A
Sales copilot
a language model with retrieval over the catalogue, three suggestions with reasoning, a cross-sell prompt
Boundary rule
Never sends an offer. The rep clicks send.
Solutions
Sales copilot and margin guard for distribution
wholesalers and distributors with a broad catalogue and several branches
What does not work today
- a rep searches the ERP, catalogues and spreadsheets by hand while the client waits for hours
- discounts are given for a quiet life and the margin has fallen three years running
- knowledge about substitutes sits in the heads of the best reps, not in the system
What you get
Three stages, each with its own value and its own price. You buy them in sequence, not all at once.
01 Process mapping
enquiry, matching, substitute, discount, quote, reply
02 Panel
- · enquiry view
- · client history
- · margin in one place
- · dashboard for the CFO
03 AI support
- · Sales copilot
- · Margin guard
- · Real availability and substitute map
Three AI modules, and what none of them will do on its own
Every module carries a boundary rule, the line beyond which the decision belongs to a person. The rule is part of the build, not a marketing claim.
B
Margin guard
flags anything below the profitability threshold and offers a higher margin alternative
Boundary rule
Never blocks an offer from going out. It warns and suggests, the rep decides.
C
Real availability and substitute map
real availability and ETA instead of the declared one, a substitute map with regional stock
Boundary rule
Does not reserve stock automatically.
Security and data
The same rules apply in every panel, whatever the industry. They are not an option to be priced separately.
- Row level security at database level
- The access rule lives in the database itself, not in application code. A bug in the application does not expose another client's records.
- Append-only audit log
- Who, what and when. An entry cannot be altered or deleted, not even from an administrator account.
- TOTP multi-factor authentication
- A second login factor from an authenticator app, without SMS codes.
- File links valid for 15 minutes
- A file address expires. Forwarded on or copied from browser history, it stops working.
- Data inside the European Union
- Database and files in an EU region. The sub-processor list is public and kept current.
- Data processing agreement
- Signed before the first data import, not after go-live.
Regulatory risks in this sector
- · classification under the AI Act as a limited risk system
- · disclosure duties towards the user
Stages and decision gates
After each stage there is a decision on whether to continue. If a stage did not deliver what it was meant to, the next one does not start.
- 01
Process mapping
Gate: continue, or stop at this stage.
- 02
Panel
Gate: continue, or stop at this stage.
- 03
AI support
Gate: continue, or stop at this stage.
Questions that come up most often
- Will this work on my catalogue if the data is incomplete?
- A messy catalogue is the rule rather than the exception, which is exactly why mapping comes first. We check how many items have a description good enough for matching, how many carry substitute links and where stock levels come from. If it turns out the data is too thin for a copilot to work on, you hear that after mapping rather than after signing for the whole project.
- How is this different from the search we already have in our ERP?
- Search in a trading system matches on codes and names, so it finds things once the rep already knows what to look for. The copilot works from a descriptive request, joins the catalogue with the client's order history and returns three suggestions with reasoning, plus a substitute when an item is out. It pays back most with new reps who do not yet carry substitute knowledge in their heads.
- Will the margin guard block an offer below the threshold?
- Never. The guard marks the line, shows the profitability threshold and proposes a higher margin alternative, but the send button belongs to the rep. A block would turn the tool into another system worked around by email. The point is that the rep sees the effect of a discount as they grant it, not a month later in a report.
- Who sees the margin, and will reps learn our purchase prices?
- Visibility is agreed during mapping and enforced by a rule in the database rather than a setting in the interface. The usual arrangement is a percentage margin for the rep without the purchase price, full figures for the sales director and an aggregate dashboard for the CFO. All three views run on the same data and differ only in what the access rule lets through.
- How do you measure the return on this?
- We measure two things before and after the pilot: average time from enquiry to reply, and average margin on an offer. The baseline is collected during mapping so there is no argument later about what we are counting from. With a catalogue above a few thousand items, the drop in response time usually shows before the margin moves, because margin reacts more slowly.
- Does this fall under the AI Act and what does that mean for us?
- A system supporting sales sits in the limited risk category, so the main duty is transparency: the user has to know a suggestion came from a system rather than a person. We mark that next to every suggestion in the panel. Because no module decides on its own, we stay out of the high risk category and its much heavier documentation duties.
Let us check whether this fits your company
Instead of a contact form we ask for five things. On that basis we come back with a price range and a proposed mapping scope, or with a straight answer that in your case the numbers do not add up.
- sector and line of business
- company size and how many people sit in the process
- systems you use today
- the process that hurts most
- rough budget