Independent editorial guide, written with AI assistance. Examples are hypothetical; no hands-on product testing or owner interviews were performed. Features and prices may change. Our editorial method.
ShiftToAGI, Shifting Founders Beyond AI. Evidence note: we checked product features against the official FourKites homepage and Digital Workforce page on 2 November 2026. Neither page showed pricing. ShiftToAGI is not affiliated with FourKites.
FourKites is a supply chain visibility and orchestration platform. It pairs live shipment tracking with named AI agents that contact carriers, answer customer status questions and adjust dock appointments. The company's own site describes the product as AI-powered supply chain orchestration. The question for an operator is whether it can take on the chasing that follows a late truck while leaving decisions with the people who are accountable for them.
A hypothetical Tuesday at a regional distributor
This scenario is fictional. Priya runs operations for a mid-sized food distributor with two warehouses and about forty outbound loads a day. At 07:40 a grocery customer emails to ask where a chilled pallet order is. The carrier portal still shows the load in transit, and it has not updated since 00:10. The customer's receiving dock closes at noon. Priya's planner is on the phone with a different carrier, and the account manager wants a reply out within the hour. No one knows the real ETA, and three people are about to give the customer three different answers.
The problem has two halves: too little information, and too many people improvising to fill the gap. The rest of this profile sets out what FourKites documents, where it might fit, and what a careful pilot could look like.
What is FourKites?
The official pages present three connected layers:
- Data and tracking. Five "Digital Twins". The one most relevant here is the Shipment Twin, which the vendor says covers tracking across modes, ML-based ETAs, carrier status, exception detection and proof of delivery (homepage).
- Orchestration. Loft, which turns plain-English SOPs into workflows that are version-controlled and auditable, with human oversight described as part of the design (homepage).
- Agents. Six named "Digital Workers" covering carrier, supplier, facility, customer, document and workflow-building tasks (Digital Workforce).
The features fall into three groups:
- General platform features: tracking, dashboards and carrier data connections.
- Ordinary automation: rule-like steps such as filing a document or moving an appointment after an ETA changes.
- AI features, as the vendor describes them: ML-predicted ETAs, parsing of unstructured supplier emails and PDFs, conversational analytics, and turning plain-language instructions into agent workflows.
For a small team, most of the value would come from the AI group. That group also carries most of the risk.
Which FourKites features matter for delay exceptions?
All five features below are summarised from the Digital Workforce page unless another source is linked.
1. Tracy, carrier follow-up. Tracy contacts carriers by email, WhatsApp, voice and SMS. She also reconciles predicted ETAs against carrier signals and escalates exceptions. In Priya's case, Tracy is the agent that would chase the silent carrier.
2. Cassie, customer status queries. Cassie triages email and chat questions about order status and answers them from Shipment Twin data. When a case needs a person, she hands it over with the conversation history. Because she speaks directly to your customers, she is the most sensitive agent in this profile.
3. Alan, dock appointments. Alan creates and reschedules dock appointments from live ETAs and reslots them when disruptions occur. The rescheduling itself is largely mechanical. The judgement sits in deciding which ETA to trust.
4. Sam and Polly, documents. Sam parses supplier emails and PDFs, links new shipments to POs and flags incomplete supplier data. Polly chases missing proof-of-delivery documents, then validates and files them.
5. Sophie and Loft, workflow building. Sophie converts a process written in plain language into a Loft workflow and assigns it to an existing agent or a new one. Separately, FourSight AI answers plain-English questions about network data (homepage).
The vendor presents these agents as autonomous and running around the clock. That describes how the product is designed. It does not show that your carriers will reply or that ETAs will be accurate on your lanes.
Who is FourKites for?
The site lists industries including food and beverage, retail, CPG, manufacturing and 3PL (homepage). Its customer stories feature large brands, and the vendor refers to running at Fortune 500 scale (Digital Workforce).
The best fit is a shipper or logistics team that moves a steady flow of loads across several modes and already loses staff hours to status calls. If you run a few dozen loads a week, assume the product is aimed above you and check fit before you invest time in a demo.
Practical applications by role
These are proposed uses that map documented features to common jobs. None of them has been validated.
- Operations lead: use Tracy and the Shipment Twin to agree a single ETA for each late load, so the team stops comparing three portals.
- Customer service manager: let Cassie handle routine order-status questions, with key accounts and claims routed to a named person.
- Dock supervisor: use Alan to move inbound appointments when a load slips, so a dead slot is released instead of held open.
- Finance or billing: use Polly's POD chasing so missing delivery paperwork does not hold up invoices.
- Procurement: use Sam to catch supplier shipment notices that are missing PO numbers or quantities before the goods arrive at receiving.
Illustrative workflow: handling a delayed outbound load
This is a proposed workflow built on fictional data. It is not a product test. Thresholds, channel order, approval gates and record fields are configuration unknowns, and you should confirm each one with FourKites.
Fictional input: Load DX-2207 is three pallets of chilled dairy, promised to the Greenfield Market DC by 11:30. The dock closes at 12:00. The last carrier ping arrived at 00:10, and the customer emailed at 07:40.
- Detect. The Shipment Twin's exception detection flags DX-2207. Under the proposed rule, a load with no carrier update for 4 hours becomes an exception. On this data the flag would fire at 04:10, more than three hours before the customer asked.
- Chase. Tracy contacts the carrier dispatcher. The proposed sequence is email first, then SMS after 20 minutes, with a maximum of 3 attempts. Whether you can configure that sequence is unknown. Use carrier business contacts only, and record which channel each contact used.
- Reconcile. Produce one exception record. Sample output: "DX-2207: carrier reports a breakdown near exit 14. Revised ETA 13:15, which is after the 12:00 dock close. Source: carrier dispatcher. Model ETA not yet reconciled. Reefer temperature: UNKNOWN." Whether temperature appears in this record at all is unconfirmed.
- Handle missing data. Any unknown cause, ETA or temperature is written as UNKNOWN and never estimated. A chilled load with an unknown temperature goes straight to the ops lead. Automated assurances about product condition stay blocked. The human account owner can send a verified delay update through the existing incident process, stating what is known and what is still being checked.
- Get human approval. The account manager approves every customer reply before it is sent. Ask FourKites whether Cassie's outbound messages can be held for approval. If the vendor cannot confirm this, keep customer replies manual for the whole pilot.
- Reschedule. Because 13:15 misses the dock window, a new slot is needed. Alan's documented behaviour is to reslot automatically. Ask whether that can become a proposal instead. If proposal-only operation cannot be verified, leave Alan disabled for this pilot and let the supervisor reschedule manually. A supervisor approves any move that displaces another booking or incurs a fee.
- Close and log. Polly chases the POD after delivery. Review Loft's audit trail every week, and ask whether it can be exported.
- Stop route. Pause the agents and return to manual handling if a customer receives a wrong ETA, if any message goes out without approval, or if carriers complain about how often they are contacted.
Pilot checks over four to six weeks, on one lane or one customer:
- the share of exceptions with a reconciled ETA before the customer asks;
- agent ETA compared with actual arrival time;
- the number of unapproved outbound messages, with a target of zero;
- staff minutes per exception, logged by hand both before and during the pilot.
Fix your baseline before you start. Do not use vendor figures as the baseline.
Copyable configuration brief (plain-English SOP draft)
Bring this to a vendor workshop or a Sophie and Loft session. The syntax is ours. It is not a confirmed Loft or AOP format.
PROCESS: Outbound delay exception, chilled and ambient loads
TRIGGER: ETA slips >30 min past promised window, OR no carrier update for 4 hours
STEP 1: Contact carrier dispatcher by email; after 20 min, SMS. Max 3 attempts per load.
STEP 2: Record cause, revised ETA, ETA source, reefer temp. Write UNKNOWN if not confirmed. Never estimate.
STEP 3: If chilled AND temp UNKNOWN or out of range: escalate to ops lead. No customer message.
STEP 4: Draft customer update. Do NOT send. Route to account owner for approval.
STEP 5: If revised ETA misses dock window: propose new slot; supervisor approves if it moves another booking or incurs a fee.
STEP 6: Log all contacts and decisions. Store carrier business contacts only; no driver personal data.
STOP: Any unapproved customer message or wrong ETA sent = pause workflow, notify ops lead.
What does FourKites cost, and what are the limits?
Numeric pricing has not been verified. Neither the homepage nor the Digital Workforce page lists prices, plans, billing intervals or units. Both pages offer links to speak with an advisor. The absence of a published price does not mean a free tier exists.
The reviewed pages also leave several questions open. Get written answers on each:
- Which agents come bundled together?
- Is usage metered, and are there credit limits?
- What access requirements apply?
- Which carrier and yard integrations are included as standard?
Other limits to plan for:
- Data dependency. The agents are only as good as the carrier signals and PO data behind them. Carriers that rarely reply will still leave gaps.
- Autonomy by default. The agents are positioned to act rather than alert. For anything that touches customers or charges, you have to design approval in yourself.
- Privacy. Voice, SMS and WhatsApp outreach produces phone numbers and message logs. Agree retention rules and a lawful basis with your own adviser before you switch these channels on.
- Accuracy. We found no independent data on ETA accuracy for your lanes.
Who should skip FourKites?
Skip it if you ship a handful of parcels through one courier. A shared inbox template and the courier's own tracking will probably do the job. Skip it too if you cannot feed it PO and shipment data, or if no one can own the pilot and approve messages. In that situation the agents would have little to work with and no one to answer to.
FAQs
Does FourKites replace my transport management system? The reviewed pages present FourKites as visibility and orchestration on top of your shipment and order data. They do not position it as a planning or tendering tool. Ask how it connects to the systems you already run.
Can I start with one agent? The pages do not say. Ask early, because a narrow pilot is easier to judge.
What should I prepare before a demo? Bring three anonymised delay cases from last month and the steps your team currently follows. Also bring a list of the messages that must never go out without approval.
Practical next action
For one week, log every delayed load and the minutes it takes. Split the time into carrier chasing, customer replies and dock changes. If the total is large and concentrated on a few lanes, take that log and the brief above to FourKites. Ask for written answers on pricing, integrations and approval controls.
Methodology: this is an AI-assisted draft based only on official FourKites web pages retrieved on 2 November 2026. The scenario, workflow, data and brief are illustrative and do not come from hands-on testing. We deliberately excluded vendor case-study figures. ShiftToAGI is an independent publisher and is not FourKites.