Effective AI Queries in Swedish

AI Boss is flexible – but like a new coworker, it performs best when you're clear about what you want. Here's a guide to phrasings that work and pitfalls to avoid.

Principle 1: Be specific about time

Less good: "How have we been selling?" Good: "How did we sell last week vs. the same week last year?"

AI can ask back if you're vague – but you save time by being concrete upfront. Use time references like:

• Today / yesterday / tomorrow
• Last week / this week
• Last month / this month
• Last 7 days / last 30 days
• Q1 2026 / March 2026
• 2024-01-15 to 2024-01-31 (date range)

Principle 2: Comparison is worth more than absolute numbers

Less good: "How many bookings do we have today?" Good: "How many bookings do we have today compared to last week?"

A number without context is hard to interpret. Always add a comparison – AI handles it happily.

Principle 3: Use real names

Less good: "Look up that lady from Lysekil" Good: "Find the guest Andersson from Lysekil"

AI has access to the full customer registry. If you know the first name – write it. If you know the phone number – even better.

25 Swedish queries that work well

Sales and economy

1. "Vad var försäljningen igår?"
2. "Hur går det idag jämfört med samma dag förra veckan?"
3. "Visa mig timfördelningen för lördag"
4. "Vilken kassa sålde mest i förra veckan?"
5. "Hur stor andel av försäljningen är online vs. kassa denna månad?"

Menu and products

1. "Vilka 5 rätter säljer sämst senaste 30 dagarna?"
2. "Vad är marginalen på varje pizza?"
3. "Vilka rätter innehåller gluten?"
4. "Höj alla ölpriser med 10 kr" (AI confirms first)
5. "Lägg till Bearnaisesås 25 kr under Tillval"

Customers and CRM

1. "Vilka VIP-kunder har inte varit på 90 dagar?"
2. "Visa Anderssons orderhistorik"
3. "Vem är vår bästa kund mätt i total spend?"
4. "Lägg till taggen 'Allergisk mot nötter' på kund nummer 42"

Bookings

1. "Vilka har bokat idag?"
2. "Boka bord för 4 imorgon 19:00 åt Johanna Eriksson"
3. "Visa väntelistan för fredag"
4. "Har vi bokningar för över 8 personer nästa vecka?"

Marketing

1. "Skicka 10 % rabatt till alla VIP som inte varit på 60 dagar"
2. "Hur presterade min senaste kampanj?"
3. "Skapa en automation: SMS till gäster som fyller år samma dag"

Staff

1. "Vem jobbar i kväll?"
2. "Föreslå ett schema för nästa vecka baserat på försäljningsprognos"
3. "Vilka är nära övertid denna vecka?"
4. "Vad är personalkostnaden i procent av försäljningen i mars?"

Common pitfalls

Pitfall 1: Too vague

❌ "What do you think?" Without context AI can't help. Ask instead: "Which dishes should I promote this weekend?"

Pitfall 2: Multiple questions in one

❌ "How did yesterday go and who works tomorrow and add pizza?" AI gets lost. Ask one question at a time.

Pitfall 3: Assuming memory

❌ (after discussing a campaign) "Send it now" If the campaign was discussed earlier in the chat it works. But in a new session AI doesn't remember. Give full context: "Send the Christmas campaign we created yesterday".

Pitfall 4: English in the middle of Swedish

❌ "Vilka customers har varit här last week?" It works – but AI responds better if you stick to one language. Start in Swedish, stay in Swedish.

Pitfall 5: Over-politeness

❌ "Excuse me, would it be possible if you could maybe check whether..." Short and direct saves time: "How many guests did we have last Friday?"

Two powerful patterns

Pattern 1: "Break down by X"

Use "dela upp" (break down) or "per X" to get a split:

• "Dela upp försäljningen per kategori"
• "Per servitör, vem tog flest ordrar?"
• "Per veckodag, vilken är mest lönsam?"

Pattern 2: "Jämfört med" (compared to)

Use "jämfört med" to force a comparison:

• "Denna månads försäljning jämfört med samma månad förra året"
• "Lördagskväll jämfört med fredagskväll"
• "Online-ordrar jämfört med kassa"

When to push the pro model

AI Boss uses two models automatically: flash (fast) and pro (deep). You don't need to choose – but for these questions AI escalates to pro:

• Analysis of photos (menu, ingredient list)
• Complex multi-step (e.g. "suggest a schedule + flag all rule violations")
• Long reports over full quarters
• Menu engineering analysis (Stars/Dogs)

See a "Thinking..." indicator for 3–5 seconds? That's AI running the pro model for a better answer.

When AI says "I don't know"

Sometimes AI says:

"I don't have access to that data right now – try checking admin settings or reach out to Vendion support."

It can be because:

• The data is in another system (e.g. payroll in Fortnox – not in Vendion)
• The question requires rules AI doesn't have (e.g. legal interpretation)
• Something is broken (contact Vendion support)

It's a feature, not a bug. AI refuses to guess rather than give wrong answers.

Next steps

• AI Boss overview (article 66)
• AI tool reference, all 95+ tools (article 78)
• Getting started with AI Boss (article 76)