esedark
Menu
Back to notes

pricing / rules / approval

Price optimization software: rules, limits and implementation costs

Before publishing a price, check its inputs, limits and approval. Define what to buy and how to accept delivery without assuming a margin improvement.

Price optimization software should help your team decide which price to propose, which evidence supports it and when it may be published. Before buying a licence or commissioning an integration, I would trace one proposed change from cost data through rules and approval to the price confirmed in the destination. A recommendation that nobody can explain, or that arrives too late, does not solve that workflow.

I would start with a bounded catalogue and channel, in recommendation mode. The first delivery should demonstrate control and usefulness to the team. Passing operational checks does not establish a margin improvement. That requires a separate commercial evaluation with its own evidence and conditions.

Choose configuration, integration or custom development

If your current system can maintain price lists, schedule promotions and approve changes under your required rules, ask for a demonstration using your exceptions before replacing it. Configuration and data cleanup may be the whole engagement. A specialist product is worth assessing when those capabilities fall short, but its recommendations should face the same checks as custom software.

An integration makes sense when the calculation already works and the missing piece is receiving costs, securing approval or confirming the published price. Custom development needs a specific uncovered requirement, such as different approval rights by category or rules across channels. Ask the proposal to identify that gap before funding an entire pricing engine.

I would not assume AI is necessary. Floors, rounding and effective dates may need explicit rules only. The agents versus workflows guide separates variable decisions from known steps. Without dependable costs, a pricing owner or capacity to review suggestions, I would scope that groundwork first and defer the recommendation engine.

Put data requirements and pricing rules in the brief

Define the decision unit: product or variant, channel, currency, price list and validity period. State which cost basis applies and what it includes. Do not confuse margin on sales with markup on cost. The supplier should document the agreed calculation, its inputs and rounding. An AI-generated explanation cannot replace that definition.

Require an identifier, update time and owner for each source. Agree when data becomes stale and what happens next. If external reference prices are used, define product equivalence, size, condition and included charges. A similar item is not automatically a comparable one. Collection and upkeep need separate budget items when they are outside the engagement.

Specify floors and ceilings, permitted changes, update frequency, active promotions and who may approve exceptions. Rules need precedence: if a promotion conflicts with a revised cost, the proposal should enter review under the agreed policy. Persuasive generated text must not choose which constraint to ignore. The AI in marketplaces guide covers the broader catalogue and drafting workflow; this engagement focuses on a traceable pricing decision.

An acceptance matrix for comparing proposals

Give each supplier the same cases. Add an owner, evidence, included/optional/excluded status and any unresolved cost to each row. This is a proposed evaluation framework, not a ranking of tested products.

Proposed checks for a price recommendation system
CaseRequired behaviourAcceptance evidence
Missing or stale costDo not produce a publishable proposal from an unknown cost basis.Pending record, reason and owner; no price change sent to the channel.
Price outside a limitReject it or route it to the specified approval without bypassing the rule.Proposed price, calculation inputs and the version of the blocking limit.
Conflicting promotionApply agreed precedence and expose the conflict.Dates, channel and affected rules; a decision to retain, cancel or review.
Inputs change after approvalInvalidate approval when relevant cost, rules or destination price changes.A new revision linked to its inputs; old approval cannot publish the revised proposal.
One channel succeeds and another failsShow each destination separately and check before retrying.Price read from each channel, identified attempt and an assigned discrepancy.
Pause or withdraw a changeStop pending writes and validate a recovery price before applying it.Paused queue, visible completed actions and an authorised restoration under current rules.

Make recommendation, approval and publication separate visible states. Accepting a suggestion is different from confirming the price a buyer sees. Completion should include a destination check and a visible incident when the result differs. Approval belongs to one specific revision, not an indefinite permission to change that product.

Hypothetical example: a promotion meets an updated cost

Imagine a retailer selling one variant through its own online store and another channel. The team approves a reduction. Before publication, an updated cost puts the proposal below the agreed floor. The system should retain the approval as history, block that write and present a new decision. This is a design example, not a client case or a measured financial improvement.

If the first destination has already confirmed the price, labelling the whole batch as failed or blindly resending it would be inadequate. I would ask to see each destination's result, pause pending changes and decide the correction using current inputs. Restoring the previous price also requires a check: it may no longer satisfy the rules.

The initial trial can run without writing prices, comparing suggestions with staff review. Measure usable proposals, rejection reasons, review time and discrepancies. Agreed criteria let you assess delivery; a historical simulation is not proof of future profitability.

Scope, deliverables and ownership

Bound the catalogue, channels, sources, frequency, users and permitted actions. An initial engagement might include a data map, versioned rules, sample import, review interface, a test destination connection, completed acceptance matrix and pause procedure. Production writes need their own acceptance decision. The guide to integrating AI into an existing SaaS covers how to separate this capability from the rest of the product.

Explicitly exclude anything not commissioned: rebuilding the store, cleaning the entire historical dataset, collecting prices from arbitrary websites, connecting every channel, unattended operation or a guaranteed margin increase. Custom predictive models, commercial campaigns and continuous support should not be assumed to be included either.

The business confirms costs, rules and approvers. The implementer supplies tests, limitations and change records. Operations receives alerts and resolves discrepancies. Agree which code, configuration, tests and exports remain in your accounts, the rights you receive and how another person takes over. When migrating rules, reconcile a sample against the old system before enabling writes.

Implementation, maintenance and exit costs

Ask for separate assessment, data preparation, licence, integration, migration, testing, training and launch items. Effort depends on variants and channels, cost quality, review frequency, promotions and permissions. A proposed model needs its own budget and evidence compared with a rules-based alternative. The AI consulting budget guide helps commission that assessment through defined deliverables.

For operations, add licences, infrastructure, usage, external reference data, human review, alerts and maintenance over the same period. Request ordinary and peak activity scenarios with explicit assumptions. I would not quote a fee without understanding scope or assign savings to an untested proposal.

Specify who adapts connectors, updates rules and reruns checks after model or catalogue changes. Separate failures against accepted requirements from new functionality. Budget for exit too: export rules, proposals, decisions and confirmed prices; revoke access; stop scheduled jobs and hand over instructions. A list of current prices without that context is insufficient.

What to send me to scope a first delivery

Send your catalogue and sales tools, channels, approximate variant count, cost source and an anonymised example of a price change requiring review today. Include limits, promotions, the approver and what happens when a destination fails. Credentials and a complete confidential catalogue are unnecessary for that initial discussion.

I can review the scope of an initial AI and automation assessment and implementation with you. You can share the workflow context or use WhatsApp. The next step is to define the recommendation the team needs and the checks it must pass before publishing a price.

Review my pricing workflow