INSIGHTS
Adding sales AI to your existing CRM and SFA
For manufacturers and wholesalers, the data often already exists in CRM, sales management systems and Excel. The opportunity is to connect that information and prepare useful questions and proposals without replacing the systems your team relies on.
1. Define the roles of your systems and AI
You can plan an AI implementation around keeping your current CRM and SFA. First, check whether the required data can be exported, whether customers and products can be matched across systems, and whether each user’s permissions can be preserved. Feasibility depends on contracts, configuration and data quality, not simply the product name.
Manufacturers selling through dealers and wholesalers serving trade accounts often need purchase history alongside meeting records. CRM manages customer relationships and interactions; SFA supports sales activities and opportunities. Products may cover both functions, while transactions and product details remain in other systems.
Decide where official records live, what AI can read, and what it should return to the user.
| System | Information and role | How AI uses it |
|---|---|---|
| CRM / SFA | Account details, meeting history and next actions | Review previous discussions and unresolved issues |
| Sales management | Orders and sales transactions | Identify changes in volume and product mix |
| Product master | Codes, specifications and commercial terms | Prepare product comparisons and suggestions |
| Sales AI | Organize information across systems by account | Present questions, suggestions and source references |
For example, a salesperson reviews and corrects an AI-drafted meeting report before it is saved to the CRM. An AI output and an approved company record are different stages of the workflow.
2. Use it to prepare a customer visit
Consider a manufacturer or wholesaler of packaging materials. Before a visit, a salesperson checks purchase history in the sales system, reviews the previous meeting in CRM, and searches the product list for suitable suggestions. AI can bring those sources together for the account.
Preparing a visit to Customer A
This is a fictional workflow example, not a customer implementation result.
| Source | Information found |
|---|---|
| Sales/order system or an Excel transaction ledger | Orders for packaging material A have declined over the last three months |
| CRM/SFA or an Excel sales report | The customer said it was adjusting inventory at the last meeting |
| Product system or an Excel product list | Specifications for products that could be compared with the current item are available |
Excel data can be included alongside CRM and SFA records. File formats and contents need to be checked so that the required fields can be extracted.
AI can suggest questions such as “How far has the inventory adjustment progressed?” and “What order volume do you expect next month?”, and prepare alternative product comparisons where relevant. The salesperson checks the sources and decides what to ask and propose.
Declining orders may reflect seasonality, returns or changes in product mix. They do not, on their own, prove that a customer has switched suppliers. AI should surface evidence for a conversation, rather than assert an explanation. Stock, delivery dates and commercial terms also need to be checked; information the AI could not verify should be marked as unconfirmed.
3. Start with the data the workflow needs
You do not need to collect all company data at once. Choose one workflow and identify the systems that contain the information it needs.
Identify the account
Use customer codes, locations and account assignments. If CRM and sales management use different codes for the same customer, create a mapping so the records can be joined correctly.
Understand changes in transactions
Use transaction dates, product codes, quantities and amounts. Account for returns and cancellations, and align units before comparing volumes: boxes and individual pieces cannot be compared directly.
Review previous conversations
Use visit dates, meeting notes and next actions to understand what was discussed and what remains unresolved. Make outdated or missing information visible to the user.
Compare products
Use product codes, specifications and commercial terms. Similar names can refer to different specifications or sales units. Match by code and check whether the product is still available or has been discontinued.
Show the source system and update date. If inventory is from the previous day, label it accordingly so it is not mistaken for current stock.
4. Check API and CSV integration requirements
APIs exchange information between systems; CSV files transfer tabular data. The right choice depends on the system and contract, how often data needs to change, and whether AI-generated drafts will be written back.
API integration
Check that your plan permits API access and that the required fields are available. Also review permissions and rate limits. Define how credentials are managed, how failed requests are retried, and how users are notified when updates stop.
CSV integration
If real-time updates are unnecessary, a daily CSV export may be enough to start a pilot. Decide who exports it and how it is transferred. Detect changes in columns, duplicates and missing fields. Include manual export and upload time when measuring the overall benefit.
Separate reading from writing
A first phase can simply read existing information and prepare visit notes. Writing meeting reports back to CRM introduces additional requirements: human review, write permissions, duplicate prevention and audit logs.
AI must preserve existing access boundaries so a user cannot see customer information they were not allowed to view before. Review retention periods, deletion procedures and the AI service’s data handling terms as well.
5. Validate a focused pilot
Define the task and expected output
For example, ask AI to summarize purchase history and the last meeting for accounts covered by one sales team. Record the existing process and time required so you have a baseline.
Check outputs using real data
Confirm that records from different customers are not mixed, information is current and permissions are respected. Check that summaries match their sources and do not add unsupported claims. Users should be able to return to the original evidence.
Measure review and correction time too
A draft produced in seconds may still require lengthy corrections. Compare total time, including data preparation, review, editing and registration in existing systems. Record missed checks, the corrections needed, whether suggestions were used in meetings, and why proposals were rejected.
Use comparable users, scope and measurement periods. If assessing revenue, account for other influences such as price changes and seasonal demand. Expand only after understanding how the workload and quality of the chosen task have changed.
Bring AI into your existing sales workflow
Intentia designs and implements sales AI around each manufacturer’s or wholesaler’s operations and systems. We work with you to identify the task, required data and implementation approach.