AI vs Manual Product Data: The Real Cost Comparison for Retailers
Before Grosvenor Flooring deployed merchi.ai, they had a 1,000-product backlog. These products were sitting in their catalogue without descriptions, imagery, or accurate attributes. Each product was invisible to search, generating no organic traffic, and converting at a fraction of its potential. Clearing that backlog manually would have taken months. With merchi.ai, it was cleared, and the retailer went on to achieve 976% online revenue growth driven substantially by the organic content that AI made possible.
That gap between manual and AI is not unique to Grosvenor Flooring. It is a structural reality that every retailer with a large or growing product range eventually hits. This article breaks down exactly where the difference lies. If you want a primer on what AI merchandising actually is before diving into the numbers, start with What is AI merchandising?.
The head-to-head comparison
| Dimension | Manual | AI (merchi.ai) |
|---|---|---|
| Speed | 30-60 min per product (writing, QA, data entry) | Under 30 seconds per product |
| Cost | £15-25 per product (copywriting, photography briefing, QA) | A fraction of manual cost at scale |
| Accuracy | Variable - depends on copywriter, brief quality, review cycles | Consistent - same rules applied to every product |
| Scale | Linear - more products means more headcount or agency spend | Non-linear - handles 10 or 10,000 products at the same rate |
| Consistency | Drifts over time - different writers, different interpretations | Locked to schema - brand voice and attribute structure never drift |
| Language support | Expensive - separate translation workflows for each language | 40+ languages from a single processing run |
Breaking down the cost
Manual product data entry
The £15-25 per product figure is not an overestimate. Here is where it comes from:
- Copywriting: A competent retail copywriter typically produces 20-30 product descriptions per day at full quality. At a freelance rate of £300-400 per day, that is roughly £10-15 per product for copy alone.
- Briefing and review: Someone has to brief the writer, review the output, request revisions, and approve. Add 10-15 minutes of internal time per product.
- Data entry: Descriptions need to be entered into the PIM or CMS. Attributes need to be mapped to the correct fields. Taxonomy needs to be assigned. For a product with 20 attributes, this is another 15-20 minutes.
- Photography and lifestyle imagery: Even basic product photography runs to several pounds per image when studio time, editing, and retouching are included. AI lifestyle imagery (placing the product in a room scene) was simply not feasible at scale before AI.
For a catalogue of 1,000 products, that is between £15,000 and £25,000 in content costs before a single page goes live. For 5,000 products, you are looking at £75,000-£125,000.
AI product data entry
With AI merchandising, the economics are fundamentally different. Costs scale with usage, not headcount. A retailer with 10,000 SKUs can process their entire range in the time it would have taken a manual team to complete 50 products. The output is consistent from product one to product ten thousand.
There are no revision cycles driven by inconsistent interpreting of briefs. There are no onboarding delays when a copywriter leaves. There are no backlog queues when a new range arrives.
Where manual still has a role
This is not an argument that human input disappears. It does not. Here is where manual effort remains valuable:
Brand-defining hero products. A flagship product that anchors a marketing campaign may warrant a bespoke, human-crafted description with more creative latitude than AI typically applies.
Complex compliance copy. Products in regulated categories (medical devices, financial services adjacent goods) may require specific legal review that goes beyond what AI configuration should handle.
Editorial content. Blog posts, buying guides, and long-form editorial are a different class of content from product descriptions. Human editorial judgment applies here.
What AI merchandising replaces is the commodity work - the 10,000 attribute fill-ins, the 1,000 baseline product descriptions, the daily feed of new products that need content before they can go live. That is the work that creates backlogs, inflates costs, and slows retailers down.
The backlog problem: why speed compounds
Speed advantage is not just about cost per product. It is about what happens when products sit without content.
A product with no description, no correct attributes, and no lifestyle imagery:
- Does not appear in Google Shopping
- Does not rank in organic search
- Cannot be filtered correctly in on-site navigation
- Converts at a lower rate even when found
Every week a product is live but undescribed is a week of lost revenue. For Grosvenor Flooring, clearing a 1,000-product backlog did not just reduce content costs. It unlocked traffic and conversions from products that had previously been invisible. The 976% online revenue growth was not from a marketing campaign. It came from products that, for the first time, had content good enough to be found and bought. Read the full Grosvenor Flooring case study.
Quality: is AI product data good enough?
This question comes up consistently. The answer is yes, with appropriate configuration.
AI product data quality depends on:
Input quality. Better product images and raw specifications produce better output. Garbage in, garbage out applies here as it does everywhere.
Schema configuration. merchi.ai is configured to the retailer’s exact attribute schema and brand voice. This is not generic AI output - it is output constrained to the retailer’s specific requirements.
Human review layer. For most retailers, AI output is reviewed by a merchandiser before publish, at least initially. Review time drops dramatically as trust in the system builds. The review function shifts from creation to curation.
In practice, AI-generated product descriptions from a well-configured system are consistently better than the average of a mixed manual team - because the manual average includes bad days, misread briefs, and rushed work from overloaded writers.
Multi-language: where the gap widens
If your product catalogue needs to serve multiple markets, the manual cost comparison changes dramatically.
Every language requires either a competent native speaker writing from scratch, or a translator working from the English source. Either way, you are multiplying the per-product cost by the number of languages.
merchi.ai generates content in 40+ languages in a single processing run. The same product is described in English, French, German, and Dutch simultaneously, with each version adapted to the linguistic conventions of that market - not just translated.
Work out what it costs your business
If you have more than 500 products and you are managing content manually, the numbers in this article apply to you. Start a free trial or book a 30-minute call and we will walk through your specific catalogue size, current process, and what switching to AI would look like in practice.
Frequently Asked Questions
Is AI product data really as good as human-written content?
When properly configured to a retailer’s brand voice, attribute schema, and category conventions, AI-generated product data is consistently high-quality - and more consistent than the average of a mixed manual team. The Grosvenor Flooring deployment produced content that drove 976% online revenue growth, which demonstrates that the quality is not just good enough for the site - it is good enough to rank and convert.
What does manual product data entry actually cost?
Including copywriting, briefing, QA, data entry, and basic imagery, manual product content typically costs £15-25 per product. For a catalogue of 1,000 products, that is £15,000-£25,000 before any enrichment or lifestyle imagery is added.
How fast can AI process product data compared to manual teams?
merchi.ai processes a product in under 30 seconds. A skilled human copywriter with full quality review takes 30-60 minutes per product. For a 1,000-product backlog, that is the difference between a few hours and several months.
Does switching to AI product data mean I no longer need a merchandising team?
No. AI handles the commodity work - baseline descriptions, attribute population, taxonomy classification, feed generation. Merchandisers shift to higher-value work: configuration, quality oversight, editorial decisions, and the bespoke content that genuinely requires human judgment. The team gets more leverage, not replaced.
Can AI product data handle my specific taxonomy and attribute schema?
Yes. merchi.ai is configured to the retailer’s exact schema - including custom attribute fields, category hierarchies, and brand voice guidelines. The system does not produce generic output; it produces output that conforms to your specific requirements.