341. Most Brands Don't Know Their SKU-Level Profit (and Why AI Categorises Transactions at 92% at Best) - Parag Mamnani, Webgility
Selling online is easy now. Making money is the hard part. Parag Mamnani says the brands that win know their numbers at SKU level, close to real time, and keep a human accountable for them.
Parag is founder and CEO of Webgility, which helps small and medium sized businesses automate their e-commerce accounting and inventory. He has spent more than 20 years in e-commerce, building online stores for merchants after the dot-com crash and later working on Amazon's Seller Central side, where the same question kept coming up: how do I close my books?
The problem is no longer how to sell
Parag told Will Laurenson that every tool needed to launch a brand is available to everyone. What separates brands now is distribution, a unique product, a community and an operation that can scale. The fastest growing brands obsess over their product and their numbers.
Why SKU-level profit matters
Most merchants lump data together to estimate profit by channel. Parag said a real answer needs clean attribution from traffic and ads through conversions, orders, returns and expenses, down to the SKU. Waiting a month or a quarter for the books means reacting too late. With numbers closer to real time, a brand can order more stock for products that perform, adjust prices to lift velocity and spot cash flow problems early. Almost every catalogue has winners, losers and loss leaders, and managing them depends on seeing which is which.
How silos stack up costs
Will described how teams make changes in isolation. One runs a promotion, another changes the shipping threshold, another changes the returns policy. Each looks sensible alone, but the costs stack. Parag agreed that end-to-end visibility is the fix. Returns are a good example, because a conversion rate win means little if it also lifts returns.
Where the profit hides
Parag sees two areas most often. The first is the middle of the catalogue. Teams focus on the top sellers and the worst performers, but mispriced or badly set up items in the middle hide the biggest opportunities. The second is surprise fees. Returns, chargebacks, marketplace fees and storage costs can land long after the order, so unless fees are attributed to the right period, channel and SKU, the P&L looks lumpy and a month can look more profitable than it was.
Accounting as the system of record
Every team uses different dashboards and spreadsheets, so Parag sees the accounting system as the unifying layer. A sensible chart of accounts and the 8 to 12 expense lines that matter, tracked over time, give a clear picture without needing to be an accountant. Webgility now has accountants on staff alongside its software and agentic features, so merchants can hand over the books and get regular reconciliation and optimisation ideas.
What AI can and cannot do
Parag has run business and personal life with agents for a couple of years and still finds that they do not know what they do not know. The best benchmarks for transaction categorisation reach about 90 to 92% accuracy, and nobody wants books that are 92% right. AI also tends to agree with the user, and the reading of tax rules can differ from person to person, so accounting needs human judgement and accountability.
On agentic commerce, Parag asked who actually requested it. Agents suit staples and discovery, while brands that want to control the experience will still need shoppers on their own site. Accountability for agents going rogue sits with the buyer, and Parag expects another spike in returns.
Key takeaways
- Track profit at SKU level, with ads, returns and expenses attributed to each product.
- Join up the teams, and check how a promotion, shipping threshold and returns policy stack up before changing them.
- Look at the middle of the catalogue and the timing of fees before chasing the top and bottom products.
- Build an accounting structure with a short list of expense lines, so non-accountants can read it.
- Use AI for speed, but keep a human accountable for the numbers.
Parag also covers good enough attribution and why perfect reconciliation has been a moving target for years. Listen to episode 341 of the Customers Who Click podcast for the full conversation.
.png)
.png)


.png)
