^{READ TIME: 5 MINUTES}

In this week’s edition of Strategies For Effective Data Leadership, I’ve invited a client and buddy of my, Jared Novack to write guest piece. Jared is the Co-Founder and Chief Data & Information Officer at Upstatement, a brand and digital product studio, where he designs the data systems that run the business. His article today is about strategies for solving the downstream data quality issue by tackling the problem at the source using AI. Data quality is a problem faced by every single one of us, so Jared’s article is well worth a read!

Find him on LinkedIn or at jared.uno

How much of your week is spent mopping up messy data? Empty rows. Missing values. Duplicate records. Mixed-up associations. Keeping it clean can feel like the whole job. But now agentic coding tools make it cheap to design interfaces that improve data quality and change how the rest of the organization thinks about the data team.

Data leaders often find themselves in the role of data janitor: mopping up data gaps in the dark of night so things can be spick and span come morning. Maybe you've tried the other variants: nagging, apologizing, or over-engineering correction tools to repair broken inputs. But these are all after-the-fact moves.

Normally, outputs get the glory: fancy dashboards, executive reports, beautiful models and warehouses to service it all. Meanwhile the inputs that feed them just piled up: forms, CRM fields, file drops, spreadsheets. Outputs get brand colors and iteration because they're visible and exec-facing. Inputs? Those are plumbing, someone else's job.

Indeed, every place data enters your organization is an interface, designed or not. Ours were mostly “not.” So we spent our energy fixing bad data after it arrived while the real opportunity sat upstream. Design the point of entry and the cleanup starts to disappear. 

Recently I've launched several tools to attack this challenge head-on. One is called "The Activator." It lets a project manager drag-and-drop a signed contract into a web interface. From there, Claude extracts all the key information from the SOW and cross-checks it against the CRM. The PM reviews the details and in one click, activates the data into all the downstream systems: setting up project spaces, linking up data sources, building invoice schedules. The human's job flips from filling out endless forms to final eyes on specifics.

This suite of processes used to take about a week, with hand-offs across at least three people. Inevitably, steps got skipped or numbers transposed. Now it's one person in one sitting. A single contained process without re-keying or other error opportunities. We even added a leaderboard so you can see who the champion activator is. Gone are my days of mopping up missed fields and data gaps. Now I'm thinking about what else can ride on that same activation step. 

Why didn't we just do this years ago? Because this was actually a crazy challenge that would take weeks of design and engineering time. Well, it used to. Claude Code has turned (hundreds of) hours of human time into a few hours of planning, testing and review. It's not instant, but agentic coding crushes the math on whether an initiative like this is worth the investment. There are no doubt dozens (hundreds?) of "great idea, but not worth the ROI" projects sitting in your backlog, all priced with last year's math. As the cost of engineering collapses, especially in system connections and integrations, that "not worth it" list shrinks by the week. It’s time to re-run the numbers.

The payoff goes beyond clean data. Own the inputs and your users (and fans) multiply: not just the execs reading dashboards, but every PM and BD rep who touches your systems daily. Instead of being the data-minder, you might start to hear: "my data team just made my life easier." You also now own critical eyeball space: a tool open on someone's screen every day, while reports and dashboards might languish off to the side. That interface you built can run in both directions: taking that contract in while handing back insights at the moment of action. Take an inventory this week: how many places does data enter your org, and how many were designed on purpose? Pick one. Claim it. Stop cleaning; start designing.

By Jared Novack

Free resources:
Data Leadership Playbook · Data Career Compass · Data Leader Assessment

Tristan Burns
Find me on LinkedIn