DEMO

Real-Time Seller Motivation — Streaming on IBM Confluent

Streams 18+ sales and HR systems through IBM Confluent in real time, then turns those live signals — joined with history and Databricks in place — into a personalized seller nudge.

Building Blocks used:

ai.engineering

data.lakehouse

automation.operate

Looking for Building Block skills?

Skills provide best practices for IBM technologies

Download skills →

Products Used

1 Review use case

Companies that run sales incentive and recognition programs sit on some of the most fragmented — and most perishable — data in the enterprise. A single seller's story is spread across a CRM that changes by the second, a collaboration tool where peer recognition happens in real time, an HRIS that lands once a night, partner files dropped on object storage, and a lakehouse another vendor already owns. Eighteen-plus systems arrive at four different velocities in four different formats, and none of them agree. Worse, the usual answer — batch everything into a warehouse overnight — means the seller gets coached on yesterday's picture: the deal they closed this morning cannot influence the nudge they receive this afternoon.

This demo makes every one of those systems a live event stream. All 18+ sources publish onto IBM Confluent across three Kafka topics — Salesforce change-data-capture and Slack events fire the instant something happens, and the batch and file sources land on the same backbone instead of a parallel pipeline nobody maintains. A single Kafka consumer dual-writes each event as it arrives: into IBM DataStax for low-latency operational reads of the live "now", and into Apache Iceberg tables on IBM Cloud Object Storage for the historical "always". On camera, a Slack recognition or a Salesforce deal update is queryable roughly two seconds after it happens, with no nightly window anywhere in the path.

The watsonx.data Presto engine then answers one SQL statement that joins that freshly streamed event with full history and the customer's existing Databricks Delta tables — read in place over open Iceberg interop, with zero rows copied and zero tables migrated, so the customer keeps Databricks while IBM lands the streaming, integration and serving layers. IBM watsonx.ai (Llama 4 Maverick) turns the unified result into a personalized nudge plus an action list priced in points that adds up to exactly the seller's remaining gap, and every task cites the signal that produced it. The same payload renders into the three surfaces sellers already work in: the program's Elevate app, a Salesforce Lightning component, and a Slack Block Kit DM. Three seller personas ship with the demo — a veteran expert, a developing contributor, and a mid-year team transitioner. All data is 100% synthetic and seeded; the streaming backbone and every engine processing it are real IBM services.

Start Here

Watch this short overview

Preview the application

Explore key screens from the demo

2 Try the interactive application

Experience the interactive demo and explore its features. A demo script is provided to help when presenting the application to partners.

Try the demo

Extra demo assets

Have questions, want a walkthrough or found a bug?

Contact the demo owner for assistance.

Submit a request →

3 Explore how it was built

The demo runs on a real-time streaming spine. Events from 18+ heterogeneous sources — Salesforce (CDC), Slack (real-time), Workday (nightly batch), object-storage file drops and a dozen more SaaS systems — publish onto IBM Confluent across three Kafka topics, so every velocity is normalized into one ordered event flow rather than a separate pipeline per source. A single Kafka consumer performs a dual-write, splitting each event as it lands into IBM DataStax for low-latency operational reads of the live "now" and into Apache Iceberg tables on IBM Cloud Object Storage for the historical "always". Publish to queryable takes roughly two seconds, with no nightly batch window anywhere in the path. The watsonx.data Presto engine then answers one SQL statement that joins all three stores: the live DataStax tables, the Iceberg lakehouse, and BIW's existing Databricks Delta tables, which are read in place through open Iceberg interop with zero rows copied and zero tables migrated. That query-in-place federation is the commercial thesis of the demo — the customer keeps Databricks, and IBM lands the streaming, integration and serving layers without a rip-and-replace. The unified result is enriched by IBM watsonx.ai (Llama 4 Maverick), which generates a personalized seller nudge and an action list priced in points that sums exactly to the seller's remaining gap, with each task citing the source signal behind it. The same payload renders into three delivery surfaces — the Elevate program app, a Salesforce Lightning component, and a Slack Block Kit DM. The whole application is containerized and runs on IBM Cloud Code Engine, and was built with IBM Bob from natural-language specifications.

Architecture

4 Download and modify with IBM Bob

Download the source code and specifications for this application to set up and run it locally. Once downloaded, open the code in IBM Bob to modify and adapt it to your own use case.

Source Code

Demo Script — Real-Time Seller Motivation on IBM Confluent

Runtime: 6–8 minutes
Audience: CDO, data architect, VP of sales operations
Setup: Open the demo URL and let it load before the call starts. The console runs in three acts; the act selector is at the top of the page.

Before You Start — The One-Line Frame

“Your seller data is spread across eighteen systems that arrive four different ways, one of them is a lakehouse you already bought from someone else, and most of it is a day stale by the time anyone can use it. We’re not going to migrate any of it. We’re going to stream all of it — and query it two seconds later.”

Act 1 — The Tangle (~2 min)

1. Open on Act 1. Eighteen source cards are on screen — Salesforce, Slack, Workday, S3 file drops, Databricks, and thirteen more.

2. Point at the velocity badges: CDC, STREAM, BATCH, FILE. Say the quiet part — a nightly warehouse load cannot serve a program where recognition happens in Slack in real time and headcount lands from Workday at 2 a.m. The mismatch is not a data-volume problem, it is a latency problem.

3. Point at the Databricks card. It is the only one badged FEDERATED · query-in-place. Flag it now: “We will come back to this one, and we will never copy it.”

4. Land the counter row: 18 sources · 4 velocities · 1 open lakehouse · 0 migrations.

Act 2 — The Unification (~3.5 min) — this is the Confluent act, give it the time

5. Switch to Act 2 and scroll straight to the live stream panel. Click Start 30-second stream. Salesforce CDC and Slack events begin landing in their Confluent topics; leave it running while you talk.

6. Narrate what is moving: The biw.salesforce.cdc and biw.slack.events counters ticking up, the per-seller bars filling for Rachel, Marcus and Maya, and rows appearing in the live event feed. “Every row on that feed is a message that just hit Kafka.”

7. Point at the cluster line under the feed — a real IBM Confluent Cloud bootstrap over SASL_SSL. This is not a recording and not a mock: it is a managed Kafka cluster accepting production-shaped traffic while we talk.

8. Follow one event end to end. Scroll up and click + Slack recognition. Watch the pipeline light up left to right: source → Kafka on IBM Confluent → the dual-write processor → IBM DataStax (live “now”) and Iceberg on watsonx.data (history “always”).

9. Read the latency chip out loud — the event lands in about two seconds.

“One event, written twice, on purpose. There is no ETL hop between ’now’ and ‘always’, and no overnight window between the seller doing something and the platform knowing about it.”

10. Scroll to the federated join panel. Walk the SQL: one statement, cassandra.biw_live FULL OUTER JOIN iceberg.biw FULL OUTER JOIN databricks.main.biw_demo. The event that landed ten seconds ago is already in the result set.

11. The money shot. Point at the gold badge: Databricks — read in place · 0 rows copied.

“Your Databricks investment stays exactly where it is. The streaming layer is new, the lakehouse is open, and the incumbent is read in place through open Iceberg formats. Nothing you put in is ever trapped — not by Databricks, and not by us.”

Act 3 — The Payoff (~2 min)

12. Switch to Act 3 and note the lock-up line: everything from here is seller-facing, and the platform stack never appears on it.

13. Rachel (Veteran Expert) loads by default. Read the nudge headline and body — generated by IBM watsonx.ai; the model badge is on the card. Then the reward bar: 800 points to go.

14. Scroll the task list. Every task is priced, and the four tasks add up to exactly the 800-point gap — the goal is never out of reach by design.

15. Click through Where the nudge lands: Elevate app → Salesforce Lightning → Slack DM → BluePoints / Thanks@IBM. One payload, four surfaces, identical copy.

16. On the Slack surface, point at the provenance chips under each task — Salesforce · deal closed, Databricks · historical influence.

“This is the whole argument in one frame: the nudge is only possible because three sources that never used to meet were streamed together moments ago, and it can show its work.”

17. Switch sellers to Marcus (Developing Contributor, 1,600 to go) or Maya (Team Transitioner, 2,100 to go) to show the same pipeline producing genuinely different coaching for a different persona.

Close

“Eighteen sources on one event backbone, queryable two seconds after they happen, joined with a lakehouse we never migrated — and a seller who gets told the one thing that moves them, while it still matters. The streaming layer is the product; the nudge is just where you can see it.”

Handling Questions

“Why streaming — why not just load it nightly?” Because the payoff is perishable. A nudge that reacts to this morning’s close lands; the same nudge tomorrow is noise. Confluent is also what lets eighteen different velocities share one backbone instead of eighteen bespoke pipelines, which is the maintenance cost nobody budgets for.

“Is this real or is it a mock?” Real engines, synthetic rows. Every frame showing a number carries a persistent SIMULATED label. Kafka, Cassandra, Iceberg, Presto and watsonx.ai are all live services; the data is seeded (SEED=20260812) so the demo is byte-identical on every run.

“Are you asking us to leave Databricks?” No, and the demo is built so you cannot accidentally imply it. Databricks is read in place, zero copy. IBM wins the streaming, integration and serving layers; the incumbent keeps the rest.

“Where does the AI actually sit?” Thin and last. watsonx.ai writes the nudge copy and the task list from data the platform already unified. It is provenance-forward on purpose — the AI serves the data story, it doesn’t replace it.