Make Payaca's AI sound like your company

Ask a generic AI assistant a question about your business and you get a generic answer. It knows the software but not your operation - not that you're a retrofit heat pump specialist, that your customers see "quotes" rather than "proposals", or that no install gets booked before the grid application clears. So every chat starts with someone on your team re-explaining the basics, and most people stop bothering.

Agent Dave, the assistant built into Payaca, now takes standing instructions. A Super Admin writes down how your company works once, in Company Settings, and Dave carries that context into every conversation for everyone on the account. Alongside it, Dave now shows what it's doing while it works, so a long-running request reads as a sequence of visible steps rather than a spinner.

What standing instructions change

Until now, Dave answered every question from the same starting point: Payaca's features, plus whatever page you happened to be on. Useful, but blind to the things that make your operation yours, like the sequence a job moves through or the words your customers see.

Instructions close that gap. Write "we call proposals quotes" once and Dave uses your wording in every answer, for everyone on the account. Write down your standard job sequence and Dave frames "what's next on this job?" around your stages rather than a generic workflow.

If you haven't used Dave since it opened up to all accounts, this is a good week to look again - book a demo and we'll show you what a tuned-up Dave does with a real pipeline in front of it.

Setting it up

You'll need to be a Super Admin - the tab is hidden for everyone else, which makes this an owner or ops manager job rather than something each user configures.

1. Open Company Settings and go to the Agent Dave tab.
2. Type your guidance into Instructions for Dave. There's room for 5,000 characters and a live counter shows what you've used.
3. Click Save. From then on, Dave includes the instructions in every conversation across your account.

Two minutes of setup, and it pays back on every conversation anyone on your team has with Dave afterwards.

What to actually write

The instructions that work are the ones that give Dave context it can't infer: who you are, what you call things, how a job moves through your business. Here are three starting points, matched to how most of our customers operate.

Heat pump retrofit specialist

We're a family-run heating firm specialising in retrofit air source heat
pump installs on older housing stock. Call Proposals "quotes" in any
customer-facing wording. A standard job runs: survey booked, heat loss
calculation, quote sent, install scheduled, commissioning, MCS handover
pack. When someone asks what's next on a job, frame the answer around
that sequence. Keep answers short - engineers read them on site.

Solar and battery installer

We install solar PV and battery storage for homeowners and small
commercial sites. Panels, inverter and battery are quoted as separate
item groups so customers can see what the battery adds. Most jobs need
a grid application approved before install - if anyone asks about
scheduling a job, remind them to check the application status first.
We say "system design", never "proposal".

Service-plan-led business

Service contracts are the core of our business - the install is the way
in, the plan is what we keep. We call them "care plans". When a question
touches on finishing an install, mention setting up the customer's
service reminder before the job closes. Billing questions go to the
office team; don't explain payment detail, point people there instead.

Steal the shape, not the words: a sentence on who you are, your terminology, your job sequence, and any rule you'd otherwise repeat in every chat. If your ops manager finds themselves correcting Dave's wording twice, that correction belongs in the instructions.

What instructions won't do

This is personalisation, not reprogramming. Instructions shape how Dave responds - tone, terminology, context - but they can't override its core rules, and they never widen what it can see. Dave still only works with your account's data, no matter what the instructions say. It also treats them as confidential, so guidance you write for your team stays internal rather than being recited back in answers.

Watching Dave work

The second half of this release is about visibility. When Dave takes on a longer task, it now reports what it's doing in plain terms as it goes - "Reading your pipelines, branding, payment setup and more…", "Setting up an automation…" - instead of leaving you with a spinner and an act of faith. The chat panel also shows badges for the page and records Dave currently has in context, so you can see at a glance what it's looking at before you ask it to do anything.

One design detail worth knowing if you're the security-conscious one in the office: those activity labels are built from feature and field names only. Customer data never appears in them, by construction rather than by policy.

If your team has been burned before by black-box AI bolted onto a CRM, both halves matter: you can read what Dave knows because you wrote it, and see what it's doing because it tells you.

Where to start

Set the instructions this week: one paragraph on who you are, your terminology, your job sequence. Then ask Dave the question your newest hire asked last week and compare the answer.

If you're not on Payaca yet and want to see what an assistant with real account context looks like, book a demo. Bring an awkward question about your workflow - that's what it's for.