Business AI Coach

Results

What it looks like when AI works inside a business.

Three engagements inside businesses we operate, in three different sectors, written up the way an owner would want to read them: the business problem, why it mattered, what changed, what AI now does, what humans no longer need to do, and the number at the end. The technology comes last, because it mattered least.

Hospitality group, 8 venues~70%less rota administration time

Commercial result

  • Rota administration time down by around 70%
  • Roughly two working days a week returned to each GM
  • Payroll corrections down sharply
  • Rolled out across the group in a few weeks

The business problem

Eight venues, eight general managers, and every one of them spending the best part of a day a week building rotas, chasing swaps, checking hours against forecasts and fixing the payroll fallout when something slipped through.

Why it mattered

Most of a day a week of senior management time per venue was going on administration rather than guests, teams and sales. Payroll corrections were a monthly argument with the finance team and a recurring source of staff frustration.

What we changed

We redesigned the process rather than bolting a tool onto the old one. Forecast demand drives a draft rota, the draft goes to the GM for judgement rather than construction, swaps and absences flow through one channel, and hours are reconciled against the schedule before payroll ever sees them.

What AI now does

Builds the draft rota from trading forecasts and staff availability, flags conflicts and compliance issues, handles the routine swap requests, and reconciles clocked hours against the schedule so only genuine exceptions reach a human.

What humans no longer need to do

Building rotas from a blank sheet. Chasing swaps over WhatsApp. Manually cross-checking hours before payroll. GMs now review and approve instead of construct.

What happened next

The first attempt, a no-code builder, was scrapped after a week because it could not handle multi-site rules. The rebuilt version became the template for how the group now approaches every management process: forecast-driven, exception-based, owned by the GM.

Technology used: Claude · Planday · Lightspeed · Slack

B2B marketplace<30 minlead-to-call time, down from around two days

Commercial result

  • Lead-to-call time from around two days to under half an hour
  • Conversion up by about a third
  • Most of a day a week freed across the sales team

The business problem

Inbound leads from venues and chefs landed in a shared inbox and a CRM nobody trusted. Qualification happened whenever someone got round to it, which was typically two days later, by which point the lead had often gone cold or gone elsewhere.

Why it mattered

In a marketplace, liquidity is everything. Every slow response was a lost listing or a lost booking, and the sales team was spending a third of its week on triage rather than selling.

What we changed

Redesigned the front of the sales process so that qualification is instant and the first human touch is a conversation, not a data-entry task. Every lead is classified, enriched and routed the moment it arrives, with a call booked for the ones that matter.

What AI now does

Reads each enquiry, classifies it by type and value, enriches it from the CRM and public sources, drafts the first response, and books qualified leads straight into the right person's calendar. Low-value or mis-routed enquiries go to a review queue rather than the sales team.

What humans no longer need to do

Reading and sorting the inbox. Copying details into the CRM. Deciding who should call whom. The team now picks up the phone to people who are already qualified.

What happened next

The first classifier mis-routed edge cases, which is why a human review queue stayed in the design. The same pattern of instant classification, human conversation, exceptions to a queue is now how the business handles supplier enquiries too.

Technology used: Claude · HubSpot · Slack

Coffee wholesale<2 dayscustomer onboarding, down from around two weeks

Commercial result

  • Onboarding time from around two weeks to under two days
  • Monthly sign-up capacity up by well over a third
  • Most steps fully self-serve
  • Planned sales-operations hire deferred

The business problem

Onboarding a new wholesale account took two weeks of back and forth: paperwork, account set-up, pricing, machine logistics, first order. Each step waited on a person, and the founder was the bottleneck for half of them.

Why it mattered

Two weeks of delay on every new account meant two weeks of revenue not recognised, a poor first impression for the customer, and a cap on how many accounts the business could open each month without hiring.

What we changed

Redesigned onboarding as a self-serve flow with human checkpoints only where judgement or a signature is needed. The customer moves themselves through most of it; the team is pulled in at the two points that matter.

What AI now does

Collects the account details, checks them, drafts the agreement from the agreed terms, sets up billing, schedules machine delivery and first order, and flags anything non-standard to a person before it goes out.

What humans no longer need to do

Chasing forms. Rekeying details into billing. Manually scheduling deliveries. The founder stopped being copied on every onboarding email.

What happened next

The agreement drafting needed a week of legal review before it went live, which is the right order to do things in. The onboarding flow is now the model for how the business handles renewals and machine servicing.

Technology used: Claude · Tally · Stripe

Want this inside your business?

The sectors differ. The problems don't. Every one of these started with a two-week look at the business and a ranked list of where AI would pay back. That is the AI Opportunity Review.

Book a call with JackAbout the Opportunity Review →