Seeking vs. Surfacing
Welcome to the journey from Visual KPI to the all new Transpara real-time operational intelligence platform. The new platform is already being battle-tested with major customers (let us know if you want to take part), but is also still in development. The documentation will be updated over time as new features are released. Stay tuned for more updates!
AI in Transpara works two ways: sometimes you ask it something, and sometimes it proactively tells you something you didn't notice or ask about. The distinction sounds small, yet it shapes almost everything about how the platform behaves and how much value it delivers.
Seeking is you asking. You have a question, you ask it in plain language, and the AI answers using live data from your model.
Surfacing is the system watching itself and telling you. Agents watch your operations and data continuously, find something worth knowing, and bring it to you before anyone has asked.
Both modes run on the same tools, leverage the same graph, and ground every answer in real data. The only difference is who starts the conversation, and that difference matters more than it appears to.
Seeking
In Seeking, you are in control and the AI acts as a fast analyst on demand.
You already seek every time you navigate to a site to check a KPI or search for a machine or metric by name. AI chat is the fastest and most flexible form of Seeking. In one question, you can ask about a single KPI, a whole site, a division, or the entire operation.
You ask how a division is performing, why a KPI dropped overnight, what happens if current trends continue, or for a dashboard that lays out a specific investigation. The AI works out what to look at, reads your live data, and comes back with an answer and the visuals behind it. Your questions can look at the past, the present, or even the future with forecasting and prediction.
What makes this new is how fast the answer arrives and how little effort it takes. "Why did overall equipment effectiveness drop on line 3 last night, and what did it cost us?" isn't a hard question for a competent engineer. It's a slow one, because answering it means pulling values from several systems, checking the calculation chain, comparing against a baseline, and doing some arithmetic on production value. That work takes hours. When the answer takes seconds instead, the questions that were never worth hours become worth asking.
Seeking has one requirement: you have to know what to ask. An operation that runs on Seeking alone still depends on somebody noticing or suspecting something and then posing the right question.
Examples of Seeking
- "What changed overnight?"
- "When did vibration start to increase on this compressor?"
- "What caused the drop in yield at Plant 2 this week?"
- "Which sites are underperforming their peers, and why?"
- "If current trends continue, which KPIs will cross their limits before the end of the shift?"
- "Build me a dashboard for the issues at Plant 3."
Surfacing
In Surfacing, the system does more of the work for you and at a much higher scale.
The simplest form of Surfacing is an alert you configure yourself: if this KPI goes above 10, tell me. The KPI raises its hand, but only because you told it when to. Agentic Monitoring removes that "because you told it to" requirement.
Tiny monitoring sub-agents run continuously against your model and data. They look for anomalies, KPIs approaching their limits, drifting data, peer outliers, stale data and more. An advisor agent combs through all of the sub-agent work to decide what is worth surfacing to users and keeps detailed records for analysis. When something is significant, the advisor agent can connect the related findings, estimate impact, and even write briefings and create dashboards.
Agents also look for patterns that no single KPI reveals on its own. Every KPI might look fine on its own while a set of related assets are all heading toward a problem you can't see in isolation. One asset drifting may be normal. Twenty related assets drifting together is a problem worth raising before any of them crosses a limit.
The case for this is arithmetic. A large operation can carry hundreds of thousands of KPIs and other metrics. No one can scan that many, and nobody tries. In practice, people watch a few dozen, or at most a few hundred, on a handful of dashboards and hope the important thing is among them. Agents watch all of it, continuously, as a system, and watching one more KPI costs close to nothing.
Surfacing also does work that Seeking can't, because working hours don't limit it. Findings accumulate overnight, and the briefing is ready before the morning meeting, so nobody has to assemble it during the meeting.
The agents rank and score what they find, and ranking is a judgment. A finding you consider trivial may land near the top, and something you care about may not clear the bar. That's why the feedback loop matters: when you mark a finding as wrong or not useful, the system applies that judgment the next time it sees similar conditions.
Examples of Surfacing
- "Eight pumps on the same cooling loop are drifting away from their normal operating range. None has crossed a limit yet, but the shared pattern points to a common condition worth checking."
- "Kiln 4 is running well below the other kilns of the same type while they hold steady, which suggests an asset problem rather than a process problem."
- "At its current rate, disk usage on the historian server will reach its limit in 13 days."
- "Fourteen KPIs at the north site stopped updating at 02:10. The timing points to a connection problem rather than a process problem."
- "Fleet availability at mine 07 fell sharply overnight. The cause traces to the underground fleet, and three dependent KPIs are cascading from it."
When you ask an open-ended question such as "Do I have anything to worry about right now?" or "How is the operation doing this morning?", you start the conversation, but you don't say what to look at. The AI scans the model and brings back what it judges most significant, just as Surfacing does. You still had to think to ask, but the answer can come from anywhere in the model.
Why an Operation Needs Both
The two modes aren't alternatives. They work together, and the combination is where the real value lies.
A finding surfaces overnight: a site is trending toward a major KPI breach, and the briefing estimates what it will cost. You read it in the morning. You want to know something the briefing didn't cover, so you reply to it: "Why do you think this is happening?" or "Is this happening anywhere else?" Now you're Seeking, but inside a context that Surfacing established. You didn't have to know where to look. You did have to decide what mattered next.
That's the loop worth designing around. Surfacing sets the agenda. Seeking works the agenda. Transpara makes findings conversational for exactly this reason: replying to a finding puts you in Seeking mode without changing screens or restating context. You can also @mention a colleague in the same thread, so the people who need to act see the evidence and the discussion in one place.
| Seeking | Surfacing | |
|---|---|---|
| Who starts | You | The system |
| Runs | On demand | Continuously or on a schedule |
| Scope | What you ask about, from one KPI to the whole operation | Everything in the model |
| Good for | The question in front of you | The question nobody asked |
| Depends on | Knowing what to look for | What the agents judge significant |
| Where it lives | AI chat in tView, tStudio, and Excel, or any MCP client | The activity feed, the KPI or site it concerns, briefing dashboards, alerts, and scheduled briefings by email or Microsoft Teams |
Most of this page describes the end-user experience in tView. The same two concepts apply when you configure and administer the system in tStudio.
Seeking in tStudio often means asking for work, not just answers: "List every KPI that has no limits or target," "Add geolocation to the plants that are missing it," or "Create a calculation that does the following." The AI assistant does the work, and the changes stay under your control.
Surfacing in tStudio means the system tells you when your configuration needs attention: KPIs that stay stale or in an unknown state, KPIs that sit out of range so often that their limits may be wrong, or a data source that has stopped responding.
Why Surfacing Is Less Common
Most industrial software vendors now offer AI, and most of that AI is Seeking. The imbalance has a reason, and the reason isn't that nobody thought of Surfacing or put the pieces in place to make it possible at scale.
Seeking is comparatively easy to ship. Add a chat window, route the question to a model, return the text. It demos well. If an answer is wrong, the user is right there to notice, and it's never clear whether the tool or the question was at fault.
Surfacing requires things a chat window doesn't:
- Operational semantics. An agent deciding whether a value is a problem needs to know the limits, the targets, the operating mode, history and what normal looks like for this asset. Raw numbers aren't enough.
- A traversable model. Explaining a finding means tracing causes through relationships. The model has to hold those relationships, whether you define them in calculations or the graph finds them itself through shared attributes, and it has to walk them fast. It has to think like an entire system that works together. That's why a knowledge graph sits underneath and is critical.
- Continuous execution. Something has to run all the time, at scale, without a person triggering it.
- Ranking that survives contact with users. Surfacing without prioritization is a firehose, and people mute a firehose within a week. Scoring, confidence, and synthesis keep it readable.
- Somewhere to deliver. Findings need a home in the product that people already look at. Otherwise they aren't findings, they're logs.
- Tolerance for being wrong in public. People hold a finding they never asked for to a higher standard than an answer to their own question. That takes grounding you can defend and a feedback loop that visibly improves.
Transpara built these in from the start, which is why Surfacing is a core part of the platform rather than an add-on. For the machinery, see Agentic Monitoring Service.
What's next?
- Industrial AI for Real-Time Operations for how AI fits into the platform as a whole.
- Agentic Monitoring Service for the agents behind Surfacing.
- Working with AI for Seeking in practice.
- Activity Feed for Surfacing in practice.