AI Consultant for Logistics Companies: Fixing Dispatch and Customer Updates Without Breaking What Works

Straight answer: an AI consultant for logistics companies earns their fee on two things only: taking manual load-matching and status-chasing off your dispatchers, and stopping customers from ringing to ask where their delivery is. If a consultant's pitch starts with "full automation" or "AI-powered fleet intelligence" before they've watched your dispatchers work a real shift, walk away. The good ones start with a stopwatch, not a slide deck.

What dispatch chaos costs you before anyone mentions AI

I've sat in enough dispatch offices to know the pattern by now. Three or four dispatchers, six phone lines, a whiteboard or a spreadsheet nobody trusts fully, and a WhatsApp group with the drivers that everyone secretly relies on more than the tracking software they paid for. The dispatchers aren't inefficient. They're firefighting. A customer calls to ask where their pallet is. A driver calls to say the drop point is closed. Another customer emails asking for a proof of delivery from three days ago. None of that is dispatching. It's customer service wearing a dispatcher's badge, and it eats the hours that should go to planning routes and matching loads.

The number that matters here isn't fleet size. It's call volume per dispatcher. A haulage firm I worked with in the Midlands had four dispatchers running a 60-vehicle fleet, and between them they were fielding around 300 inbound calls a day. My colleague and I logged a full week of calls with them before touching anything. 41 percent of those calls were customers or their own sales team asking "where is it." Not a complaint, not an exception, just a status check that someone had to stop and manually look up.

The case that changed how I sell this

That Midlands client had already been burned once. A software vendor had sold them a "fully automated dispatch AI" eighteen months earlier for a five-figure sum, and it lasted eleven weeks before the dispatchers quietly went back to the whiteboard because the system kept assigning loads to drivers who were on their rest break. Nobody had built in HGV driving hours rules. The AI wasn't wrong exactly, it just didn't know what it didn't know, and the vendor had never sat in that office to find out. Dispatch and fulfilment teams with a result like this can write a logistics guest post.

So we didn't touch load assignment. We started with the 41 percent, the status calls. We connected their existing tracking (they were on a Samsara telematics setup) to a simple automated WhatsApp and SMS update flow: booking confirmed, driver assigned with ETA window, out for delivery, delivered with a photo link, and a proactive delay message the moment a job slipped more than 30 minutes behind schedule, sent before the customer had to call and ask. Nine weeks of build and testing. Total cost including my time and a developer for the integration came to just under 14,000 pounds.

Inbound calls dropped from roughly 300 a day to just under 160 within the first month, and stayed there. The dispatchers didn't lose their jobs. They got their afternoons back to plan the next day's routes instead of repeating the same "it's on the lorry, mate, should be with you by 3" conversation forty times a day. That's the win. Not headcount reduction. Time reallocation.

Where a good AI consultant starts (and it's rarely with the AI)

Before anyone recommends a tool, they should be asking to see your actual call logs, your delivery exception rate, and how many status updates your team sends manually per day. If a consultant can't tell you your current "where is it" call percentage within the first meeting, they haven't done the diagnostic work and they're about to sell you a solution to a guess.

The order that works, every time I've run it:

• Map the exact journey from booking to delivery confirmation, including every point a human currently has to check something manually
• Log real call and message volume for at least five working days, not a "typical" estimate from memory
• Separate what's a data problem (drivers not updating status) from what's a communication problem (customers not being told without asking)
• Fix the data problem first, always, because automated updates built on bad data just automate the confusion faster
• Only then automate the customer-facing update layer
• Leave dispatch decision-making (load matching, route sequencing) as a human-plus-suggestion tool, not full automation, unless you're running a fleet over 200 vehicles with standardised routes

Dispatch automation: what works at this scale

For fleets under 100 vehicles, full AI-driven load assignment rarely earns its cost. The variables (driver preference, vehicle suitability, customer relationships, one-off exceptions) are too messy for a smaller data set to learn well, and your dispatchers already hold that knowledge in their heads. What does work is a suggestion layer: the system proposes three viable driver-load matches based on location, hours remaining, and vehicle type, and a human picks one in ten seconds instead of working it out from scratch in three minutes. Route optimisation tools like Route4Me or Onfleet do this reasonably, and they're a fraction of the cost of a custom-built dispatch AI.

For fleets over 150 to 200 vehicles running fairly standardised routes (parcel delivery, multi-drop retail, regional distribution), fuller automation starts to pay for itself because the pattern repeats enough for the model to learn something real. That's a different project with a different price tag, usually starting around 30,000 pounds and running months, not weeks.

Customer updates: the channel matters more than the AI

Here's the bit most consultants skip past too fast. It's not about which AI model drafts the message. It's about which channel your customers check. B2B logistics clients (the warehouse manager, the retail buyer) almost always prefer email with a tracking link they can forward to their own team. Consumer-facing last-mile deliveries respond far better to SMS or WhatsApp because that's what gets read within minutes, not left in an inbox. Mixing this up is the single most common reason a "smart update" system gets ignored and the calls come right back.

Twilio and WhatsApp Business API cover most of the technical delivery. The actual work is writing update templates that sound like a person, not a system: "Hi Dave, your delivery from Hartlake Logistics is running about 25 minutes behind schedule due to traffic on the A14, new ETA 2:40pm" beats "DELAY NOTIFICATION: SHIPMENT #48291 STATUS UPDATE" every single time, and it's the difference between a customer who trusts you and one who calls to double check anyway.

What this costs

For a small to mid-size fleet (20 to 80 vehicles), a proper diagnostic plus a customer update automation build typically runs 8,000 to 18,000 pounds over 6 to 10 weeks, depending on how messy your current tracking data is. Day rates for an experienced AI consultant in this space in the UK sit between 900 and 1,600 pounds. If someone quotes you a flat "AI dispatch system, 3,000 pounds, live in a week," ask what data source it's pulling from, because it almost certainly isn't yours, and it definitely hasn't been tested against your exceptions. If you're weighing up whether to hire for a project like this or bring someone in longer term to build the whole thing out with your team, I've written more on what that decision looks like and what it should cost on my AI implementation coaching page.

The uncomfortable part nobody selling this tells you

Automating customer updates does not eliminate the phone calls entirely, and any consultant promising it will is either inexperienced or lying to close the deal. What it does is shift the calls from routine status checks to genuine exceptions, meaning the calls you still get are harder, angrier, and take longer, because the easy ones have already been handled automatically. Your dispatchers need to be trained for that shift, not just handed fewer calls and left to figure it out. The Midlands client we worked with saw their average call handling time go up slightly, from 3 minutes to just under 5, because the remaining calls were the messy 15 percent of jobs that had gone wrong. Fewer calls, harder calls. Nobody puts that in the case study slide, but it's the honest trade you're making, and it's still worth it.

Questions to ask before you sign a contract

• Ask to see the call log analysis before they propose anything, not after
• Ask what happens when the telematics or booking data is wrong or missing, because it will be, and the plan needs to account for that
• Ask whether they're keeping dispatchers in the decision loop for load matching, or trying to remove them, and ask why
• Ask for one reference client at a similar fleet size, and call them
• Ask what the customer update templates look like before build starts, not after, because rewriting them later costs more than getting them right first

Logistics is one of the few sectors where "AI project" doesn't mean chatbots or content generation, it means fixing the plumbing between your booking system, your telematics, and your customer's phone. That's a less glamorous sell than most consultants want to make it, which is exactly why the ones who focus on it tend to be worth hiring.

Two pages that follow on from this: hiring an AI consultant (what to ask, what to expect) and AI automation consultant (what the work looks like week by week).

Every industry guide I have written is collected at AI consultant by industry.

Frequently asked questions

How much does an AI consultant for logistics dispatch cost in the UK?

Day rates typically run 900 to 1,600 pounds, and a full diagnostic plus customer update automation project for a 20 to 80 vehicle fleet usually comes to 8,000 to 18,000 pounds over 6 to 10 weeks.

Will AI replace dispatchers in logistics companies?

Not at fleet sizes under roughly 150 to 200 vehicles. AI works best as a suggestion tool that speeds up load matching and handles routine status updates, freeing dispatchers for exceptions and route planning rather than replacing them.

What's the fastest win an AI consultant can deliver for logistics customer updates?

Automated proactive status messages (booking confirmed, driver assigned, out for delivery, delayed) sent by SMS or WhatsApp before the customer has to call. In one project this cut inbound status calls by nearly half within a month.

Does automating customer updates reduce call volume to zero?

No, and anyone claiming that isn't being straight with you. It removes the routine "where is it" calls but the remaining calls tend to be genuine exceptions, which take longer to handle, so dispatchers need training on that shift, not just fewer calls to answer.

Useful references

• Anthropic's own documentation
• the NIST AI Risk Management Framework