Enterprise Intelligence Group
How to Build an Assortment Decision Support Agent in ChatGPT
A practical example of a ChatGPT agent that prepares evidence for SKU add, retain and delist decisions without replacing merchant judgment.
For enterprise retailers, workspace agents are the most relevant ChatGPT pattern for repeatable team workflows — reusable agents that can work across connected applications, operate on schedules, and follow workspace-defined permissions and approval controls. Availability and exact functionality depend on the retailer's plan, connected applications and administrator configuration.
Assortment decisions are rarely based on one number.
A low-selling item may serve an important customer mission. A profitable SKU may duplicate another product. A new launch may be underperforming because it never received the agreed distribution. A slow-moving regional item may be essential in a particular store cluster.
This is why simplistic AI recommendations — "delist the bottom 10%" — are dangerous.
The commercial value is not in automatically removing products. It is in assembling the evidence, history and exceptions required for a merchant to make a better decision.
An Assortment Decision Pack Agent can support add, retain, expand, reduce or delist reviews by preparing a structured evidence pack for every nominated SKU. It does not make the final assortment decision. It makes the reasoning easier to see, challenge and preserve.
Why this is an achievable early agent
Assortment optimization can become technically complex when it includes demand transfer, space elasticity, localization, constraints and predictive modeling. But the first agent does not need to solve the entire optimization problem.
A valuable initial version can: gather the available evidence · apply agreed screening rules · retrieve prior decisions · identify missing context · surface relevant exceptions · compare products consistently · generate a draft decision pack · capture the merchant's final rationale.
This creates value before the retailer attempts a fully automated assortment-optimization platform.
What does the agent actually produce?
For every SKU being reviewed, the agent creates a structured record containing: product and supplier information · sales, units, margin and rate of sale · store and cluster distribution · availability and out-of-stock context · promotional dependency · shelf-space or facing information where available · customer need or assortment role · comparable and potentially substitutable products · launch date and agreed evaluation period · previous add, retain or delist decisions · supplier commitments or contractual considerations · known local-market exceptions · evidence supporting each available option · information still required before a decision is made.
The outcome is a consistent decision-support pack, not an autonomous decision.
A high-level ChatGPT agent setup
1. Establish the decision scope
The agent should be configured for a clearly defined decision. For example:
Prepare an assortment decision pack for SKUs nominated for review in the selected category. Evaluate each SKU against the retailer's approved evidence framework. Present the case to retain, expand, reduce or delist. Identify exceptions and missing evidence. Do not select the final action or update any assortment system.
The agent must also understand that the retailer's approved decision framework — not a generic model — is authoritative.
2. Define the evidence framework
EIG would work with the merchandising team to document the factors actually used in assortment decisions:
- Performance — sales · units · gross margin · rate of sale · trend · promotional dependency. - Distribution and execution — authorized stores · selling stores · availability · out-of-stocks · weeks on sale · store-cluster performance. - Assortment role — customer need served · price tier · brand role · pack size · flavor, format or attribute · destination versus convenience item · good-better-best position. - Commercial context — supplier terms · funding · contractual commitments · private-label strategy · upcoming reformulation or packaging change. - Constraints and exceptions — minimum assortment requirements · local-market demand · regulatory considerations · seasonal relevance · space or fixture constraints · new-product incubation period.
The purpose is not to create one universal scoring formula. It is to capture how this retailer evaluates assortment choices.
3. Connect the relevant sources
A controlled implementation may use: product master data · approved SKU-performance exports · store and cluster attributes · assortment and distribution files · planogram or facing data · promotional history · supplier documentation · previous range-review decks · decision logs · category strategy documents · approved merchant rules and exceptions.
Workspace agents can be configured to use connected tools and information for repeatable workflows. The available applications, actions and scheduled capabilities depend on the organization's ChatGPT plan and administrator settings.
4. Configure the agent's workflow
Step 1: Receive the review population. The user provides a category and a list of SKUs — or asks the agent to prepare packs for products already nominated by an approved process.
Step 2: Validate the data window. The agent confirms the evaluation period, whether products have had enough weeks on sale, whether promotional periods distort the result, and whether distribution and availability are sufficient for a fair comparison.
Step 3: Assemble the evidence. The agent brings together performance, distribution, space, supplier and decision-history information.
Step 4: Compare like with like. Products are compared within relevant groups rather than against the entire category — equivalent pack sizes, similar price tiers, same customer need, same store cluster, comparable launch maturity, similar promotional support.
Step 5: Retrieve historical context. The agent searches previous reviews for answers such as: Why was the product originally listed? Was a performance threshold agreed? Was expansion conditional on supplier support? Has this product been reviewed before? Was a previous delist delayed? Is it part of an agreed assortment strategy?
Step 6: Present decision options. For each SKU, the pack presents evidence for retain, expand, reduce distribution, reposition, review after a defined period, or delist. The agent may identify a provisional evidence-led option, but it should clearly distinguish analysis from the final merchant decision.
Step 7: Flag exceptions and uncertainty. Examples: insufficient weeks on sale · poor availability invalidating rate-of-sale comparisons · missing supplier terms · conflicting product attributes · unrecorded local-market exceptions · incomplete substitution information.
Step 8: Capture the approved decision. After the merchant decides, an approved workflow can record the final decision, decision owner, rationale, evidence considered, exceptions applied, effective date and review date. Writing that record back to a business system should happen only through an approved application or action and under the retailer's required confirmation policy.
Example decision-pack output
- SKU under review — product, supplier, category, subcategory and current distribution. - Performance summary — the most relevant commercial measures, with comparison groups and period definitions. - Assortment role — what customer need the product serves and whether another SKU already serves it. - Decision history — what was previously decided and why. - Evidence by option:
| Option | Supporting evidence | Risks or exceptions |
|---|---|---|
| Retain | Stable cluster performance | High promotional dependency |
| Reduce | Weak performance in small stores | Strong in urban stores |
| Delist | Duplicated need and declining margin | Supplier agreement requires review |
| Reassess | New launch with limited weeks | More evidence needed |
- Missing information — anything the merchant needs before approving an action.
- Human decision — a clearly reserved section for the responsible merchant.
The most valuable output may be the decision record
What is currently ranged can be read from the assortment and distribution files. Why each item is there can be checked only if the reasoning was recorded when the decision was made.
Without a decision record, the same assortment debates return every review cycle. New merchants repeat old analysis. Supplier commitments are forgotten. Local exceptions become permanent without scrutiny.
By capturing the final rationale, the agent creates a growing body of Assortment Intelligence: why products entered the range · why they remained · why distribution changed · why an exception was approved · why they were eventually removed.
That history can improve future reviews, accelerate onboarding and provide governed context for more advanced assortment optimization later.
How EIG helps
EIG works alongside merchants, category managers, space planners and analysts to define the complete workflow: observe the current assortment review · map the real decision criteria · document the exceptions · identify trusted data and knowledge sources · design the agent instructions · establish human approval points · build the output template · define how final rationales are retained · test the agent against previous assortment decisions · measure its usefulness before expanding the scope.
We are vendor-agnostic and can design the workflow around ChatGPT or another approved enterprise AI environment. The objective is not to force a retailer into a predetermined platform. It is to build a controlled decision-support capability around the retailer's existing merchandising process and commercial judgment.
Ready to build a more consistent assortment-review process?
Start with one category and one upcoming range review. EIG can map the current process, capture the merchant decision framework and design a practical Assortment Decision Pack Agent around your existing information. Book a Retail Intelligence Discovery Workshop to turn your next assortment review into a controlled, measurable AI pilot.
→ enterpriseintelligencegroup.com/Contact (Or take the free AI Maturity Assessment → /AIMaturityAssessment)
EIG brings practical consulting expertise across retail, CPG and distribution, connecting people, process, data and technology.
The EIG Library · Insights · AI Maturity Assessment · Contact