Business intelligence has a blind spot baked into its DNA, and we stopped noticing it because we had no choice.
BI works on structured data. Rows and columns. Numbers that fit in a cell. Revenue, units, dates, quantities. Give it a clean star schema and it will slice and total and trend it beautifully. That is what it was built for, going back to the relational ideas Edgar Codd laid down in 1970 and the warehouse patterns Ralph Kimball formalized two decades later.
But look at what that leaves out. The call where a customer said they were frustrated. The email where a buyer hinted they were shopping around. The support ticket that explained why an order got cancelled. The contract clause that changed the margin. None of that fits in a column, so BI never saw it. We built an entire discipline around the half of the company that was easy to count, and we quietly agreed to ignore the half that was hard.
That second half is where AI changes the math.
The half we could measure
To be fair to BI, the quantitative half is genuinely important, and getting it right was hard. Consolidating data from a dozen systems, agreeing on what a number means, making it fast and trustworthy. That work is real and it is the foundation everything else sits on. I am not waving it away.
But notice what kind of question structured data can answer. It is very good at what happened. Revenue fell. Orders dropped. Margin compressed in one plant. It can show you the shape of the change with total precision.
What it cannot tell you is why. The why is almost never in the numbers. The why is in a sentence someone said on a call, or a complaint that came in three weeks before the orders stopped. BI hands you a perfect description of a problem and stays silent on the cause, because the cause lives in text it was never able to read.
The half we threw away
Here is the part that should bother you. By most estimates, something like four-fifths of what a company stores is unstructured. Text, mostly. Emails, documents, transcripts, notes, tickets, chat logs, contracts.
So the discipline we call “data” has, for forty years, mostly meant the small structured slice, while the large unstructured majority sat in inboxes and file shares as dead weight. Not because it was worthless. Because we had no economical way to interrogate it at scale. Reading ten thousand support tickets to find a pattern was a project, not a query.
That constraint is gone. A capable language model reads unstructured text the way BI reads a column. You can ask ten thousand tickets what customers complain about most this quarter and get an answer in minutes. The thing that was uneconomical is now cheap. And the moment something uneconomical becomes cheap, the whole calculation of what is worth doing changes.
Where the value actually shows up
Let me make this concrete, because “use your unstructured data” is the kind of advice that sounds good and changes nothing.
Think about customer churn. Your warehouse can tell you, with precision, that a customer’s orders fell off a cliff in March. Useful, and too late. The early signal was not in the order data. It was in a call in January where the buyer sounded annoyed, and an email in February that went unanswered. Those signals existed. They were just locked in text nobody was reading as data.
Now you can read them. An agent can scan call transcripts and flag the accounts where the tone shifted, weeks before the numbers move. That is not a faster dashboard. That is a question BI structurally could not ask, because the input was invisible to it.
Or take a sales call where a customer raises a concern. The old world: the rep writes a note, maybe, and the concern evaporates. The new world: pull the concern out of the transcript, take it to the warehouse to check whether it is even true, and find that the quality metric the customer is worried about actually improved last quarter. Two halves of the company, the qualitative worry and the quantitative truth, joined for the first time. That join was impossible when one half was unreadable.
Fusion is the real prize
The single sentence I want you to take from this: the value is not in the unstructured data alone, it is in fusing it with the structured data you already trust.
Qualitative on its own is suggestive but soft. A customer sounding unhappy is a hint, not a fact. Quantitative on its own is precise but mute. The order drop is a fact with no explanation. Put them together and you get something neither half could give you. The hint tells you where to look, the numbers tell you whether it is real, and together they tell a story you can act on.
This is the thing BI could never reach, by construction. It only had one half. The fusion is the new capability, and it is worth more than any speedup to the dashboards, because it answers the questions that actually drive decisions: why is this happening, what should we do, who is about to leave.
I will allow myself one example from the messy end. A company can compare a product that is losing money at one site against a near-identical product making money at another, using the structured cost data, and then read the maintenance notes and shift logs to find the operational reason for the gap. The numbers find the anomaly. The text explains it. You needed both, and until recently you could only have one.
What this means for your data strategy
If you run analytics or data for a company, this reframes the work in front of you.
Your warehouse is not the finish line. It is half the picture, the half you have spent years perfecting, and it is necessary but no longer sufficient. The next leg is bringing the unstructured majority into reach: the transcripts, the tickets, the emails, the documents. You do not have to cram them into a star schema. You have to make them queryable by an agent, which is a different and lighter task.
Be careful with the soft signals. Qualitative data is suggestive, and a model can over-read it, finding a pattern that is really just noise. So treat the text as the thing that raises a question and the structured data as the thing that confirms it. Hint first, verify second. That discipline keeps the fusion honest.
And widen who you think of as a data source. The contact center, the inbox, the meeting recordings, the contract repository. For years those were outside the data team’s remit because they were unreadable. They are readable now, and they hold the answers your dashboards have always pointed at without being able to give.
Running Tallyfy for over a decade, the pattern that stuck with me is that the important information about a business is mostly written in prose, scattered across conversations, not sitting in a tidy table. We built BI for the tidy tables because that was what we could handle. The prose was always where the truth lived. We finally have a way to read it. That is the real shift, and it is much bigger than another dashboard.
