The weekly review, written from your own records.

Kelynto does one job each week. It writes the review of sales against plan that your planners owe leadership: what your records explain, with the record behind each estimate, what is on file near your stores, what is still open, and what is coming against plan. Your planner reads it, decides and publishes.

Product tour.

Sixteen stops in the order a week runs. Every one is a real screen of the product on a synthetic retailer.

To see it working on your own questions, book a 20-minute conversation.

  1. Step 1 of 16

    The draft is waiting

    The week opens on a drafted review: the files it was produced from, actual, plan and variance, then the week in three parts on the shortfalls and lifts of each store and category: what your records show with an estimated impact, what is on file near your stores, and the amount not explained. The byline says who drafted it and that every figure in the commentary is a figure of this week's data.

    The planner reads it before changing anything.

    Review screen in the sample workspace (synthetic data): the files the review was produced from, actual, plan and variance for the week, and the week in three parts on the shortfalls: recorded with an estimated impact, on file and not credited, and not explained
  2. Step 2 of 16

    The review, written

    The byline says who drafted it: Kelynto with the AI writer, Claude (Anthropic). The model read the week's evidence ledger, asked for three calculations first, and wrote the review in sections: the summary, what is recorded with an estimated impact, what is on file and not credited, what is not explained, questions to resolve, proposed next steps and what is coming up. Every figure in it is printed by Kelynto from the entry the sentence cites; the model typed none of them.

    "How this draft was written" opens under the text: the calculations the model asked for and why, then each claim with the entries it rests on. Here 26 claims were checked and one was dropped. The review is shown as it was written.

    The planner reads the draft, checks a claim against its evidence, and asks for a new draft with a focus if the emphasis is wrong.

    Review commentary in the sample workspace (synthetic data), written by the AI writer, Claude (Anthropic): the summary of the week against plan, what is recorded this week with estimated impact, and what is on file and not credited, with the records on file for the coming weeks beside it
    The same review further down: a hypothesis labelled as one, the questions to resolve, the proposed next steps and what is coming up
    How this draft was written, opened under the review: 25 claims checked, 1 dropped and 3 calculations run, the calculations the model asked for and why, then each claim with the evidence entries it cites
  3. Step 3 of 16

    Check each cause against its evidence

    A recorded cause opens to the records it rests on: here the in-stock file, store by store, each store's rate that week against its usual rate. Its amount is labelled an estimate. A record on file opens the same way, as context with its place and dates and no amount. Movements inside the amount not explained are listed with buttons to name the cause.

    The planner confirms, rejects or corrects each recorded cause, and says whether a record on file mattered.

    Review screen in the sample workspace (synthetic data): a confirmed stockout opened to its evidence, the in-stock records of nine stores, each against its usual rate, with the estimated impact at each store and the file the records came from
  4. Step 4 of 16

    Follow a number down to the store

    Explore shows where the week moved by region, category and store. Opening a store shows its last thirteen weeks against plan and its own week in three parts.

    The planner answers "which stores?" without leaving the review.

    Explore screen in the sample workspace (synthetic data): stores listed by variance to plan, with one store opened to thirteen weeks of actual against plan and its own week in three parts, a weather record on file with no amount and all of its shortfall not explained
  5. Step 5 of 16

    Set two weeks side by side

    Compare puts any two weeks next to each other at any level, each as its own review states it, with the net gap and the shortfalls apart, and marks what differs. A second view lists the last weeks of one store, region or category.

    The planner sees what changed since last week before anyone asks.

    Compare screen in the sample workspace (synthetic data): two weeks side by side for the whole workspace, with plan, actual, variance, the net gap, the shortfalls, recorded causes, the amount not explained and its share of the shortfalls for each week, and the change between them
  6. Step 6 of 16

    Turn the review forward

    Outlook sets what is on file for the weeks ahead against the latest forecast, with the original plan beside it: the change on file with its range, the outlook it gives, and the items with no amount, counted and shown rather than guessed. The impact is listed by cause, region, store, category and week, and every item can be adjusted, marked as already in the plan or ignored. A table says what evidence reaches each week.

    A scenario changes named assumptions, here replenishing earlier everywhere in scope, and shows the outlook as on file beside the outlook with those assumptions and the difference between them. It is labelled a conditional what-if and can be issued with the outlook.

    The planner walks into Monday with the next four weeks argued, and issues the outlook with its decision, owner and due date.

    Outlook in the sample workspace (synthetic data): four weeks against the latest forecast, with the original plan beside it, the change on file with its range, the outlook, the items with no amount, the impact by cause, and the first items with their decision buttons
    Outlook scenario in the sample workspace (synthetic data): replenish earlier, every store and category, from the week of Oct 26, with the outlook as on file, with these assumptions and the difference, labelled a conditional what-if, and the table of what the evidence reaches week by week
  7. Step 7 of 16

    See what is on file near your stores

    Watch shows the records near your stores for the days ahead on a map and a list, filtered by kind, dates, region, store and distance. Each record names its place, its dates and its source. Your team can record an event or a competitor opening itself.

    The planner marks each item as new to them, already known or not relevant.

    Watch map in the sample workspace (synthetic data): filters by kind of record, dates and distance, a list of records near stores with place, dates and source, and the map beside it
  8. Step 8 of 16

    Look up the last times it happened here

    Impact Memory lists the earlier weeks a condition was on file at a store, each with actual and plan for the weeks around it and what else was on file.

    The planner reads the history as a record of what was on file and what sales did.

    Impact Memory in the sample workspace (synthetic data): one earlier occurrence of a heat advisory at one store, with actual and plan for the weeks around it and what else was on file, under a notice that it is a record and not a cause, an average or a forecast
  9. Step 9 of 16

    Ask the workspace a question

    The assistant answers from the workspace's own records with one short sentence first, and the evidence beneath it. Each sentence carries a numbered source that opens the record behind it, and a record on file is listed as context with no amount.

    The planner checks an answer against its record in one click.

    Assistant panel beside the review in the sample workspace (synthetic data): the question what explains the gap to plan, answered with the shortfalls that have no recorded cause, a numbered source and the evidence beneath it
  10. Step 10 of 16

    Publish and send it up

    Publishing lists the items above the workspace's threshold that have no decision yet, and readers are told how many were left. The review downloads as a Word document, a PDF or a PowerPoint deck, or copies with formatting for email and slides; from a published review the same menu gives the recap for leadership as a workbook, slides and email text. Send emails it to members and approved addresses, each getting the review they are entitled to, or drafts the message for you to send when no mail relay is set up.

    The planner sends leadership the review in the form they already use.

    Review screen in the sample workspace (synthetic data) with the export menu open: copy with formatting or as plain text, the review as a Word document, PDF or PowerPoint, and the recap for leadership as a workbook, slides and email text, offered once the review is published. Above it, the week after week line, saying the amount not explained has been above the threshold for the eighth week running
  11. Step 11 of 16

    Two questions once it is published

    The published review asks the planner two questions: how many minutes the review took, and whether they would use this every week, from 1 to 5.

    Both answers are kept with the week's review, and the second feeds Score and learn.

    Published review in the sample workspace (synthetic data): the end of the commentary, with the cause a planner added in their own words and estimate, and a short form asking for the minutes the review took and a one to five answer to would you use this every week
  12. Step 12 of 16

    Every decision is kept

    The decision history lists each label, each cause a planner added or changed and each edit to the commentary, with who, when, and what was on record before and after. A decision cannot be edited or removed from the list.

    Anyone who reads the whole workspace can see who decided what, and download it.

    Decision history of a review in the sample workspace (synthetic data): an edit to the commentary, a cause a planner added with their own estimate, and a confirmed cause, each with who and when, and a button to download the whole history
  13. Step 13 of 16

    Check whether it is earning its place

    Score and learn holds the three pilot measures against their targets, each counted from what your planners did. The sample workspace shows the measures and their targets; the counts come from your own planners.

    The planning leader reads how the reviews are serving the team, from the team's own decisions.

    Score and learn page in the sample workspace: three measures with their targets and no figure, under a notice that nothing is counted in a sample workspace
  14. Step 14 of 16

    When an export is wrong, it says exactly where

    Every file is checked and shown before anything is loaded. This file had unreadable dates and text where numbers belong, so it cannot be loaded as it is and the existing data stays unchanged. The preview shows how each column is read and what needs fixing.

    The analyst fixes the export and uploads it again.

    Upload preview in the sample workspace (synthetic data): a sales file that cannot be loaded because 16 of its 30 rows had errors, with how each column of the file is read
  15. Step 15 of 16

    A changed file waits for a decision

    Under a source profile, a correction to a published week is held and nothing of it loads. The delivery shows what it would change, row by row and store by store, and what it does to the published review's variance. Here it was sent by an API key, so the analyst decides.

    An admin or analyst who did not send it approves it with a reason, or rejects it.

    Data, Deliveries in the sample workspace (synthetic data): a corrected sales file sent by an API key under a source profile, held because it changes a published review, with each row it would change before and after
  16. Step 16 of 16

    Every change is on the record

    The audit log lists uploads, runs, edits, publishing, labels, settings and sign-ins in words: who, what and when. The admin can check from the screen that no event was changed after it was written, and download the log.

    The admin can show who changed what.

    Admin audit log in the sample workspace (synthetic data): a check that every event links to the one before it, filters, downloads as CSV or JSON Lines, and the newest events with who and what changed

Four parts, one weekly loop.

Explain, Watch, Act, Score. Each part is a screen your planners use.

01 Explain

What happened last week and why, with evidence

Actual against plan, with the stores and categories under plan and over plan added up separately, then the week in three parts. Recorded this week, with estimated impact: a stockout, a store closure or a promotion that ran, read from your own records, credited only where its records sit, each with the record behind it one click away. On file this week, not credited: weather, local events, road work and competitor openings near a store, listed with place and dates as context for the planner. Not explained: every shortfall no recorded cause is credited with, with the movements inside it named first.

The same three parts can be followed down by region, category and store, searched and filtered, and any two weeks can be set side by side. Each review names the files it was produced from.

Every level adds up to the one above, a strong store never hides a weak one, and each estimate is labelled as one.

Kelynto Explore screen in the sample workspace (synthetic data): stores listed by variance to plan, with one store opened to thirteen weeks of actual against plan and its own week in three parts, a weather record on file with no amount and all of its shortfall not explained

02 Watch

What is on file near your stores

Weather, local events, road work and competitor openings for the days and weeks ahead, on a map and a list that stay in step, each with its place, its dates, its source and when the source issued it. Your team records what only it knows, such as a competitor opening. Alert rules tell a person in the app, and by email where the deployment has a mail relay, when a record of the kind they chose comes on file near their stores.

Impact Memory keeps what was on file at each store beside what sales and plan were in those weeks, and lists the earlier times a condition was on file there.

Each record is context for the planner, with its place, its dates and its source.

Kelynto Watch map in the sample workspace (synthetic data): records near the stores for the days ahead, each with its kind, place, dates and source, listed beside the map

03 Act

The planner decides, and the decision is kept

Your planner confirms, rejects or corrects each recorded cause, says whether a record on file mattered, names the cause of a movement that is not explained, and edits the drafted commentary. A named cause is the planner's word, so its amount stays in what is not explained. A cause the review missed can be added in the planner's own words, with stores, dates and the planner's own estimate, before or after publishing; it is listed apart as recorded by a planner and never changes the review's amounts. Follow-up actions get an owner and a due date.

When corrected figures arrive after publishing, the published review stays as it was. The week is run again with every decision carried over, and the restated review says what it replaces, when and by how much, and is offered to the people who received the earlier one.

The published review goes out as a Word document, a PDF or a PowerPoint deck, by email to members and to addresses at domains your admin has approved, or as a read-only link when your admin has switched links on. It can also be copied into email or slides.

Every decision, edit and publish is kept with who made it and when, and can be downloaded.

Kelynto decision history of a review in the sample workspace (synthetic data): an edit to the commentary, a cause a planner added with their own estimate and a confirmed cause, each with who and when

04 Score

Whether it is earning its place

Score and learn holds three measures, taken from what your planners actually did: how much of the draft they changed, how often their labels agreed with what the review recorded, and whether they would use it every week. For each reviewed week it also shows what the review said, what planners decided, and the plan and actual sales of the weeks that followed, stated as facts.

Each figure is counted from your planners' own edits and decisions, and shows the number of reviews or labels it rests on.

Kelynto Score and learn page in the sample workspace: draft edit rate, planner agreement and would use every week, each with its target and no figure, under a notice that nothing is counted in a sample workspace

Screens show Harborview Markets, a demonstration retailer with synthetic data.

Look ahead against your plan

The Outlook.

The review, turned forward.

  • Beside your own plan. For a period of one to thirteen weeks the planner chooses, what is on file that will move sales, set beside your own plan or latest forecast as you loaded it, by cause, for every store and category. An item already in your plan is not added again.
  • By cause. Promotions on the calendar, sized from earlier promotions of the same category; store closures on file; holidays, from the same holiday a year earlier; and stock-out risk where a store and category's in-stock rate fell well below its usual one. Weather, local events, road work and competitor openings for the period are listed beside them with place and dates.
  • Scenarios as conditional what-ifs. Named assumptions set beside the outlook: replenish earlier, cancel or move a promotion, trading hours, or a planner's own.
  • Issued, then looked back on. The planner marks each item (adjust the forecast, already in the plan, ignore), and the outlook is issued as a snapshot with its decision, owner and due date. Once the weeks have sales on file, the look-back sets what happened beside the estimate.

Each estimate says what it rests on, and a planner's decisions can be exported as a file for your forecasting system.

Who writes the words

The engine computes every number.

The words are written by Kelynto's built-in writer, or by Claude when the AI writer is switched on for the workspace. Either way every figure is printed from the week's evidence.

  • From the week's evidence. The evidence holds every figure, estimate and record of the week, each with its subject, period, direction and source. Both writers work from it and from nothing else.
  • The AI writer cites, Kelynto prints. Claude, by Anthropic, writes typed claims: fact, estimate, hypothesis, context, question, proposed action and what is coming up. Each figure cites its evidence entry, and Kelynto prints the figure in its place.
  • Each claim is checked against the entries it cites. A claim that does not hold is dropped, and a text that does not hold as a whole gives way to the built-in draft. The review shows how the draft was written, claim by claim.
  • The planner reads and publishes. The planner edits the draft, can ask for a new one with a focus, and publishes it. Every version stays.
  • A choice, per workspace. The AI writer is switched on by the workspace's admin, or for a partner's clients under the partner's recorded agreement with them. An admin can choose the company's own model instead, with its key sealed, and sets the audience, the length and a house style.

From last week's workbook to what leadership reads

Around the weekly review.

  • Workbook import A planner gets from last week's workbook to a first review the same morning. The upload takes Excel and CSV, and reads a recap sheet laid out as a grid, with stores down the side and weeks across the top. The workbook's own totals are read as control totals.
  • The recap for leadership From a published review, Export gives back the recap the planner already sends: a workbook with the week in three parts per region, store and category, 16:9 slides, and an email-ready text. Every figure is a stored value of the week.
  • What this means for next week A panel states in words what carries into next week: each recorded cause and what its records say of it, what is still open as a question, the open follow-up actions, and the records on file for the week after.
  • Memory across weeks This week's statement is set beside the week before's, measure by measure, and the review says when a recorded cause or an open amount has been on record at the same place for several weeks running.
  • Notes from the stores A store or district manager leaves a short, dated note for one of their own stores, from a form that fits a phone. It is shown on the review as context, with who wrote it and when.
  • The Monday email A plain email per person with the published review of their own stores in brief: the three parts, what changed since the week before, what carries into next week, the open follow-ups and a link.

What leadership gets

  • A written review By region, category and store, in the planner's own final words, as a document, a deck or an email.
  • The record behind every recorded cause Store by store, one click from the sentence that states it.
  • The whole week accounted for What the records explain and what is still open add up to the week's variance to plan, every week.
  • Score and learn How the reviews are serving the team, counted on your own weeks.

Files in, nothing installed

Works with what you already run.

Kelynto works from weekly files in a documented contract, with your own column names. It reads the standard exports of these point-of-sale and back-office systems as they are, and maps every column on the first upload.

  • Lightspeed Retail (R-Series and X-Series)
  • Square for Retail
  • Shopify POS
  • NCR Counterpoint
  • Retail Pro
  • Microsoft Dynamics 365 Business Central
  • Clover and RICS, with a ready mapping
  • An open-to-buy plan by month and class
  • A weekly recap workbook

From a planning system. There is a file recipe for the weekly exports of Anaplan, Blue Yonder, Logility, o9, Oracle Retail, RELEX and SAP IBP: which export carries what, how that system defines the week, and how promotions and availability appear. Your planning system or warehouse sends the weekly file to the inbound API or leaves it in a folder Kelynto collects from, with its control totals beside it.

The names above are trademarks of their owners. Kelynto reads the files these systems export.

For your admin, your analyst and your IT team

Run the workspace yourselves.

Data in and out

  • Four ways in. Upload in the app, send files to the inbound API, let Kelynto collect them on a schedule from an SFTP folder or a storage bucket your admin names, or have the first load done with you. Every file goes through the same check before anything is loaded.
  • Ways out. A read-only API, signed webhooks when a review is published or data is late, and a reporting export as CSV or Parquet for a BI tool.
  • Data health. When each feed last delivered, what is late or missing, and what is held for a decision.

The files a pilot needs

Governed delivery

  • A source profile for each recurring export. A named, versioned contract made from a file that loaded: its columns, the field and type of each, and the workspace's calendar and currency.
  • Control totals, to the cent. The rows that would load must reconcile exactly to what the export says it holds, or the delivery is held with every difference.
  • A changed file waits for a person. A file whose shape changed, or that would change a published review, is held and shown row by row, before and after, for an admin or analyst to decide with a reason.
  • A delivery history. Every file from every channel, with its profile version, content hash, control totals, outcome, who sent it and who decided.

Where the checks sit in the week

Kelynto Data, Deliveries in the sample workspace (synthetic data): a corrected sales file sent by an API key under a source profile, held because it changes a published review, with each row it would change before and after and what it does to the review's variance
A corrected sales file, held because it would change a published review, with each row it would change before and after. Synthetic data.

Access by region, district or store

  • Set by an admin. A member can be limited to the stores of some regions, districts or stores of your stores file. API keys and webhooks can carry the same limit, and your identity provider's groups can set it.
  • Their stores, everywhere. A limited member reads their own stores on every page, document, email, share link, export, API answer and assistant answer.
  • Enforced three times. A restrictive policy in the PostgreSQL database on every table keyed by store, the same filter on every query, and a closed list of the routes a limited member can reach.

Administration

  • Audit log. Every change in words, filtered, checked against its hash chain from the screen, and downloaded as CSV or JSON Lines.
  • Retention and deletion. Retention per kind of data with a preview of what a change would remove, a full export of everything the workspace holds, and deletion that ends with a certificate listing what was removed.
  • Sign-in and people. Single sign-on set up by the admin with a test sign-in first, SCIM provisioning, a page that shows what each role can do, and a usage report.

Every control with its status and evidence

Ask the workspace

  • Plain questions, answered from your records. What explains a gap at a store, which stores missed most, what changed against another week, what is on file around a store, which actions are open, and what planners decided.
  • Every sentence cites its record. A numbered source opens the record it came from, and every amount, share, count and date in an answer must match that record before it is shown.
  • Fixed rules, the asker's own access. Answers are put together from the records by fixed rules, and the assistant reads only what the person asking is allowed to read.
Kelynto assistant panel beside the review in the sample workspace (synthetic data): the question what explains the gap to plan, answered with the shortfalls that have no recorded cause, a numbered source and the evidence beneath it

Beside your planning system.

Kelynto works from the plan and forecast you already have, and writes nothing back.

  • Your planning system keeps the plan. Forecasts, replenishment, allocation, pricing and orders stay where they are. Kelynto explains last week against the plan and sets what is coming beside it.
  • Files in. A store list and weekly sales against plan by store and category, as CSV or Excel with your own column names.
  • Or on a schedule. Your planning system or warehouse sends the weekly file to the inbound API, or leaves it in a folder Kelynto collects from. No agent is installed on your side.

A narrow job, done in full.

What Kelynto does, and what it leaves to the systems built for it.

  • The review, not the forecast. The Outlook sets what is on file for the weeks ahead beside the plan or forecast you load, and a planner's decisions on it can be exported for your forecasting system.
  • Written commentary, not dashboards. The review is a document a person signs. The labeled findings can be exported as CSV or Parquet for your BI tool.
  • Weekly totals by store and category. The grain a weekly business review is written at, with no transaction, shopper or item-level data to hand over.
  • Rules you can read. Each cause is stated by a rule with a stated threshold, so every recorded cause can be read and challenged.

Read how Kelynto weighs evidence, or see the measures a pilot is judged on.

Five ways to take the next step.

  1. Explore

    See the product

    The product tour shows real screens of the product on a synthetic retailer, in the order a week runs.

    Take the product tour

  2. Evaluate with your data

    Run it on your own weeks

    Eight sealed weeks of your own history, run once and scored against bars fixed before anyone sees a result, then about eight live weeks.

    See what a pilot involves

  3. Review the security

    Check every control

    Kelynto Managed on Microsoft Azure or in your own Azure account, what crosses the boundary, and every control with its status and evidence.

    Open the Trust center

  4. Offer it to your clients

    Build it into your service or product

    For planning services, retail software vendors and consultancies: the partner API, embedded views and your brand, starting with a sandbox of synthetic clients.

    See the partner platform

  5. Talk to us

    Twenty minutes with the founder

    Your process and your questions. Or write to marc@kelynto.com.

    Book a 20-minute conversation