Tuaha Jawaid
Tuaha runs discovery, the audit and delivery at Kastling. His background is consumer goods marketing and sales on the agency side, working on brands including Unilever, Red Bull and Toyota. He spends most of his time with operations leaders, mapping how work runs today before anyone proposes software.
LinkedInWritten by Tuaha
AI adoption statistics for 2026: what the data says for mid-size businessesUS employers with 250 or more employees report AI use of about 37%, against under 20% at those with four or fewer, and overall firm-level use was 17% to 20% between December 2025 and May 2026 (US Census Bureau Business Trends and Outlook Survey). Other surveys report higher figures, such as 88% of surveyed organizations in 2025 or 62% of US adults using generative AI, because each one measures a different population and definition of use. AI agent deployment was still in the single digits across nearly all business functions in 2025, and each figure links to its source, rechecked on 3 October 2026.Read AI agent development services: what you are actually buyingAI agent development services should deliver one defined workflow that works inside your existing systems, with limited access, a person approving consequential actions, testing on your own past cases, monitoring, and a handover you can run yourself. The model is the smallest part of the bill. Scope, integrations, evaluation and operations decide the cost, and a proposal that skips any of them is selling a demo.Read AI workflow automation: what it costs to run and what it savesAI workflow automation cost comes from five areas: discovery or audit scope, implementation, model usage and infrastructure, integration licenses, and support and change. Model usage is often a smaller recurring line than the licenses around it, while support, exception handling and adoption are the lines buyers underestimate. This guide for companies of roughly 20 to 200 people explains each area, and savings count as cash only when they change hiring, overtime or outside spend.Read AI automation handover checklist and runbook templateAn AI automation handover should leave your team able to operate, check, pause and recover the workflow without relying on its builder. Use this checklist to request the evidence behind each responsibility, then copy the runbook template to document one workflow before accepting it.Read How to calculate the ROI of AI automationCalculate the ROI of AI automation by measuring the current process first, then valuing hours saved at loaded cost, errors avoided and faster cycle times, and setting that against the full cost to build, run and support the system. Payback in months equals one-time costs divided by monthly net benefit, and ROI over a period equals total benefit minus total cost, divided by total cost. Express the result under conservative assumptions, and count saved hours as money only when they change hiring, overtime or contractor spend.Read How to write an RFP for an AI implementation projectAn AI implementation RFP needs nine sections: the business problem, the workflow, systems and data, constraints, the response format, weighted criteria, timeline, commercial format and submission rules. This guide for companies of roughly 20 to 200 people includes a copyable template, the questions to put to every bidder and a weighted scoring table. An RFP is the wrong tool when you do not yet understand your own workflow, so a requirements brief and a discovery call come first.Read AI integration services: what they include and how to hire a providerAI integration services connect AI to the systems your business already runs on, such as your CRM, ERP, email and document stores, and build it into a defined workflow with people kept in the decisions. A typical engagement moves from a free discovery call to a separately scoped paid audit, a proposal, development, integration and training, and optional maintenance. Scope, system access and data quality drive cost far more than company size, and some projects do not need a provider at all.Read AI proof of concept vs pilot vs MVP: which first step fits your project?Use a proof of concept to test whether AI can do the task on your data, and a pilot to test whether people can use it in live work. An MVP is the first version people rely on every day, and for AI automation a pilot that passes and stays on often becomes that release. This guide is for operations and IT leaders at companies of roughly 20 to 200 people choosing a first step.Read What happens in an AI readiness assessmentAn AI readiness assessment (often called an AI readiness audit) is a structured review of specific workflows, the data and systems behind them and the people who run them, to decide where AI or automation will help and what must be fixed first. A good one ends with named, prioritized opportunities and a roadmap, not a maturity score or a deck of industry trends. This guide is written for companies of roughly 20 to 200 people.Read What drives the cost of an AI readiness auditAn AI readiness audit (also called an AI readiness assessment) has no fixed price because cost follows scope: how many workflows and teams are reviewed, how many systems and data sources are checked, and how deep the security review goes. A discovery conversation is normally free, while the audit is a paid, separately scoped engagement, so any single published price is a guess. This guide for companies of roughly 20 to 200 people explains the drivers and how to compare proposals.Read AI workflow automation: a practical guide for operations teamsAI workflow automation puts AI models inside a defined business process to read, classify and prepare work, while fixed rules, integrations and named people handle the steps that must be predictable. It suits repetitive, document-heavy or request-heavy work such as intake, routing, document checks and approvals. This guide is for operations leaders at companies of roughly 20 to 200 people, mainly in the US and Canada: start with one measurable workflow, keep consequential approvals with an accountable person, and expand only once the numbers show it works.Read Build, buy or integrate: deciding on custom business softwareBuy software for standard processes, integrate when your existing systems already hold the data, and build only where the process sets you apart. Companies of 20 to 200 people usually end up with a hybrid: buy the core system of record, then integrate or build the edges around it. Compare the options on 3 to 5 year total cost of ownership, not on the first invoice.Read How to choose an AI implementation partner: 10 questions to askChoose an AI implementation partner by asking each shortlisted firm the same ten questions and favoring the one that studies your workflow first and puts ownership in the contract. The questions cover scoping, testing, human control, ownership, lock-in, security, integration feasibility, who does the work, how success is measured and what happens after launch. Then verify with reference calls and a small, paid first piece of work before committing to a large build.Read What factors affect the cost of custom software developmentScope sets the cost of custom software: how many user roles it serves, how many systems it connects to, how much data must migrate and which security rules apply. Technology choices and hourly rates matter far less than most buyers expect. Budget for the years after launch as well, because hosting, support and maintenance continue for as long as the business depends on the software.Read Fixed price vs time and materials vs retainer: how to pay for an AI projectUse a fixed fee when scope is settled and testable, time and materials with a cap while work is uncertain, and a retainer for support after launch. Most AI projects combine all three in sequence. This guide is written for buyers at companies of roughly 20 to 200 people.Read Automating FMCG order entry and validation before the ERPTo automate order entry for FMCG, read each distributor or retailer order into standard fields, then check price, trade terms, promotions and credit before it reaches the ERP. Clean orders pass straight through, while exceptions go to the sales operations owner with the evidence attached. Credit, pricing and promotion decisions stay with a named person, in companies of roughly 20 to 200 people as much as in larger ones.Read Retailer deduction management: matching short payments to proofDeduction management is identifying, validating and resolving the amounts a retailer subtracts from a supplier’s payment by matching each one to the invoice, proof of delivery or promotion. Automation can read remittances and retailer backup, match each deduction to those records and route the unsupported ones to the right owner before the dispute window closes. Whether to dispute, write off or settle, and the conversation with the retailer, stay with people.Read Trade promotion approval process: how to speed it up with automationA trade promotion approval moves a brief through mechanics, funding, price and legal checks to a named approver before it reaches a retailer. It typically stalls on incomplete briefs, numbers that do not match the plan, and approvers waiting on each other by email. Automation can check the brief, validate the numbers, route it to the right approver and record the decision, while the commercial call and any exception to price rules stay with a person.Read Internal tools and customer portals: planning the first releaseThe first release of an internal tool or customer portal should cover one workflow for one group of users, end to end, rather than a thin version of everything. Decide roles, permissions and how people sign in before design starts, because those choices are expensive to retrofit. Then measure adoption against the work actually moving through the tool, not against logins.Read Carrier vetting checklist: verify authority, insurance and identityCarrier vetting means verifying, before the first load, that a US trucking company is who it says it is, holds the authority and insurance it claims and is paid correctly. AI can read the carrier packet, certificate of insurance and W-9, compare them with the FMCSA record, and send each mismatch to the person who owns it. Approving a carrier, judging safety history and any fraud determination stay with named people.Read Delivery exceptions: what AI can automate and what stays with peopleDelivery exception management can be partly automated: AI classifies the exception, assembles order context from the TMS, WMS and ERP, drafts the customer update and routes the case. Claims decisions, credits, refunds and carrier disputes stay with a named person who reviews a prepared case instead of building one from scratch. The workflow works from carrier tracking feeds, EDI 214 messages and customer emails.Read How to audit freight invoices: carrier and accessorial chargesA freight invoice audit checks each carrier invoice, line by line, against the rate confirmation or contract, the bill of lading and the proof of delivery before payment is approved. In the US, a shipper has 180 days from receipt of a motor carrier bill to contest it under 49 U.S.C. 13710, so the clock matters as much as the check. AI can read the invoice, line up the linehaul, fuel and accessorial charges against the agreement, and flag every difference with its evidence, while whether to dispute, short pay or accept stays with named people.Read Commercial invoice vs packing list vs bill of lading: what must matchShipping document verification compares the commercial invoice, packing list, bill of lading or air waybill and purchase order for one shipment, field by field, before the freight moves. AI can read the documents, line up the fields and route each mismatch to the operator who owns that shipment. Customs classification, valuation and the filing itself stay with a licensed broker and your own people, and the customs references in this guide are for US imports.Read Off-the-shelf vs no-code vs custom software: when to use eachBuy off-the-shelf software for standard processes, use no-code or low-code for team tools, and build custom software when the process is specific to you or outgrows a platform. Govern every no-code app like any other business system, and test data volume and integration depth before committing, because that is where platforms stop. Companies of roughly 20 to 200 people usually end up using all three.Read Client onboarding automation for accounting, law and consulting firmsClient onboarding automation combines a single intake form, document checklists and automated field checks with a person who still makes every acceptance, conflict and risk decision. Done well, it replaces chasing missing documents and re-keying the same details into three systems with one record that is ready for a reviewer to approve. The result shows up as fewer touches per client and a shorter gap between a signed engagement and billable work.Read How to run a conflict check at a law firm: what to search, who decides and what software can doA conflict check is a search of a law firm’s records, run before a matter opens, for relationships that could stop or limit the firm from taking the work. Software and AI can build the name list, search every system at once, rank the hits and route each one to the right reviewer. Whether a conflict exists, whether it can be waived and whether to accept the work stay with lawyers.Read Client document collection for accounting, legal and advisory firms: the best way from request to ready fileThe best way to collect documents from clients is one secure portal, a request list built for that client, automatic matching of each upload and specific reminders. Client document collection automation does that matching, checking and chasing and hands the professional a ready file, while the professional reviews the documents, decides what they mean and owns the advice.Read Real estate lead qualification with AI: faster replies, fair criteria, the right agentReal estate lead qualification with AI reads each inquiry, matches it to a property and an agent, asks a few fixed questions and routes it within minutes. It qualifies only on budget, timing, property fit and financing status, never on protected characteristics or their proxies, and the agent owns the relationship, the advice and every decision about the person.Read Lease abstraction: pulling dates, rent steps and clauses into a lease registerLease abstraction means reading a lease and every amendment and recording its dates, rent schedule, expense terms and key clauses in a lease register, with the source of each value. AI can read the PDFs, propose the values, run the consistency checks and route each gap to the lease administrator, accountant or counsel who owns it. Interpreting ambiguous clauses, judging whether an option is reasonably certain to be exercised and exercising options stay with people.Read Rental application processing automation: document checks, screening and what stays with the leasing managerRental application processing automation reads each uploaded document against the application, flags what is missing or inconsistent and prepares a complete file, while a person approves or denies. Done well, it cuts the back and forth with applicants and the number of people who touch one file before a decision is made. This guide is for US property management companies, and the payoff shows up as a shorter time from application to a decision ready file.Read Product data validation checklist for ecommerce catalogs: checking supplier data before it goes liveProduct data validation means checking every supplier feed, spreadsheet and sell sheet against your catalog rules before an item reaches the ecommerce platform or PIM. AI can read messy supplier files, test attributes, GTINs, images, dimensions, compliance fields and pricing, and propose fixes. A merchandiser still approves anything that changes how a product is described, priced or classified.Read Retail order exceptions: how to handle and automate them without losing the customerOrder exception management means catching a payment hold, stockout, address failure or pricing mismatch as it happens and routing a prepared fix to the person who owns the decision. Automation gathers the order, stock and customer context and drafts the fix, while refunds above a limit, fraud declines and goodwill credits stay with a person who reviews a prepared case rather than searching five systems for the facts.Read How to write a software requirements brief (with a template) a development partner can useA software requirements brief is a two to four page decisions document stating the problem, the users, what the first release must do and how success is measured. It also covers how the process runs today, the systems and data involved, the constraints and who decides, and it leaves design to later. Written that way for a company of roughly 20 to 200 people, it produces comparable quotes, a realistic plan and far less rework.Read In-house vs agency vs AI partner vs freelancer: how to staff an AI projectFor a first AI workflow, a specialist partner with named internal owners is usually the fastest route, and in-house hiring makes sense once steady work can keep a team busy. An AI project also needs a business owner and a process owner from inside your company, plus AI engineering, integration, data, cloud and change skills that you can hire, contract or bring in through a partner. Whichever route you take, your company should own the code, the accounts and the documentation.Read Telecom order fallout and order to activation: checking orders before they reach provisioningTelecom order fallout is an order that stops moving between signature and activation and needs manual work, and much of it traces back to an incomplete order from sales. Order to activation automation reads each signed business order, checks site details, number porting data, credit status and serviceability, and sends only complete orders to provisioning, returning the rest to sales with a specific request. Exceptions, design choices and customer commitments stay with a provisioning lead.Read AI ticket routing for telecom support: the right queue on the first tryAI ticket routing reads a support request, extracts the account, service and location, checks known outages and service levels, and sends it to the right queue on the first try. This guide is for telecom and ISP support teams of roughly 20 to 200 agents. Routing does not decide refunds, retention offers or complaints, which stay with a person, and done well it cuts the misroutes and transfers a customer sits through before someone can help.Read Three-way match: automating PO, goods receipt and invoice matching with tolerancesThree-way match compares the purchase order, the goods receipt and the supplier invoice before payment, so a company pays only for what it ordered and received at the agreed price. Automation can read each document, apply price and quantity tolerances and send every exception to the person who owns it. Approving payment above tolerance, accepting a price variance and changing supplier bank details stay with named people.Read Which business processes to automate firstThe best first process to automate is a frequent, well understood, medium consequence process with one accountable owner, not the one people complain about most. Choose it by scoring three to five candidates on volume, repetition, judgment, data availability, systems access, cost of errors, ownership and measurability, then scope the highest as a single workflow with a recorded baseline. This guide is for operations leaders at companies of roughly 20 to 200 people.Read
See how an idea becomes a working solution.
Explore how we scope, build and hand over AI projects, with clear decisions at each stage.