A new supplier or competitor price list arrives as a PDF or Excel file. The product names resemble those in your own catalogue, but are written slightly differently. Volumes, units and brands vary. Your procurement team investigates every line again.
This work is more than file processing. The core question is recognition: which external product corresponds to which item in your own range?
AI can prepare that matching. The employee makes the decision.
Want to first assess more broadly which process is suitable for an AI agent? Read Automating business processes with AI: where do you start?.
Why product matching remains difficult
An exact product code or EAN makes matching straightforward. In practice, this information is often missing or differs between sources.
Common differences include:
- abbreviations and spelling variations;
- millilitres versus litres;
- pack size or sales unit;
- brand, sub-brand and private label;
- incomplete descriptions;
- the same product name for different variants.
A solution that only compares text literally will miss many matches. A system that compares too broadly will suggest incorrect products. You need multiple layers of recognition and a clear threshold for human review.
Three layers for a reliable match
QuotePilot uses three types of matching:
1. Exact matching
The agent first looks for hard matches, such as EAN, product code, product name, brand and volume. An exact match requires the least interpretation.
2. Fuzzy matching
The system then recognises small differences in spelling, notation and units. “Heinz ketchup bottle 1000ml” can, for example, be linked to “Heinz Tomato Ketchup 1L”.
3. Semantic matching
When words do not match literally, the agent compares meaning. This helps with synonyms, alternative names and product descriptions that mean the same thing.
For each line, the agent shows the best candidate, an alternative and, where relevant, a white-label or private-label option.
A confidence score helps, but does not decide
Every suggestion receives a confidence score. That score is not an automatic truth. It helps the employee quickly identify which lines are likely to be correct and where review is needed.
You determine in advance:
- which score is sufficient to propose a match directly;
- when multiple candidates should be shown;
- which product groups always require review;
- who may accept or adjust a match;
- how corrections are stored and reused.
This removes recurring search work without changing the catalogue without human judgement.
Your catalogue and ERP remain the source
QuotePilot does not replace your catalogue or ERP. The agent uses product information from your own landscape and links external lines to it.
The checked result can then be exported or returned through an integration to the process you already use. In the VHC Jongens case, the solution consists of a matching engine and a management portal with Microsoft SSO, statistics, synonym management and catalogue data supplied through an ERP integration.
Your existing system continues to determine which products, prices and attributes are authoritative.
Live at VHC Jongens
VHC Jongens receives supplier and competitor price lists in different formats. With up to 75 new products per week, manual matching became increasingly difficult to scale.
rb2 built MatchPoint, the technology behind QuotePilot. The solution reads price lists, finds the best candidates for each product and lets employees accept or adjust the match. Previous decisions are recorded and reused later.
The publicly confirmed results are qualitative:
- less recurring manual work;
- more consistent product matches;
- reusable corrections and decisions;
- human control over every decision.
There is no publicly validated savings percentage. We therefore do not claim one.
Explore rb2’s published cases.
Which numbers should you measure in advance?
Before automation, record:
- number of price lists and lines per week;
- active handling time per list or line;
- percentage of lines that can be matched exactly at once;
- percentage requiring human review;
- number of recurring corrections;
- time from receipt of the file to usable output.
Choose one primary KPI. For many teams, this is handling time per price list. First-time-right rate and exception rate then help explain why the time changes.
An exception rate does not need to fall to zero. Uncertain or business-critical matches should remain with your people.
How to determine whether QuotePilot fits
In a free 30-minute conversation, we assess whether the matching process has enough volume, repetition and ownership.
If it is promising, the paid Prove session follows. Together with the people who know the work, we examine sources, catalogue data, exceptions, systems and the outcome that matters. The result is to build, improve the prerequisites first, or not proceed.