AI agent for product catalog operations
Keep product facts, merchandising changes and feeds in agreement
Built the growth engine that took cnvrg.io to its acquisition by Intel.
A product description can be polished while its size, category or feed value is wrong. I personally build catalog workflows that connect approved product facts to reviewable copy and merchandising changes, then check the accepted destination instead of assuming a successful request fixed every surface.
Choose which source owns each product fact
The product team, store and feed can disagree about the same item. I map product and variant identities, field authority, locales and downstream destinations before defining a change. Copy can be drafted from approved facts; unavailable specifications stay missing rather than becoming invented features.
The scope may cover titles, descriptions, categories, merchandising attributes or feed consistency. Price, stock, promotions, publication and order operations are separate decisions with different APIs and business consequences. They enter the implementation only when expressly agreed and supported.
A catalog action needs a field and destination
| Situation | What I implement | What we inspect |
|---|---|---|
| Descriptions repeat vague claims | Source-backed copy drafts with fact references | Accurate details and unresolved missing information |
| Items use inconsistent categories | A reviewed mapping and exception queue | Accepted category per product, with ambiguous cases retained |
| Store and feed disagree | Field-level comparison using stable product/variant IDs | The changed source and the downstream accepted value |
| A batch needs ongoing maintenance | Approved changes, current-state checks and failure reconciliation | Exact field updates, partial results and named repair owner |
An ordinary rule may resolve a known code or category. An agent can help interpret permitted product text and prepare a proposed description or classification. It should not guess a missing safety claim, compatibility statement or variant relationship.
Keep a copy edit separate from changing price or publishing
A local draft has not changed the store. Saving a product field is a real write, even when the product remains unpublished. Updating a variant price, inventory quantity or publication state is a different operation; permission to edit a description is not permission to change those values.
I inspect the selected commerce platform, installation scopes, product model, supported mutations and feed connections before building. Product updates do not automatically publish an item or establish variant-price support. Customer and order analysis needs its own protected-data and history access where applicable.
Illustrative example: one product title, two variant identities
A feed record resembles the store’s parent product, but its identifier belongs to one variant. This is an illustrative acceptance case, not an executed client change. The workflow preserves both identifiers, shows the differing attributes and routes an ambiguous mapping for review rather than overwriting every variant from the parent text.
Once the mapping is accepted, it applies only the agreed field change to the correct destination. The subsequent readback checks that object. Feed propagation may lag or fail independently, so the downstream check remains open until there is evidence for that surface.
Work with Yaniv
Built the growth engine that took cnvrg.io to its acquisition by Intel.
Make batches recoverable without overwriting newer work
I personally build the selected source checks, proposal queue, integration and acceptance tests. Each change record keeps the source fact, object/variant identity, field, prior value and accepted new value. A timeout or partial batch result is reconciled before retrying.
Recovery checks the current destination before restoring a prior value; a later valid edit must not be erased blindly. The operating rules name who handles ambiguous facts, rejected fields, new categories, changed permissions and source conflicts. Publication and commercial changes stay with their designated owners.
Deliver a usable operating process, not a pile of generated copy
The handoff includes field/source definitions, approved mappings, integration configuration, test cases and exception/recovery instructions. I can operate and improve the agreed catalog process when ongoing work is included. Your product and merchandising owners approve facts and consequential choices; Advisory is the separate advice-only offer.
Measure accepted field changes, unresolved source conflicts, feed discrepancies and review/correction effort using the same eligible product population. Sales and margin changes need separate observation. If product truth is not ready, I can personally investigate that gap before scaling content production.
For broader store growth and commercial priorities, see ecommerce growth.
Questions before the build
Can you create product copy and implement approved changes?
Yes. I personally build the agreed catalog workflow, prepare source-backed copy or mappings and implement approved supported changes. We define the fields, destinations, acceptance checks and operating responsibilities for the actual client or project.
Does this include Shopify?
Shopify can be a selected implementation where the installed app, granted scopes and current supported APIs fit the work. I inspect the product/variant model and downstream feeds before choosing operations. Other commerce platforms are assessed against their actual capabilities.
Can it change prices, inventory and publication status?
Only when those separate operations are explicitly scoped, supported and authorized. Product-description access does not grant price, stock or publication authority. I define the resource, limits, reviewer and recovery behavior before implementing a consequential change.
What happens when the product facts are incomplete?
The workflow identifies the missing fact and the source or person needed to resolve it. It can prepare a draft using supported information, but does not fill a missing specification, compliance claim or compatibility fact with a plausible invention.
Will updating the store fix every feed immediately?
Not necessarily. Downstream feeds can have separate processing, mappings and errors. We check the agreed destinations and keep propagation or acceptance gaps visible. A successful source write is not proof that every sales channel changed.
How does this connect to customer analysis and lifecycle work?
Data analysis examines eligible order/customer evidence and its limits. Lifecycle operations coordinates permitted customer actions such as recovery or renewal. Catalog operations owns product facts and approved merchandising changes; customer messaging or order changes need their separate scope and permissions.
Personal strategy and implementation
Choose the catalog inconsistency worth fixing first
Bring the product source, store/feed destinations and one permitted example of a mismatch. I will define the field authority, build or investigation, review rules and acceptance evidence for the scope.
Choose the connected work