Real-time streaming data: a CDO playbook for getting it right

0:00Welcome to the Leaders Insights Podcast.

0:07Today's episode, real-time streaming data, a CDO playbook for getting it right.

0:12Most companies I talk to swear they're real-time now.

0:15Are they lying or do they just not know what the words mean?

0:19Bit of both.

0:20They've bought the plumbing and confused it with the water.

0:23Collecting streaming data, a continuous flow of events instead of a nightly batch upload,

0:28is easy.

0:29Acting on it in seconds is where 90% of them quietly fall over.

0:34Start me at zero.

0:35What actually is streaming data and why should an executive care?

0:38Picture your old approach.

0:40Every night at 2 a.m., you dump the day's sales into a warehouse and look at it tomorrow.

0:46Streaming flips that.

0:47Each swipe, click, or sensor reading arrives the instant it happens, and you can react

0:52while it still matters.

0:54The care part is simple.

0:55A fraud alert tomorrow morning is a police report.

0:59A fraud alert in 400 milliseconds is a declined transaction.

1:03Fine, but if it's so obviously better, why do so few actually do it?

1:07Because they build the cathedral before they know if anyone wants to pray in it.

1:11They stand up a huge event pipeline.

1:14Kafka, the tool most people use for moving these event streams around, and then go looking

1:19for a use case.

1:20Backwards, you end up with a very expensive river flowing into a field nobody farms.

1:26So what's the right sequence?

1:27Give me the playbook.

1:29Step one, find a decision that's currently too slow and cost you money for being slow.

1:35Not a dashboard someone wants prettier.

1:37A decision.

1:38Uber Eats reroutes couriers in real time because a cold burrito is a refund.

1:43That's the shape you want.

1:45How does a CDO, the chief data officer, the person who owns how the company uses its data,

1:51spot that decision?

1:52Ask one question everywhere.

1:54Where do we already know something useful but act on it too late?

1:59Retailers know a carts being abandoned.

2:01Banks know a cards behaving oddly.

2:04The knowledge exists.

2:05The reaction lags.

2:07That gap is your first project.

2:09Say you found it.

2:10What's step two?

2:11Measure the actual latency.

2:13The delay between an event happening and you doing something about it.

2:17Most people have never measured it.

2:19They guess a few minutes.

2:21Then you check and it's 40.

2:23Clavio, the marketing platform, has published that speeding up abandoned cart messages materially

2:29lifts recovery.

2:30Vendor numbers so salt them.

2:32But the direction is real.

2:34You can't shrink a number you haven't looked at.

2:36This is where people reach for exotic architecture, isn't it?

2:40And that's the trap.

2:42Step three is deliberately boring.

2:44Build the thinnest possible slice that touches one decision end to end.

2:48One event, one rule, one action.

2:51Prove it moves a metric.

2:53Resist the engineer who wants to stream all 900 event types on day one.

2:57He's not solving your problem.

2:59He's decorating his CV.

3:01What does one rule even look like in practice?

3:04If a high value customer's payment fails twice in five minutes, ping a human before

3:08they rage quit.

3:10That's it.

3:11No machine learning, no neural anything.

3:14A rule a 12-year-old could write.

3:16Running on live data beats a genius model that reports on Tuesday.

3:20When do the fancy predictions come in, then?

3:22Step four.

3:23Once the pipe is proven and trusted, now you layer on models that score events as they

3:28arrive.

3:29Predict churn.

3:30The share of customers about to walk.

3:32At the moment, they wobble.

3:34Not in a quarterly review.

3:35But you've earned that complexity by shipping something small first.

3:39Where do these projects actually die?

3:41Ownership.

3:42The data team builds the stream.

3:44Then nobody in operations is on the hook for acting on it.

3:47So it hums away, technically alive, achieving nothing.

3:51The corporate equivalent of a treadmill used as a coat rack.

3:55Every real-time project needs a business owner whose number improves or doesn't.

3:59Give me the one thing an exec should do Monday morning.

4:02Pick a single decision your company makes too late and put a stopwatch on it.

4:07Measure the real lag from event to action.

4:10Don't buy anything.

4:11Don't hire anyone.

4:12Just get the number.

4:13Nine times out of ten, it's embarrassing enough to fund the whole project by itself.

4:18That's a wrap.

4:19Fresh CDO briefings drop daily at MBA-training.com.