreadsheet work businesses actually do every day. Financial modelling. Completing templates. Debugging formulas. Creating visualizations. Real tasks. Real pressure. Real evaluation.
And sitting above GPT-5.2 on that leaderboard is a Nigerian startup.
Above Gemini 3.1 Pro. Above DeepSeek V3.2. Above GLM-5.0, Kimi K2.5, Qwen3.5, and other models backed by some of the most well-funded AI research teams on the planet.
A team in Lagos, building with significantly less funding, significantly less infrastructure, and significantly less global attention than every company it just outranked, put its model on the same leaderboard and came out ahead.
That is not a small thing. That is the kind of result that makes you stop and actually pay attention to what Abiodun Adetona is building.
The result did not happen by accident. It happened because Abiodun spent months doing something that most AI companies skip in the rush to ship: listening carefully to the people actually using the product and rebuilding based on what they found.
The engine behind the score is DAX-1, Decide’s proprietary AI system developed specifically for business intelligence and data work. SpreadsheetBench V2 is designed to test exactly the kind of tasks businesses throw at AI tools every day, and DAX-1 handled them better than GPT-5.2 and Gemini on the official evaluation.
Decide Agent 2.0 is now available on the web, in Excel, in Google Sheets, and through its API.
Before the benchmark result, before the global leaderboard, there was a much simpler observation that Abiodun could not stop thinking about.
He grew up watching businesses, especially African businesses, make decisions that felt confident in the moment and turned out to be completely wrong in hindsight. Not because the people making them were careless. Because the information they were working from was incomplete, delayed, misread, or just sitting unexamined in a spreadsheet nobody had time to properly analyse.
The data existed. It always existed. The problem was always the gap between having numbers and understanding what those numbers were actually saying in time to do something useful with them.
Traditional business intelligence tools were built for companies with dedicated data teams, large IT budgets, and analysts who spend entire careers interpreting outputs. They were not built for the founder running a mid-sized business in Abuja who needs to know by Thursday whether to expand her branch or wait another quarter.
They were not built for the finance manager trying to understand why this month’s numbers look different from last month’s without spending three days rebuilding a model.
Abiodun’s answer was to build an AI that could sit on top of a business’s existing data: spreadsheets, financial records, databases, and answer real questions in plain language. Not produce dashboards that need explaining. Not generate charts that require interpretation. Actually answer the question you asked, with the reasoning behind the answer, based on your specific data.
“Most business failures are not caused by bad ideas,” he said. “They are caused by good ideas executed on bad information.”
Here is what Decide AI actually does in practice, stripped of any jargon.
You connect it to wherever your business data lives. Your sales records. Your financial system. Your inventory logs. Your spreadsheets. Then you ask it questions the way you would ask a knowledgeable colleague.
Which products are actually making money this quarter? Which customer segment is leaving fastest? If the current growth rate continues, when does this branch break even?
Decide reads the data, understands the question in context, and gives you an answer in plain language, with the reasoning, not just the conclusion. You do not need to know what a pivot table is. You do not need to export anything, build a formula, or wait for an analyst to come back to you. You ask, and the system tells you what your data is actually showing.
The difference between this and a generic AI assistant is important. A generic AI answers based on what it has been trained to know about the world in general. Decide answers based on your specific business data, your numbers, your customers, your history.
The intelligence is grounded in what is happening inside your business right now, not in general knowledge about businesses.
That specificity is what the SpreadsheetBench V2 evaluation was testing. Not general intelligence. Not how well a model writes an essay or explains a concept. Whether it can actually do the spreadsheet work that businesses need done, and do it accurately, reliably, and better than the alternatives.
Decide Agent 2.0 passed that test ahead of every major competitor on the leaderboard.
The types of problems Decide AI solves will feel familiar to anyone who has run or worked inside a business.
Inventory decisions that should be straightforward but somehow always end up being wrong because the sales data is spread across three different files and nobody has time to reconcile them.
Pricing questions that never get properly answered because building the model to test scenarios takes longer than the window to actually act on the result.
Customer analysis that reveals something important too late because nobody was watching the right signals. Financial planning that relies on assumptions nobody has properly stress-tested because the tools to do it properly are either too expensive or too complicated.
These are not exotic problems. They are the daily friction of running a business. And Decide AI is built to remove that friction, specifically for African businesses operating in environments with more volatility, more informal data collection, and more complexity than the enterprise software market typically designs for.
One early customer, a retail business running multiple locations, had been tracking inventory across branches using WhatsApp messages and shared Google Sheets. They knew the system was broken but had not found anything better that was actually simple to use.
After integrating Decide, they could see stock levels across all branches in real time, get alerts before stockouts happened, and understand which branches were performing differently and why, without anyone having to compile a report.
“I finally feel like I know what is happening in my business,” the founder told Abiodun.
That feeling of actually knowing, rather than assuming, is what Decide AI is built to give every business owner who uses it.
The global AI race has a familiar cast of characters.
OpenAI. Google. Anthropic. DeepSeek. Alibaba. Companies with billions of dollars in compute budgets, research teams of hundreds, and the kind of infrastructure that makes the work of building competitive AI models feel almost inevitable when you have access to it.
Decide AI does not have any of that. What it has is a specific, deep understanding of a real problem, the problem of business intelligence for companies that have data but cannot use it, and months of patient iteration based on feedback from the businesses actually using the product.
That combination was enough to land Decide Agent 2.0 above GPT-5.2 and Gemini 3.1 Pro on a globally recognised evaluation.
Abiodun’s longer ambition goes beyond individual businesses. He sees Decide AI eventually serving as an intelligence layer for African business at sector level, giving not just individual companies but investors, ecosystem players, and policymakers clearer, real-time pictures of what is actually happening across entire industries and markets on the continent.
The data to build that picture already exists. It is scattered across thousands of businesses, unconnected and mostly unread. Decide AI is building the infrastructure to surface it in ways that are actually useful.
Nigeria just showed up on a global AI leaderboard. Above companies that have been doing this longer, with more money, and with more people.
The team that made it happen is in Lagos. And they are just getting started.