Build
Your first process mapping: pick the right table, answer three questions
For anyone adding a database or file source for the first time: how to tell whether a table holds a process, which table to pick in a database, the three questions Mapping asks, and how to read the sample under the fields to know the mapping is right.
This page covers the sources where you pick the table yourself: CSV and XLSX files, and databases such as MySQL and PostgreSQL, and online spreadsheets. SAP, Kingdee, Yonyou, Xero, HubSpot, Salesforce and the like have a fixed mapping: there is nothing to set.
First, check that the table holds a process
Process data records how something moves forward step by step: a column that tells each thing apart (an order number, say) and a time for each step. Tables come in three shapes:
- One row per case, a time column per step. An orders table with order_no, ordered_at, paid_at and shipped_at has one row per order and a time each for ordering, payment and shipping. Purchase orders, expense claims and sample registers often look like this.
- One row per status change. An order status history with order_id, status and changed_at has several rows per order, one for each change. Approval records, ticket transitions and activity logs look like this too.
- Not process data: one row per thing, with one time. A customers table with id, name, industry, region_id and created_at has one row per customer and only records when it was created. Product, employee and region tables are the same. A table with only a created time and an updated time is not one either: the updated time says the row was last changed, not that a step happened.
When you pick the third kind, Mapping opens with "This table reads as a list: one row per thing, with one time. It does not record steps." and an empty form, instead of a guess. When the database has tables that look like process records, they are listed under "These tables look like process records:"; Read this table returns to Connect with that table picked.
Find the right table in a database
- The table Connect picks for you already prefers tables with several time columns or a status column, usually the main document table. Start with it.
- Tables named like order, po, invoice, ticket or application are usually documents: one row per case.
- Tables named like history, log, status, flow or record are usually change records: one row per status change.
- Tables named like customer, product, user, region or dept are usually lists. Leave them out.
- When a process is spread over several tables, read the main document first, such as orders. Add a lower level such as order lines once the main one reads well, under Object levels (optional) in Advanced.
Mapping: three questions
Before Mapping opens, Duaer has the AI read the first rows and fill in a suggestion. This takes about half a minute, and "The AI is reading the data. This takes about half a minute." shows under the buttons. A list is not sent to the AI and opens within seconds. Once the suggestion is in, check three things:
- Which column tells each thing apart? This is the Case id column: different for every order, application or ticket, such as SO-1001. Do not pick a region, status, type or department, which many rows share. In a table with one row per status change, do not pick a row id that differs on every row either.
- What are the steps, and when did each happen? Under Rows of the table, choose the table's shape. With One row per case, a time column per step, select Add step and name each time column as a step, such as Ordered for ordered_at. With One row per status change, choose the Status column and the Changed-at column.
- What kind of process is it? Process kind is usually recognized already, such as purchase to pay or order to cash. If unsure, choose None; you can change it later.
Everything else is under Advanced, such as object levels, linked objects, the value map, who did a step and the owner email. You can leave it alone the first time. Advanced opens by itself when the AI's suggestion uses any of it. A choice the suggestion is unsure about is marked "Unsure, check this".
Read the sample to know the mapping is right
As you change a field, the first 200 rows are worked out again under the fields, such as "First 200 rows: 200 cases, 2.3 steps each on average", with two cases' steps, such as "SO-1001: Ordered → Paid → Shipped".
- A sensible number of cases, two or more steps each and steps in an order that makes sense: the mapping is usually right. In a table with one row per case, the cases should be close to the rows.
- "200 rows give only 3 cases: the case id column is likely wrong.": you probably picked a column many rows share, such as a region or a status. Pick one that differs for every case, such as the order number.
- "Most cases have one step: the step or time columns are likely wrong.": you probably added one step only, or read a status history as one row per case. Add more time columns as steps, or choose One row per status change.
With either warning, Next stays on Mapping. Fix the fields, then go on.
When the AI's suggestion was not used
Mapping then says why at the top, and the form was guessed from the column names and values: check each field. Retry AI suggestion asks again, and the new suggestion replaces the form.
- "No AI model is enabled": the organization has no text model it can use now.
- "The AI did not answer (it timed out, ran out of quota or its model source failed)": select Retry AI suggestion a little later.
- "The AI's answer could not be used": the answer could not be read, or named columns the table does not have.
- "The AI read …, but …": the AI's reading of the table does not fit the data, for example one row per case with one step where the rows look like a status history. Check the three questions above yourself.
Ids, people, emails and amounts are never shown to the AI.
Next
Once the preview has no issue marked as blocking, set the schedule and who can see the cases, and confirm. Every step in full is in Add a process source.
Questions
Our system only has list tables. What now?
A list records what things there are, not how something moves forward. Look in the same system for a document table (orders, applications, tickets) or a status or activity log. If there is none, export a spreadsheet with a time for each step and upload it as a file source.
My table has several time columns and a status column. Which shape do I choose?
If every step has its own time column, choose One row per case, a time column per step. If there is one status column and one changed-at column with a row per change, choose One row per status change.
Can I change the mapping later?
Yes. Before you confirm, select Mapping in the step bar to go back. After, choose Edit in the source's more actions to open the wizard again.
In this section
Add a process source in Duaer
A process source tells Duaer where to read the steps of your cases: a CSV or XLSX file, files dropped in an S3 or SharePoint folder, PostgreSQL, MySQL, Microsoft SQL, Oracle, TiDB, OceanBase, KingbaseES, SAP HANA, IBM Db2, Dameng DM8, Teradata, Vertica, Snowflake, BigQuery, Databricks, Microsoft Fabric, Amazon Redshift, Azure Synapse Analytics, ClickHouse, Trino, a Feishu Base, or Feishu, DingTalk or WeCom approvals. The wizard takes you through columns, mapping, a preview of issues and a schedule; after you confirm, Duaer reads new rows on that schedule.
ProcessSee the process in your data and generate workflows in Duaer
Duaer reads the steps every case went through from the tables, databases or Feishu apps you already have: where cases wait too long, where they branch and who did each step. From that it generates three kinds of workflow, Watch, Coordinate and Execute, and replays each over past days before you publish it.
ProcessProcess findings and SOP conformance in Duaer
Process findings list the steps where cases wait too long or are done again, and which kind of case they happen to most. Propose any of them as a card rule and compare the 30 days before and after it was adopted. SOP conformance compares the SOP in your knowledge base with the steps cases really took.