Nigerian AI Startup Decide Unveils DAX-1 Model to Make Spreadsheet Editing Faster, Cheaper

By Adedapo Adesanya

Decide, the Nigerian startup building AI tools for spreadsheet automation, is taking steps to develop some of its own artificial intelligence capabilities, reducing its reliance on third-party models.

Startups can build AI products without spending years training a model from scratch by integrating existing models and building products around them. But as usage grows, depending on third-party models can become increasingly expensive, pushing some startups to develop their own models or AI infrastructure.

The company has introduced DAX-1, its first specialised AI model, designed to edit spreadsheets faster and at a lower cost than the larger models the startup previously relied on.

Before developing DAX-1, Decide used a mix of AI models from OpenAI, Anthropic and Google to power different parts of its product. The startup has now begun bringing some of that intelligence in-house with a model built specifically for spreadsheet tasks.

DAX-1 is designed to handle tasks including fixing formulas, removing duplicates, correcting spreadsheet errors and formatting workbooks.

According to Mr Abiodun Adetona, Decide’s co-founder, DAX-1 represents how the company is evolving, with the model being the first major step in that direction.

“DAX-1 represents a shift in how we are building Decide. We don’t want to simply wrap general-purpose AI models and depend on them for every spreadsheet operation. We believe some of the most repetitive and high-volume spreadsheet tasks can be handled more accurately and economically by models trained specifically for those workflows,” he told Business Post.

According to Mr Adetona, the model, which was post-trained specifically for spreadsheet editing, achieved 91.7 per cent accuracy on its benchmark, while running five times faster and at 97 per cent lower cost than Anthropic’s Fable 5.

“We post-trained it specifically for spreadsheet editing, and on our benchmark it achieved 91.7 per cent accuracy while being 5× faster and 97 per cent cheaper to run than Anthropic’s Fable 5,” he added.

The company said the results demonstrated the model’s ability to handle spreadsheet-editing tasks with greater speed and cost efficiency than the competing model.

“That efficiency matters because our goal is to make AI-powered spreadsheet automation economically viable at scale.”