AI Controls#

The AI Controls decide how hard the system works to match each product line in an incoming email to the right product in your catalogue. Open them from the AI Connect / settings page — click the Controls card.

There are only three switches. Everything else (how the matching is batched, when to look deeper, how many candidates to re-check) is handled automatically — you never have to tune it.

The three switches#

Maximum accuracy#

Makes the system look at your entire catalogue when matching a product, instead of stopping early once it finds a strong match. Turn this on when products are being matched to the wrong code and you'd rather be thorough than fast.

Smarter head matching#

When matching the product head (the top-level product family), the system normally matches on the name alone. With this on, it also considers each head's full/external name and its units, plus the unit the buyer asked for, and falls back to each head's description when it's still unsure — useful when several heads have similar short names but mean different things, or when the unit (e.g. metres vs pieces) is what tells them apart.

Smarter code matching#

The same idea, applied to the product code (the specific item under a head). With this on, the system reads code descriptions to tell apart codes that look alike by name.

All three are off by default. You can turn on any combination.

Will this cost more?#

The honest answer: a little, but the system is built to keep that cost as low as possible — and it does it for you automatically.

Reading descriptions uses more of the AI's effort (and therefore costs your firm more) than reading names alone. So the system only spends that effort when it's actually worth it:

• It first matches using names only — fast and cheap.
• It looks at descriptions only when it's genuinely unsure — when its confidence in the best name-only match falls below about 70%.
• Even then, it doesn't re-read the whole catalogue. It only re-checks the handful of candidates it was torn between — the few that looked plausible — not every product you sell.

So "smarter matching" mostly improves accuracy on the tricky lines, while the easy, obvious matches stay just as cheap as before. When the AI is confident, it never wastes money double-checking.

You don't get a knob for the 70% threshold, the batch sizes, or how many candidates get re-checked. That's deliberate — fewer options means the system can keep optimizing cost for you. The right amount of extra effort is decided per email, automatically.

When is "Maximum accuracy" worth it?#

For most firms, leave Maximum accuracy off. The default already finds strong matches quickly.

Consider turning it on if:

• Your catalogue is very large — roughly more than 500 product heads — and
• You're seeing products matched to the wrong code often enough that accuracy matters more than speed and cost.

The trade-off: with Maximum accuracy on, the system considers far more of your catalogue for every line, so each email takes a bit longer and costs a bit more to process. On a big catalogue that extra thoroughness is usually what fixes mis-matches; on a small catalogue it rarely changes the result, so it's not worth the cost.

A good approach: leave everything off, watch your AI Inbox for a few days, and only switch on the control that targets the problem you're actually seeing (wrong head → Smarter head matching; wrong code → Smarter code matching; missing the right answer entirely on a huge catalogue → Maximum accuracy).

Which emails to process#

Your listening inbox receives mail addressed to you directly (To), in copy (CC), or silently (BCC / list mail / forwards). Mail sent directly to you is always processed. Two switches control the rest:

• Process emails you're CC'd on — read & stage mail where your inbox is only in the CC line.
• Process emails you're BCC'd on — read & stage mail that reached your inbox but lists you in neither To nor CC.

Both are on by default. Turn one off if those copies are mostly noise and you don't want to spend tokens on them.

Blacklisted senders#

Some senders never send anything worth processing — a marketing list, a noisy notification address, a whole domain. The Blacklisted senders screen (the Senders card in AI Controls) lets you skip them. Blacklisted mail is dropped before the AI reads it, so it costs zero tokens.

You can block three ways:

• Email address — one exact sender, e.g. promo@acme.com.
• Domain — everyone at a domain, e.g. @acme.com (or just acme.com).
• Wildcard pattern — use * as a wildcard for anything more specific, e.g. sales*@*.xyz.com or raju@*.

Use the search box to find an existing rule, Add to create one, and the row actions to edit or remove a rule. Removing a rule means that sender's mail will be processed again.

A known customer or supplier is never treated as spam automatically — but a blacklist rule always wins. If you blacklist a sender, their mail is dropped even if they're in your records, so only blacklist senders you're sure you never need.

Nothing changes about review#

These controls only affect how matches are found and which mail is processed. Every processed email still waits in the AI Inbox for a person to review and approve — see Reviewing & approving. Better matching just means fewer corrections for you to make.