Skip to main content
How Can We Help?Contact

Retail & Consumer

Earn the repeat visit instead of buying it, and keep the shelf full when they come.

We join what your point-of-sale, loyalty, online, inventory and pricing systems each know. Offers go to the customers they will move, at a depth that pays; promotions are judged by the margin they leave; forecasts reach the item and the store; and every AI tool that talks to shoppers or sets a price has an owner, a test and a record, inside the pricing and privacy rules of each state you sell in.

Talk through your loyalty and sales data See two loyalty results

Grocery and specialty retail · Restaurant brands · Consumer brands · Online and direct-to-consumer

A shopper’s phone beside a store front; the offer on the phone, the gap on the shelf, the price tag and the receipt at the checkout light in turn while a cart rolls along the pavement and fills1234
  1. 1LoyaltyLapsing member: a morning offer, at the depth that pays
  2. 2DemandSaturday’s gap seen on Thursday; the order is in
  3. 3PricingPromotion judged by the margin it left
  4. 4MeasurementThe visit counted against the offer that earned it
Illustrative: one offer, followed from the phone to the shelf to the receipt.

Where margin leaks between the offer and the shelf

Four labels we find on the shelf at almost every retailer and restaurant brand we meet. Each one reads the same way: what shoppers notice, what head office sees, what sits underneath, and what it costs.

  1. Offers go to customers who were coming anyway

    What shoppers notice
    The same offer as everyone else, often on the thing they already buy every week.
    What head office sees
    Redemptions look healthy, but most of the visits behind them would have happened without the offer.
    Underneath it
    Offers are chosen from segment averages and judged by redemptions, with no holdout group to show what changed.
    What it takes from margin
    Discount spent on visits you already had, while members who are drifting away get the same offer as the regulars.
    See Aisle 3: how an incremental visit is counted

  2. Promotions report lift, and the margin goes missing

    What shoppers notice
    A deep deal one week, then the item back at full price and short on the shelf the next.
    What head office sees
    The promotion report shows units up; the category’s margin report shows something else.
    Underneath it
    Lift is measured against last week instead of a baseline, and nothing subtracts what the deal took from similar items or from the weeks after.
    What it takes from margin
    Promotions renewed quarter after quarter that lose money once the full receipt is read.
    See Aisle 3: one promotion rung up five ways

  3. The shelf is empty while the system says it is full

    What shoppers notice
    The item they came for is missing, so they buy it elsewhere or leave without it.
    What head office sees
    On-hand counts show stock, so no order goes in; stores find the gap by walking the aisle.
    Underneath it
    Forecasts are made by category and week rather than by item and store, and inventory records drift away from what is on the shelf.
    What it takes from margin
    Lost sales nobody records, emergency orders, and markdowns on the stock that sat in the back room instead.
    See Aisle 4: the work in supply chain and stores

  4. AI reaches shoppers before anyone signs off

    What shoppers notice
    A chatbot answering for the brand, a price that seems to change by person, or a review summary nobody checked.
    What head office sees
    Teams switching on AI features inside the commerce, marketing and service tools they already pay for.
    Underneath it
    No list of what is in use, no owner for each tool, and this year’s state pricing and disclosure laws not yet mapped to any of it.
    What it takes from margin
    A disclosure missed in New York, a pricing practice a regulator asks about, or a complaint that becomes a headline.
    See Aisle 6: checking an AI tool before launch

Two loyalty programs, printed as receipts

Both are restaurant brands, and both set out to grow loyalty without deeper discounts. Each receipt lists who the client was, what was built and the totals it rang up. The figures are the client’s own; the clients are described, not named.

Receipt 1 of 2

A national quick-service restaurant chain

Marketing & loyalty: rewards optimization

The right offer, at the right depth and the right time

A large multi-market network with strong morning traffic, mobile order-ahead and drive-through. Its loyalty program already worked, but results varied widely by market, daypart and visit habit, so national offers did not always produce the best local outcome. Offers were broad-based rather than tuned to a member’s lifecycle or to how they actually visited: the commuter morning against the weekend afternoon. First-party transactions and loyalty data were combined with public, aggregated spend signals to find where a targeted incentive would convert best; eligibility had to work the same way in the app, in mobile order-ahead and in the store; and clean A/B tests with guardrails had to show promotional spend going down while loyalty growth and repeat visits went up, not the other way round.

Built

  • Lifecycle segments and propensitiesNew, routine and lapsing members, scored for conversion, morning-daypart frequency, basket lift and offer fatigue.
  • Offer selection with capsThe incentive (points multiplier, bundle or dollars off), its depth and its timing, chosen by lifecycle and behavior, with margin and frequency caps enforced.
  • Spend reallocationPromotional spend moved toward the cohorts and offers that performed best, to stop over-discounting.

Totals

  • Loyalty sign-ups, in the targeted dayparts and markets+14%
  • Repeat visits, among active members receiving tuned offers+11%
  • Discount dollars per incremental visit−8%

Reported by the client.

Case study: Loyalty rewards optimization for a QSR chain

Receipt 2 of 2

A regional fast-casual dessert brand

Marketing & loyalty: personalization across app, email and text

Offers tuned to the neighborhood, the time of day and the lifecycle

A neighborhood footprint, rotating seasonal flavors and a gamified loyalty program built to turn casual drop-ins into weekly visits. Loyalty performance was steady but under-used: offers were not tuned to neighborhood, daypart or lifecycle, which left acquisition and repeat frequency on the table. Transactions and loyalty data were combined with public, aggregated spend signals in a privacy-safe way and turned into propensities marketers and operators could use, with consistent eligibility across channels, quick A/B tests with clean readouts, and guardrails that protect margin and limit offer fatigue.

Built

  • Privacy-first harmonizationAutomated ingestion of transactions, loyalty events and public spend and rewards panels, with privacy controls designed in.
  • Orchestrated deliveryOffers sent through the app, email and text, with dashboards by role so marketers and operators could iterate every week.
  • Creative rotationFlavor call-outs against points multipliers, rotated to learn which nudge works by lifecycle and daypart, improving redemption without deeper discounts.

Totals

  • New-member sign-ups, in the targeted cohorts+23%
  • Repeat-visit frequency, among active customers+6%
  • Customers moved to the new app and its loyalty point shop, alongside the program120,000+

Reported by the client.

Case study: Loyalty personalization for a fast-casual brand

Read a promotion the way finance reads it

Start with one promotion, rung up five ways: every line is true, and only the last total says whether to run it again. Then pick a price tag to see how each number behind offers, sales and the shelf can be counted, and what to agree before the next report.

One promotion, rung up five ways

A 12-pack of sparkling water, 25% off for one week. Press each key in turn: the receipt adds a line and the total changes.

Ring it up

Illustrative figures, checked line by line. The same receipt can come out ahead with vendor funding or a shallower discount.

Promotion review

12-pack sparkling water, 25% off, one week. Amounts in dollars.

  • Normal week: 1,000 sold at 6.00, margin 2.40 each2,400.00
  • Promotion week: 3,000 sold at 4.50, margin 0.90 each2,700.00
  • 1 Units against a normal week+200%
  • 1 Sales against a normal week+7,500.00
  • 2 Item margin against a normal week+300.00
  • 3 Similar items: 700 fewer sold, at 2.00 margin−1,400.00
  • 4 Next two weeks: 500 fewer of the item, at 2.40−1,200.00
  • 5 Rest of the basket: margin on the extra trips+900.00

Lift, as the report reads it+200%

Run it again: units tripled.

Margin gained+300.00

Run it again: 300 dollars better than a normal week.

Margin gained−1,100.00

Think again: it took more from the brands beside it than it added.

Margin gained−2,300.00

Stop: shoppers stocked up and skipped the next two weeks.

What the promotion earned−1,400.00

The full receipt: 1,400 dollars of margin lost on a promotion reported as a 200% lift.

Seven numbers on the price tags

Discount per incremental visit

Rung up as

  1. Per redemption. The discount given, divided by the offers redeemed.
  2. Per incremental visit. The discount given, divided by the visits a holdout group shows the offer added.
  3. Which discount. The face value of the offer, or the margin given up after any vendor funding.

Who acts on it

Loyalty and CRM teams choosing the depth of an offer, and finance setting the program budget.

Pulled from

the loyalty platform, point-of-sale transactions, the offer and campaign log

The arithmetic

10,000 members get a 3-dollar offer and 1,000 similar members are held back. 2,000 redeem, so the program spends 6,000 dollars. 43% of the offer group visit that week against 38% of the holdout, so the offer added about 500 visits: 3 dollars per redemption, and 12 dollars per incremental visit.

Agree before the next report

Per redemption or per incremental visit, and face value or margin given up?

Repeat rate

Rung up as

  1. Which customers. Loyalty members only, or every card and device that can be matched.
  2. Which window. A second visit within 30 days, 90 days or a year.
  3. Which starting group. Everyone who visited once, or only customers new in the period.

Who acts on it

Marketing and the leadership team, as the read on whether customers come back.

Pulled from

the loyalty platform, point-of-sale and online orders, the customer match tables

The arithmetic

Of 20,000 people who joined in March, 7,000 visit again within 30 days and 11,000 within 90 days. The repeat rate is 35% or 55%, and lower again once guests who never joined are counted.

Agree before the next report

Members only or every matched customer, which window, and which starting group?

Comparable sales

Rung up as

  1. Which stores. Stores open a full year or 13 months, and whether remodeled or relocated stores stay in.
  2. Which sales. In-store only, or online and delivery orders credited to a store.
  3. Which calendar. The same weeks last year or the same dates, which moves holidays and the extra week in a 53-week year.

Who acts on it

Investors, lenders and the board, and often every store manager’s bonus.

Pulled from

point-of-sale, the online and delivery platforms, the store master list

The arithmetic

40 stores open a full year sold 52.0 million dollars against 50.0 million last year: up 4.0%. Credit the orders picked up in those stores to them, 5.2 million this year against 4.6 million, and it is 57.2 against 54.6: up 4.8%.

Agree before the next report

Which stores, which channels, and which calendar?

Average ticket

Rung up as

  1. What it includes. Before or after discounts, and with or without tax, tips and delivery fees.
  2. Per what. Per transaction, per order or per guest.
  3. Which orders. All channels together, or the counter, drive-through, app and delivery apart.

Who acts on it

Pricing, menu and upsell decisions, and how a new offer is judged.

Pulled from

point-of-sale, the app and the delivery platforms

The arithmetic

1,000 orders bring in 14,000 dollars before discounts, with 1,200 dollars of discounts and 900 dollars of delivery fees. The average ticket is 14.00 before discounts, 12.80 after them, and 13.70 with delivery fees counted.

Agree before the next report

Before or after discounts, which fees, and per order or per guest?

Forecast error

Rung up as

  1. Which error. The average percentage error by item, or the total error weighted by sales.
  2. Which level. Item by store by day, or category by week, where errors cancel out.
  3. Which forecast. The system’s forecast, or the one planners changed.

Who acts on it

Replenishment and planning, and whether stores trust the forecast enough to order from it.

Pulled from

the forecasting and replenishment systems, point-of-sale, inventory records

The arithmetic

Three items forecast at 90, 60 and 20 units sell 100, 50 and 10. The errors are 10%, 20% and 100% of what sold: an average of 43%. Weighted by sales, 30 units of error over 160 sold is 19%. Added up as one category, 170 forecast against 160 sold is 6% off.

Agree before the next report

Which error, at which level, and the system’s forecast or the planners’?

In-stock rate

Rung up as

  1. By the record. Items the inventory system shows on hand.
  2. By the shelf. Items actually found on the shelf in a scan or a walk.
  3. Weighted how. Every item counted the same, or each weighted by what it sells.

Who acts on it

Store operations, replenishment, and the conversation with suppliers.

Pulled from

inventory records, shelf scans and audits, point-of-sale

The arithmetic

A store carries 10,000 items. The system shows 9,700 in stock: 97%. A shelf walk finds 9,200 on the shelf: 92%. The 800 missing items are fast sellers worth 11% of sales, so weighted by sales the store is 89% in stock.

Agree before the next report

By the record or by the shelf, and weighted by sales or not?

Shrink

Rung up as

  1. At cost or at retail. The loss valued at what the stock cost, or at its selling price.
  2. Known or unknown. Recorded damage, spoilage and write-offs, or only the gap a stock count finds.
  3. Against what. A share of sales, or of inventory value.

Who acts on it

Loss prevention, store managers and finance, and where cameras and locked cases go.

Pulled from

stock counts, write-off and markdown records, point-of-sale

The arithmetic

A store sells 20 million dollars a year. Its count finds a gap of 300,000 dollars at retail, or 180,000 at cost, and recorded spoilage and damage add 100,000 at retail. Shrink is 1.5% at retail on the count alone, 2.0% with the recorded losses, and 0.9% at cost.

Agree before the next report

Cost or retail, recorded losses in or out, and against sales or inventory?

Which promotion gets renewed every quarter without anyone reading the full receipt? Bring it to a 30-minute call, and we will show you what it takes to read every event the way finance would, every week.

Read one promotion with us

The work, department by department

Four departments, the way a retailer or restaurant brand is run rather than the way its systems are built. Each item names the number it moves; where a published result exists, it is linked.

Dept. AMarketing and loyalty

  • Offers chosen by who they will move

    Lifecycle segments and propensity models choose the offer, its depth and its timing for each member, with margin and frequency caps, and a holdout group in every campaign so the visits it added can be counted.

    The number it moves: Discount per incremental visit, and repeat rate

    Case study: Loyalty rewards optimization for a QSR chain

  • Personalization across app, email and text

    Offers and creative rotated by lifecycle and time of day across your own channels, with consent and opt-outs honored in every one, and weekly readouts that marketers and operators act on.

    The number it moves: Sign-ups and repeat visits in the targeted cohorts

    Case study: Loyalty personalization for a fast-casual brand

Dept. BMerchandising and pricing

  • Promotions judged by the margin they leave

    A baseline for every promoted item, with what the deal took from similar items and from the weeks after, so next season’s calendar is built on what each event actually earned.

    The number it moves: Incremental margin per promotion

    First step: AI & Data Assessment Sprint

  • Markdowns timed by sell-through

    Clearance and seasonal markdowns recommended by item and store from how fast each is selling and the weeks left, with the merchant approving every price.

    The number it moves: Full-price sell-through, and markdown dollars

    First step: AI & Data Assessment Sprint

Dept. CSupply chain and stores

  • Forecasts by item, store and day

    Demand forecast at the level orders are placed, with promotions, weather, local events and holidays in the model, and planners’ changes tracked so the better forecast wins.

    The number it moves: Forecast error weighted by sales, and emergency orders

    First step: AI & Data Assessment Sprint

  • Shelf gaps found before shoppers find them

    Items that should be selling and are not, flagged by store each morning, so phantom inventory is corrected and the order goes in before the weekend.

    The number it moves: In-stock rate by the shelf, and lost sales

    First step: Data Platform Accelerator

  • Schedules built from the forecast

    Hours planned from forecast traffic and posted inside the notice periods the fair workweek laws set, with managers free to adjust.

    The number it moves: Labor hours against traffic, and pay for late schedule changes

    First step: AI & Data Assessment Sprint

Dept. DOnline and customer service

  • A service assistant that knows the order

    Answers about orders, returns and store hours from your own policies and order data; it says it is an assistant, and hands off to a person with the context when the question needs one.

    The number it moves: Contacts resolved without a hand-off, and repeat contacts

    First step: AI & Agent Pilot

  • Product content and review summaries that hold up

    Product descriptions drafted from approved attributes, and review summaries drawn only from real reviews, each checked by a person before it is published.

    The number it moves: Time to list a product, and content corrections

    First step: AI & Agent Pilot

In every department

  • One customer across channels, with consent attached

    Point-of-sale, loyalty, online and app records matched to one customer, with each person’s consents and opt-outs carried with the record, so every offer and audience respects them.

    First step: Data Platform Accelerator. The practice behind it: Data Engineering & Platforms.

  • An AI inventory and a pricing review

    Every AI tool and AI feature in use listed with an owner, and anything that sets or personalizes a price checked against the state rules before it goes live.

    First step: AI Governance Quickstart. The practice behind it: Strategy & Governance.

  • Teams who read the numbers the same way

    Training by role on your own data, so merchants, marketers and store managers read one promotion the same way and act on it.

    First step: Adoption Review. The practice behind it: AI & Data Literacy.

The margin leaks somewhere else in your business? Bring the number that worries you most to a 30-minute call.

Talk through the number that worries you

The rules on prices, offers, faces and schedules

Personalized pricing is where the rules moved fastest this year: three states passed limits in 2026, on top of New York’s disclosure law. The rest depend on where you sell and what you collect. Each rule links to its source.

On the price tag

  • New York’s algorithmic pricing disclosure law (General Business Law §349-a) (opens in a new tab)

    Reaches
    Any business that sets a price for a New York shopper with an algorithm that uses that shopper’s personal data.
    Where it stands, October 2026
    In force since Nov 10, 2025. A broader ban, the One Fair Price Act, passed the legislature on Jun 4, 2026 and was awaiting the governor when last checked.
    What your data has to show
    The disclosure sits next to the price, in the same place and at the same time. Prices set only from market conditions such as demand, supply or inventory fall outside it; penalties reach 1,000 dollars per violation once a notice to cure has passed.

    Wording the law requiresTHIS PRICE WAS SET BY AN ALGORITHM USING YOUR PERSONAL DATA.

  • The 2026 state laws limiting personalized prices (opens in a new tab)

    Reaches
    Maryland: food retailers with stores of 15,000 square feet or more, and grocery delivery services. Connecticut: retail sellers and delivery services. A third state: anyone pricing groceries and other foodstuffs, including household, health and beauty and pet supplies.
    Where it stands, October 2026
    Maryland in force since Oct 1, 2026. Connecticut’s takes effect in 2027. The third state’s grocery ban starts Aug 1, 2027, after a one-year pause on new electronic shelf labels in food stores that begins Feb 1, 2027.
    What your data has to show
    A shopper’s personal data cannot be used to raise their price. Loyalty and rewards discounts with published terms, posted promotions and cost-based differences are allowed, so the data behind each price change has to show which one it is. In Connecticut, others who raise an online price this way must say so.

    Wording the law requiresTHIS PRICE WAS INCREASED USING YOUR PERSONAL DATA.

  • California’s law on common pricing algorithms (AB 325) (opens in a new tab)

    Reaches
    Businesses that use or share a pricing tool drawing on competitors’ data.
    Where it stands, October 2026
    In force since Jan 1, 2026.
    What your data has to show
    Using or distributing a common pricing algorithm as part of an agreement that restrains trade is unlawful, as is using one to pressure others into a recommended price. Know what data your pricing tool learns from, and whose.

In the offer and the message

  • State consumer privacy laws

    Reaches
    Businesses large enough to be covered, in the 20 states with a comprehensive privacy law in force in 2026.
    Where it stands, October 2026
    Indiana, Kentucky and Rhode Island joined on Jan 1, 2026.
    What your data has to show
    Shoppers can opt out of targeted advertising and the sale of their data, and several states require honoring a browser’s opt-out signal. Sensitive data, such as precise location or health purchases, needs consent in most of these states.

  • Loyalty program notices in California and Colorado (opens in a new tab)

    Reaches
    Loyalty and rewards programs with members in California or Colorado.
    Where it stands, October 2026
    In force. Colorado’s loyalty program rules since Jul 1, 2023.
    What your data has to show
    California asks for a good-faith estimate of what a member’s data is worth to the program, and how it was worked out. Colorado asks for the categories of data sold or used for targeted ads, who receives them, and each partner’s benefit.

  • California’s regulations on risk assessments and automated decisions (opens in a new tab)

    Reaches
    Businesses covered by the California Consumer Privacy Act.
    Where it stands, October 2026
    In force since Jan 1, 2026; notices for automated decisions from Jan 1, 2027; risk assessment details for 2026 and 2027 due Apr 1, 2028. On Jan 28, 2026 the Attorney General began a review of how retailers and others use personal data to set prices.
    What your data has to show
    A written risk assessment before selling or sharing personal information, which covers most targeted advertising, and before processing sensitive data. Automated tools that make significant decisions, such as hiring, need a notice before use and, in most cases, a way to opt out or appeal.

  • The TCPA and the FCC’s consent rules, for text programs (opens in a new tab)

    Reaches
    Brands that send marketing texts or calls.
    Where it stands, October 2026
    Opt-outs by any reasonable means, honored within ten business days, since Apr 11, 2025. The rule that one opt-out stops every kind of message is delayed to Jan 31, 2027.
    What your data has to show
    Each member’s written consent kept with the record, and an opt-out made in any channel carried to every system that can text them.

In the store

  • Illinois BIPA, and local rules on face recognition in stores (opens in a new tab)

    Reaches
    Stores that collect face, voice or fingerprint data: across Illinois; with signs at the door in New York City; and banned in places open to the public in Portland, Oregon.
    Where it stands, October 2026
    In force. Illinois limited damages to one recovery per person in August 2024.
    What your data has to show
    Written notice and written consent before a face template or voiceprint is collected, a published retention schedule, and no sale of the data. Cameras that count people without identifying anyone are another matter, and worth designing that way.

  • Fair workweek and predictive scheduling laws (opens in a new tab)

    Reaches
    Large retail and food service employers in Oregon, New York City, Chicago, Seattle, Los Angeles, San Francisco and other cities.
    Where it stands, October 2026
    In force; Los Angeles County’s since Jul 1, 2025.
    What your data has to show
    Schedules posted 14 days ahead in most places and 72 hours ahead for retail in New York City, with premium pay for changes after that. A schedule built from a forecast has to be built inside those windows.

Online and at checkout

  • The FTC’s rule on consumer reviews and testimonials (opens in a new tab)

    Reaches
    Every business that shows, buys or writes reviews.
    Where it stands, October 2026
    In force since Oct 21, 2024. The FTC sent its first warning letters under it in December 2025.
    What your data has to show
    No fake or AI-written reviews from people who do not exist or never used the product, no undisclosed reviews from staff, and no hiding the bad ones. Summaries drawn only from real reviews.

  • Bot disclosure laws in California, Maine and Utah (opens in a new tab)

    Reaches
    Chat and voice assistants that talk to shoppers.
    Where it stands, October 2026
    California’s since Jul 1, 2019; Maine’s since Sep 23, 2025; Utah’s narrowed on May 7, 2025.
    What your data has to show
    California: a bot used to encourage a sale says clearly that it is a bot. Maine: say so wherever a shopper could think it is a person. Utah: answer truthfully when asked, and say so up front in high-risk exchanges.

  • PCI DSS version 4.0.1 (opens in a new tab)

    Reaches
    Every merchant that takes card payments.
    Where it stands, October 2026
    All requirements mandatory since Mar 31, 2025.
    What your data has to show
    Every script on a payment page authorized and checked, and changes to the page detected. Marketing and analytics tags on checkout pages count as scripts.

For orientation only, not legal advice: other federal, state and local rules may apply, and what applies depends on where you sell, what you collect and how you use it. Checked October 14, 2026.

Where the data sits, system by system

The systems we read from, set out like a planogram, each with what it carries and the rules that ride along with it.

  • Point of sale

    Carries Every transaction, basket, discount and tender, by store and hour.

    Rules that ride along PCI DSS for card data; state privacy laws once a sale is tied to a person.

  • Loyalty and CRM

    Carries Members, points, offers, redemptions, consents and opt-outs.

    Rules that ride along Loyalty notices in California and Colorado; text and email consent rules.

  • Online store, app and delivery

    Carries Orders, sessions, carts, prices shown and delivery fees.

    Rules that ride along Pricing disclosure and personalized-pricing limits; PCI DSS on checkout pages.

  • Pricing and promotions

    Carries Regular prices, promotions, markdowns, vendor funding and the calendar.

    Rules that ride along State pricing laws; price advertising rules.

  • Inventory and replenishment

    Carries On-hand counts, orders, receipts, transfers and forecasts.

    Rules that ride along Few personal-data rules; traceability records for certain foods.

  • Workforce management

    Carries Schedules, hours, time punches and labor forecasts.

    Rules that ride along Fair workweek laws; automated hiring rules where it screens applicants.

  • Cameras and store sensors

    Carries Footage, people counts, shelf images and alerts.

    Rules that ride along BIPA in Illinois, signs in New York City, Portland’s ban on face recognition.

  • Customer service

    Carries Contacts, chats, returns, refunds and complaints.

    Rules that ride along Bot disclosure laws; consent to record; state privacy laws.

Before an AI tool talks to a shopper or sets a price

Pick a use case to see the job it does, the questions to answer before launch, the rules worth reading, the conditions for going live, and who would usually sign off.

Loyalty

Personalized offers in the loyalty app

The job it does: Chooses which offer each member sees, its depth and its timing.

Before launch, answer

  • Uses a shopper’s personal dataYes
  • Sets or changes a priceOften
  • Talks to shoppers directlyNo
  • Identifies a person by face or voiceNo
  • Decides about a worker or applicantNo

Rules worth reading

  • State privacy laws: targeted-advertising opt-outs honored, and sensitive data such as precise location used only with consent.
  • California’s notice of financial incentive, and Colorado’s loyalty program disclosures.
  • If the offer sets the member’s price: New York’s disclosure law, and the loyalty-discount terms in the 2026 state pricing laws.

Conditions for going live

  • A holdout group in every campaign, so the visits it added are counted.
  • Margin and frequency caps on every offer.
  • Discount terms published, and the same for every member who qualifies.

Who signs off (suggested)

Department head

First step: AI & Data Assessment Sprint. Case study: Loyalty rewards optimization for a QSR chain

Pricing

Prices that change by shopper online

The job it does: Sets a different price for a shopper based on their data.

Before launch, answer

  • Uses a shopper’s personal dataYes
  • Sets or changes a priceYes
  • Talks to shoppers directlyOften
  • Identifies a person by face or voiceNo
  • Decides about a worker or applicantNo

Rules worth reading

  • New York: the required disclosure next to every price set this way.
  • Maryland’s ban for large food retailers, in force now; Connecticut’s and a third state’s grocery ban in 2027.
  • California: the Attorney General is reviewing the practice under the state privacy law.

Conditions for going live

  • A written decision, state by state, on where it may run, before launch.
  • Every price logged with the inputs that set it.
  • Prices compared across groups, so no protected group pays more.

Who signs off (suggested)

Leadership and counsel

First step: AI Governance Quickstart

Pricing

Markdown and clearance recommendations

The job it does: Recommends when and how far to mark down seasonal and slow stock, by item and store.

Before launch, answer

  • Uses a shopper’s personal dataNo
  • Sets or changes a priceYes
  • Talks to shoppers directlyNo
  • Identifies a person by face or voiceNo
  • Decides about a worker or applicantNo

Rules worth reading

  • Price advertising rules: a “was” price has to be a real, recent price.
  • Little else, provided no personal data goes into the price.

Conditions for going live

  • The merchant approves every price.
  • Each recommendation shows the sell-through and weeks left behind it.
  • Results compared with the merchants’ own markdowns for a season.

Who signs off (suggested)

Tool owner

First step: AI & Data Assessment Sprint

Supply chain

Demand forecasting and replenishment

The job it does: Forecasts sales by item, store and day, and proposes the order.

Before launch, answer

  • Uses a shopper’s personal dataNo
  • Sets or changes a priceNo
  • Talks to shoppers directlyNo
  • Identifies a person by face or voiceNo
  • Decides about a worker or applicantNo

Rules worth reading

  • Few rules reach a forecast that uses no personal data.
  • For grocers, the traceability records required for certain foods from Jul 20, 2028.

Conditions for going live

  • Accuracy measured on your own history, weighted by sales, before go-live.
  • Planners can override, and every override is tracked.
  • A fallback for the days the forecast fails.

Who signs off (suggested)

Tool owner

First step: AI & Data Assessment Sprint

Online

A shopping and service assistant on the website and app

The job it does: Answers questions about products, orders and returns, and suggests items.

Before launch, answer

  • Uses a shopper’s personal dataOften
  • Sets or changes a priceNo
  • Talks to shoppers directlyYes
  • Identifies a person by face or voiceNo
  • Decides about a worker or applicantNo

Rules worth reading

  • California, Maine and Utah: shoppers told they are dealing with a bot.
  • Advertising law: what the assistant says about a product has to be true and supported.
  • State privacy and recording laws for any chat transcripts kept.

Conditions for going live

  • Answers only from approved product data and policies.
  • Says it is an assistant, and offers a person at any point.
  • Refunds and exceptions go to staff.

Who signs off (suggested)

Department head

First step: AI & Agent Pilot

Online

AI-written product descriptions and review summaries

The job it does: Drafts product pages and summarizes customer reviews.

Before launch, answer

  • Uses a shopper’s personal dataNo
  • Sets or changes a priceNo
  • Talks to shoppers directlyYes
  • Identifies a person by face or voiceNo
  • Decides about a worker or applicantNo

Rules worth reading

  • The FTC rule on consumer reviews: no fake reviews, and summaries that do not misstate what reviewers said.
  • Advertising law: every claim in a description needs support.

Conditions for going live

  • Descriptions drafted only from approved product attributes.
  • Summaries drawn from all reviews, not only the good ones.
  • A person approves before anything is published.

Who signs off (suggested)

Tool owner

First step: AI & Agent Pilot

Stores

Cameras that flag theft or match faces

The job it does: Flags suspicious activity on camera, or matches faces against a watch list.

Before launch, answer

  • Uses a shopper’s personal dataYes
  • Sets or changes a priceNo
  • Talks to shoppers directlyNo
  • Identifies a person by face or voiceOften
  • Decides about a worker or applicantNo

Rules worth reading

  • Illinois BIPA: written consent before a face template is collected.
  • New York City: signs at every entrance. Portland, Oregon: face recognition banned in places open to the public.
  • State privacy laws treat biometric data as sensitive.

Conditions for going live

  • Activity detection without face matching wherever that does the job.
  • A person reviews every alert before anyone is approached.
  • Error rates checked across groups before launch, and footage kept for a short, written period.

Who signs off (suggested)

Leadership and counsel

First step: AI & Data Assessment Sprint

Stores

Staff schedules generated from the forecast

The job it does: Builds shift schedules from forecast traffic and sales.

Before launch, answer

  • Uses a shopper’s personal dataNo
  • Sets or changes a priceNo
  • Talks to shoppers directlyNo
  • Identifies a person by face or voiceNo
  • Decides about a worker or applicantYes

Rules worth reading

  • Fair workweek laws: schedules posted 14 days ahead in most covered places, with pay for late changes.
  • Rest periods between shifts where the law sets them.

Conditions for going live

  • Schedules locked inside the notice period, with every change logged.
  • Premium pay worked out automatically for any change.
  • Managers can adjust, and see why the schedule looks the way it does.

Who signs off (suggested)

Department head

First step: AI & Data Assessment Sprint

People

Screening applicants for seasonal hiring

The job it does: Ranks or screens applicants for store and distribution roles.

Before launch, answer

  • Uses a shopper’s personal dataYes
  • Sets or changes a priceNo
  • Talks to shoppers directlyOften
  • Identifies a person by face or voiceNo
  • Decides about a worker or applicantYes

Rules worth reading

  • New York City Local Law 144: a bias audit and notice before an automated hiring tool is used.
  • Illinois: notice when AI is used in hiring decisions, since Jan 1, 2026.
  • California: a notice and, in most cases, a way to opt out or appeal for automated hiring decisions from Jan 1, 2027.

Conditions for going live

  • A recruiter decides; the tool only ranks.
  • Selection rates checked by group before each season.
  • Applicants told, and given a way to ask for a person.

Who signs off (suggested)

Leadership and counsel

First step: AI Governance Quickstart

Stores

An operations assistant for store staff

The job it does: Answers store staff questions from policies, planograms and procedures.

Before launch, answer

  • Uses a shopper’s personal dataNo
  • Sets or changes a priceNo
  • Talks to shoppers directlyNo
  • Identifies a person by face or voiceNo
  • Decides about a worker or applicantNo

Rules worth reading

  • Few rules apply while it answers from your own procedures and holds no customer data.

Conditions for going live

  • Only current, approved procedures, each with an owner.
  • Every answer cites the procedure and its date.
  • Questions about a customer or a refund go to a manager.

Who signs off (suggested)

Tool owner

First step: AI & Agent Pilot

A starting point for the conversation with counsel. What applies depends on where you sell, what you collect and how the tool is used; your counsel and privacy lead decide. The rules in full: Aisle 5.

Suggested sign-off. More than one may be needed: leadership approves anything that personalizes prices, identifies people or screens applicants, the department that runs the tool approves how it is used, and counsel decides what has to be disclosed.

  • Leadership and counsel: a senior executive, counsel and the privacy lead.
  • Department head: the head of marketing, merchandising, stores or service, with security and privacy.
  • Tool owner: a named owner, access controls and a check every quarter.

A use case not on the list, or one already live without these conditions? Thirty minutes is usually enough to see what review it needs.

Run one AI idea past the rules

Pick a lane: three ways to start

Three first steps for three different buyers. The first two are our assessment and executive offers, sized to one program or one leadership team, and neither commits you to the next.

  1. Offer and promotion economics

    For a CMO, loyalty lead or head of pricing with one program to fix first

    Two weeks, one offer program · Runs as the AI & Data Assessment Sprint

    One loyalty program or promotion calendar traced from the offer to the visit and the margin; the leaks on one list with what each costs; and a plan in order, with the holdout tests that will prove it.

    Start with the offer sprint

  2. An AI shortlist for the leadership team

    For a CEO, COO or chief digital officer with more AI ideas than budget

    Several hours, one leadership team · Runs as the Executive AI Workshop

    Your leadership team works through where AI would pay first, from pricing and promotions to the shelf and the service desk, and leaves with a ranked shortlist and one first move.

    Plan a workshop for your team

  3. Data Maturity Assessment

    For an IT or data lead who wants a quick read first

    Free; done in about six minutes

    Fifteen questions on lineage, quality, ownership, access and platform. The result is instant and needs no email.

    Check your data foundation

After either of the first two: a first build on one offer, one category or one group of stores, measured against a number you already report, such as repeat rate, promotion margin or in-stock rate, then a decision to scale, change or stop.

AI & Agent Pilot · Data Platform Accelerator

Questions from marketing, merchandising and IT

Do we need a customer data platform before we start?

No. We start from point-of-sale, loyalty and online order data as they are, match customers well enough to run a clean test, and build the shared customer record as part of the work where it pays for itself.

How do you prove an offer earned the visit?

With a holdout: a random group of similar customers who do not get the offer. The difference in visits between the two groups is what the offer added. Every campaign we design has one, sized so the result can be read in the weeks the offer runs.

Are member-only offers caught by the new pricing laws?

Usually not, when the terms are published, open to every member who qualifies and the same for each of them: the three state pricing laws passed in 2026 all allow loyalty discounts on those terms. An offer whose amount is set for one member from their data is another matter, and in New York it may need the disclosure. We map each offer type with your counsel before launch.

We are a franchise system. Whose data is it?

The franchise agreement decides that. We work from what it says about point-of-sale and loyalty data, build reporting each franchisee sees for their own restaurants, and keep anything shared across the system aggregated where the agreement asks for it.

Will our customer data train someone else’s model?

No. We build inside your cloud accounts and under your security controls, nothing learns from customer data unless you decide it should, and the AI services are set up so that providers neither keep nor train on what you send them.

Do chat transcripts and session recordings carry privacy risk?

They can. Several states treat recording a conversation without consent as wiretapping, and lawsuits over website chat and session replay are common. We set consent, disclosure and retention for transcripts before an assistant goes live, and keep only what you need.

Do you work with restaurant brands as well as retailers?

Yes. Both published results on this page are restaurant brands’ loyalty programs: a national quick-service chain and a regional fast-casual brand. The same work applies to grocery, specialty and online retail, and to consumer brands selling direct.

How soon would we see a result?

An offer test is read in the weeks the offer runs, so the first answer usually comes inside the first cycle. A forecast or promotion model is judged against your own history before it changes an order or a price.

Bring the offer you are least sure pays for itself.

The weekly offer that always redeems, the promotion that keeps being renewed, the item that is never on the shelf on Saturday, or the AI pricing idea nobody has signed off. Thirty minutes with the lead who would run the work, and one change to make first.

Book 30 minutes for your brand

Discounts up, visits flat? Thirty minutes on your offers, your promotions or your pricing rules.

Book a call for your brand