AI Portfolio Manager: Full Market Test and Five Business Reviews

Full-market test · 14 November 2026 · Prices at the 13 November regular-session close

The investment decision

Momentum is the only candidate I’d take into the Monday decision. Momentum is the only finalist that reaches 15% in the USD currency stress at Friday’s close before personal taxes. Its margin is narrow: R40.59 was only R0.31 below the test ceiling, and a 20% dividend haircut would take it below the hurdle. Most important: the base case needs some rerating. With 7% EPS growth and the current multiple unchanged, Momentum earns 13.2% in rand rather than 18.8%.

• South Africa: Momentum leads Absa on earnings quality, profitability and entry valuation. Absa gives more dividend income, but Friday’s quote prices it above the stressed USD ceiling.
• US: Adobe is the closest USD name. Its price is only about 1.4% above the modelled entry ceiling, but the return depends on paid AI features protecting the franchise and on a partial rerating.
• Australia: Sonic beats Worley on the investment case, but at Friday’s prices neither clears 15% in stressed USD. Sonic requires better per-share earnings and an 18× exit valuation; Worley requires stronger cash conversion.

AI-generated paper portfolio. This is the 14 November test edition: zero positions and zero orders. New selections are made on Mondays only.

Portfolio and S&P 500 comparison

The first funded lot in the actual pilot will be US$10,000. VOO gets the same timed USD contribution, with fund distributions reinvested. The matched local total-return index is a separate test of stock selection. Today's valuation scenarios are not benchmark returns.

Market coverage and selection metrics

I applied the size test to every captured inventory row: 4,204 fell below the floor and two lacked usable market caps. These are ticker rows, not unique businesses. The US inventory includes foreign issuers and alternative share classes; the dedicated US value screen narrowed that pool. Funds, duplicate economic exposures, non-operating shells and unreliable quotes are not eligible for a final pick. The Australia and South Africa supplemental valuation pages do not cover every name above the size floor; a ticker that was not underwritten is untested, not rejected.

Sources: US inventory · ASX inventory · JSE inventory · US value screen · ASX valuation cross-check · ASX low-P/E screen · JSE valuation cross-check · ECB FX reference rates. The size test used 13 November euro-reference cross rates, about US$0.679 per A$1 and US$0.059 per R1.

How a cheap stock became a candidate

1. Start with size and tradability. For a US$10,000 paper lot, confirm there is a dated regular-session price and sufficient trading activity. Do not rank a small quoted spread or a thin last trade as an actionable opportunity.
2. Check the earnings denominator. For software, use GAAP earnings and check cash after stock compensation. For banks, compare return on ordinary equity, ordinary book value and credit losses. For insurers, compare operating earnings, solvency and value of new business. For resources, normalise commodity prices and sustaining capex before treating low P/E as cheap.
3. Require a route to higher value per share. Estimate a three-year selling price from earnings growth and a defensible exit multiple. Separate operational growth, dividends and a change in the price investors will pay for earnings. Model all cash flows on the same annual timeline, apply trading costs, and show FX separately.
4. Check the political and macro mechanism. Ask what a policy, regulation or institutional change would do to unit costs, sales, funding or default losses. Change the assumptions when outcomes are visible, not when a plan is announced.

Finalists on the same return basis

¹ For ZAR and AUD, the stress assumes 3% annual depreciation against USD. The 15% local threshold is the original screen; the USD column tests the risk to the dollar-denominated portfolio. Returns include dividends and 0.15% transaction costs each way, before personal taxes. Ceilings are approximate; the Momentum ceiling is rounded down to the cent for execution.

What the base case needs from the market

Holding the entry multiple fixed isolates the dependence on rerating; earnings growth and dividends stay the same. None of the five clears 15% without some rerating. Adobe’s gap is the largest.

What you would actually own

The strongest case is not always the cheapest multiple. Each review looks at where profits come from, what management does with the cash, and what would make those earnings more valuable per share.

1. Momentum Group: insurance earnings at a discount

What it does. Momentum sells protection, retirement savings, investment products and employee benefits. Momentum and Metropolitan serve different customer segments; Guardrisk provides specialist insurance structures, while investment management, health administration and general insurance broaden the earnings base. Its business model connects advisers and employer relationships to products and asset management. Group business overview.

Where the profit comes from. FY26 normalised headline earnings included R1.42bn from Corporate, R1.19bn from Investments, R1.15bn from Metropolitan Life, R1.04bn from Guardrisk and R1.01bn from Retail. Together these five contributed approximately 82% of group earnings. Operating profit was R5.97bn and shareholder investment return R1.09bn. Insurance profitability depends on claims, customer retention and expenses relative to pricing; investment businesses also earn fees on client assets. FY26 segment and operating results.

How it is reinvesting. FY26 strategic capital deployment totalled R875m: R301m in India, R285m in African operations, R137m into Consult’s growth plans, R60m for Zestlife and R28m for broker-facing insurtech Control, among other investments. The African injections supported local solvency; they were not all expansion spending. Total shares repurchased during the financial year cost R1.6bn, including the completed R1bn programme. Capital deployment, page 4.

The commercial rationale is specific: better adviser tools should improve distribution productivity, specialist acquisitions can widen underwriting capabilities, and investment-platform integration can retain more of a customer’s lifetime relationship. My test is whether those investments increase profitable business per adviser and earnings per share, rather than merely adding policies, assets or headcount. Repurchases below economic value can help existing shareholders, but do not build the operating franchise by themselves.

How durable is its position? I see an established distribution advantage rather than an unassailable moat. An adviser relationship, accumulated underwriting experience and a long-lived policy book are costly to replicate. They also provide recurring access to customers whose protection and retirement needs evolve. But independent advisers can recommend competitors, customers can lapse policies and investment mandates can move. The most useful evidence is persistency and profitable new-business growth, not the size of the existing book.

Where it is heading. Management’s Impact strategy reached its R7bn earnings ambition a year early; the next strategy through FY2030 is due within the following twelve months. Metropolitan’s turnaround emphasised digital systems, costs and sales quality; Guardrisk added underwriting capabilities and bolt-on acquisitions. The immediate weakness is new-business economics: sales grew 18%, but value of new business grew only 5%, with a 0.5% margin. Management’s results and outlook.

Investment conclusion. The valuation case rests on modest growth in an already profitable franchise. Stronger new-business margins would support the proposed rerating; growing volumes without better economics would not. Track new-business value, policy retention and solvency alongside EPS.

Political and economic transmission: unemployment of 33.7% and a 7.25% policy rate can pressure recurring contributions and policy retention. Better electricity, logistics and municipal services could strengthen household and business cash flows. The Auditor-General’s finding of minimal improvement, with 151 of 417 national and provincial auditees receiving clean audits, does not justify a governance premium in the base case. Look for sustained improvements in procurement controls and service delivery before raising assumptions. Stats SA Q3 labour market · SARB September decision · Auditor-General.

Model decision: revisit on Monday. The R58.43 sell target uses 7% annual growth in diluted normalised headline EPS, a 45% payout and a 9× exit. With the multiple unchanged, the model earns 13.2% a year in rand; at the assumed rerating it earns 18.8%.

2. Adobe: paid creative workflows versus AI substitution

What the business does

Adobe supplies the tools used to create, manage and distribute digital content. Photoshop, Illustrator and Premiere serve professional creators; Acrobat handles documents, signatures and collaboration; Express targets simpler, faster content creation. Its enterprise products connect customer data, marketing campaigns, commerce and content delivery. The business model is predominantly recurring software subscriptions, with upgrades, additional products and usage creating opportunities to increase spending per customer.

Adobe now reports one operating segment, not separate Digital Media and Digital Experience profits. It discloses subscription revenue for two customer groups. In Q3, Creative & Marketing Professionals generated US$4.65bn, up 13%, and Business Professionals & Consumers US$1.91bn, up 16%. Those are revenue disclosures, not separate profit contributions. Group revenue was US$6.76bn and GAAP operating profit US$2.35bn, a 34.8% operating margin. Q3 results · Reporting structure.

Core profit drivers

Retention is the foundation; paid expansion is the growth engine. A professional workflow can combine authoring, stored assets, collaboration and enterprise controls. My view is that replacing that entire workflow is harder than replacing one image-generation task. Adobe earns attractive margins when it retains those customers and sells more capability without a proportional increase in development and distribution costs.

That operating leverage is not automatic. Q3 gross margin fell to 88.7% from 89.3%, while R&D increased 18% to US$1.29bn, faster than revenue. AI training and hosting sit within its cost base. More AI usage only creates shareholder value if incremental subscription and usage revenue exceeds the associated compute, development and acquisition costs. Q3 filing.

How it is reinvesting

Adobe is spending on both internal development and adjacent capabilities. Nine-month R&D reached US$3.60bn. Its US$1.87bn Semrush acquisition, completed in April, adds search and AI-discovery tools to enterprise marketing. The strategic logic is to connect content production with whether customers actually discover and engage with it, increasing the value of the wider platform. Acquisition announcement · Acquisition accounting.

The November MAX announcements show the product direction: Firefly combines Adobe and partner models with agents that execute multistep creative work. Creative Cloud adds agent-driven workflows and previews access through third-party agents. Adobe also confirmed its acquisition of Topaz Labs, bringing image and video enhancement capabilities. These investments defend the editing and production layer even when customers use someone else's underlying AI model. Firefly expansion · Creative Cloud and Topaz update.

Durable position and direction

My thesis is that the workflow is the defensible asset. Professional familiarity, integrated production tools and enterprise deployment can support retention. The counterargument is equally important: easier AI tools can reduce the need for specialist software seats, and a large free audience does not guarantee paid conversion.

The next phase is a move from selling individual tools toward orchestrating creative and marketing work. Judge it by recurring revenue growth excluding acquisitions, paid conversion, customer retention and operating margins after AI costs, not by feature announcements.

Political and economic transmission: copyright, training-data and commercial-use rules can either favour rights-cleared professional workflows or raise the costs of delivering AI. Stickier inflation can curb customer budgets and compress software multiples. The base assumes no favourable change in copyright law and no rate-driven return to a premium multiple. BLS October CPI · BLS October PPI.

Model decision: watch below US$239.75. The US$365.72 sell target starts at FY26 GAAP guidance of US$18.145 per share and assumes 8% annual growth and a 16× exit. The test does not credit the gap between adjusted and GAAP earnings as free cash.

3. Absa: a bank recovery, not a below-book bargain

What it does. Absa provides personal banking, mortgages, cards, business finance, payments, corporate lending, markets and advisory services. It has 13.4 million customers and banking operations in ten African countries. Its three pan-African divisions are Personal and Private Banking, Business Banking, and Corporate and Investment Banking. Absa business structure.

Where the profit comes from. H1 headline earnings were R6.19bn in Corporate and Investment Banking, R4.11bn in Personal and Private Banking and R2.74bn in Business Banking, before R234m of head-office and other losses. CIB therefore provided about 48% of group headline earnings. Business Banking generated the highest divisional ROE, 24.6%, versus 19.2% in CIB and 15.2% in personal banking. The medium-term group ROE target is 16–19%. Segment results and strategy.

The earnings engine has three moving parts: the spread between lending yields and funding costs; transaction, advisory and trading income; and credit losses. A growing loan book creates value only when its risk-adjusted income exceeds the extra funding, capital and operating costs. In H1, net interest income grew 3% and non-interest income 6%; pricing competition and lower rates in African markets compressed margins. Better customer payment behaviour reduced personal-banking impairments. H1 earnings release.

How it is reinvesting. IT-related spending rose 7% to R8.8bn, targeting digital infrastructure, cybersecurity, cloud, data and AI. The reported total includes ongoing technology costs, not simply new growth investment. Investment priorities. Absa reports that automated SME onboarding cut account-opening time from two days to under 30 minutes, while customer information from 32 systems was consolidated into one platform. Those are concrete improvements to acquisition and servicing, rather than an abstract AI ambition. Technology outcomes.

How durable is its position? My assessment is that the strongest advantage lies in integrated business relationships: operating accounts, collections, payroll, trade finance and currency services can become embedded in a client’s processes. A multinational also benefits from one bank spanning several markets. This gives Absa opportunities to cross-sell and gather deposits, but it does not guarantee pricing power. Margin pressure demonstrates that customers still have alternatives. The investment case is stronger if digital improvements deepen primary-bank relationships and lower servicing costs, not just increase app activity.

Where it is heading. Absa is integrating its pan-African businesses, modernising platforms and allocating resources toward stronger-return activities. Near-term guidance remains about 15% ROE and a 55% payout, with costs expected to grow slightly faster than revenue. Management outlook. That leaves a measurable gap between present profitability and ambition.

Investment conclusion. Absa offers an income-and-efficiency case, not a rapid-growth story. The rerating needs improving cost-to-income, resilient credit quality and returns moving toward management’s target while capital remains strong. Until then, a discount to higher-return banking peers is justified.

Macro link. Reliable municipal services and logistics could make more borrowers creditworthy. No reform premium is included: watch impairments, loan growth and net interest margin, which can move in opposite directions when funding costs rise. SARB September decision.

Model decision: watch. Reconstructed trailing diluted headline EPS is R30.497. At R230.09, the shares trade at 1.10× ordinary book, not the screener’s 0.98×. The R308.74 target uses 6% annual earnings growth, a 55% payout and an 8.5× exit.

4. Sonic Healthcare: lab scale must reach per-share earnings

What the business does

Sonic runs pathology laboratories, diagnostic imaging and smaller clinical-service businesses. Doctors refer patients for tests; Sonic collects samples, processes and interprets them, and bills patients, insurers or public health systems. Its largest profit engine is laboratory medicine, not hospitals or pharmaceutical discovery. The network combines local clinical relationships with regional processing capacity. Sonic business history.

Where the profits come from

FY26 laboratory revenue was A$9.32bn, with A$1.65bn underlying EBITDA at a 17.7% margin. Australian radiology contributed A$1.04bn revenue and A$249m EBITDA at 24.0%. Other activities contributed A$34m EBITDA. Laboratory medicine therefore generated approximately 85% of group underlying EBITDA. Europe supplied 46.5% of operating revenue, Australian/New Zealand laboratories 20.4%, and US laboratories 18.9%. Appendix 4E, revenue and segment tables.

The economic engine is tests per laboratory, revenue per test and the cost of delivering each result. More samples can spread facilities, equipment and specialist staff across greater volume. Higher-value genetics and specialist diagnostics can improve the mix. My interpretation is that local network density matters more than simply operating in more countries: a distant acquisition creates little advantage unless its referrals, purchasing or processing can be integrated.

How it is reinvesting

FY26 cash investment included A$631m in property, plant and equipment, A$105m in intangibles and A$616m in acquisitions. LADR also required A$371m of newly issued shares. These commitments are substantial relative to A$1.41bn operating cash flow and A$534m dividends paid. Cash-flow and equity statements.

The reinvestment has identifiable uses. LADR expands Germany’s network and logistics capability; Germany is consolidating 11 operating members into three divisions. A Melbourne hub laboratory is due in mid-2027. In the US, more than 70% of dermatopathology volume runs through the proprietary PathologyWatch platform, helping distribute specialist work. Ten radiology sites opened over two years; management expects them to become accretive about 18 months after opening. FY26 annual report, CEO review.

Durability and direction

I see a defensible service network rather than unlimited pricing power. Clinical reputation, referral relationships and accredited processing capacity are difficult to recreate quickly. Governments and insurers, however, can change reimbursement faster than Sonic can reduce its cost base.

That tension is visible now. US operating initiatives are expected to add A$25–30m to FY27 earnings, while Swiss fee reductions are expected to remove CHF20m of annual revenue. The UK Hertfordshire and West Essex contract’s intended margins have moved to the second half of FY28 after delayed testing transfers. Annual report, regional review.

My assessment: Sonic has durable demand, but must earn better returns on the infrastructure and acquisitions already funded. The key evidence is higher EPS and cash generation after interest, leases and investment, not another acquisition-led revenue increase. That supports a watchlist position, rather than paying today for every promised efficiency.

Macro link. The RBA’s November cash rate of 4.85% makes the economics of capital allocation more important. Reimbursement is the larger direct political risk: fee restraint can absorb the benefit of growing test volumes.

Model decision: watch. The A$26.93 target assumes 6% annual EPS growth, a flat A$1.08 annual dividend and an 18× exit. Sonic clears 15% locally by a narrow margin but earns 12.1% in the USD stress.

5. Worley: a growing backlog is not yet growing cash

What the business does

Worley helps customers plan, design, procure, construct and commission complex energy, chemicals and resources assets. It sells engineering and delivery expertise rather than owning the oilfield, power plant or mine. Work can start with feasibility advice and extend into execution and operations. This gives it several opportunities to earn fees during an asset’s life. Worley’s delivery model.

Where the profits come from

FY26 energy supplied 53% of aggregated revenue, resources 27%, and chemicals 20%. Energy revenue grew 8% and resources 6%; chemicals fell 22% following European project cancellations. The Americas supplied 52% of revenue. Group underlying EBITA was A$734m, on A$12.02bn aggregated revenue. Its 6.1% margin rises to 9.0% when procurement revenue is excluded. These are revenue shares, not sector profit shares. FY26 results.

Profit depends on winning work at sensible rates, keeping skilled teams productively employed and controlling delivery costs. Procurement and construction can increase the contract value without adding proportionate margin. A larger sales figure can therefore conceal weaker economics. The accounting bridge also matters: FY26 underlying profit after acquired-intangible amortisation was A$327m, versus A$395m before it and A$238m statutory net profit. Appendix 4E.

How it is reinvesting

The May investor presentation earmarked approximately A$70m over two years for capabilities in integrated gas, nuclear, power, data centres, energy-transition materials and digital tools. Investment includes specialist hiring, delivery hubs, technology partnerships and early consulting work intended to lead to larger delivery mandates. Its AI tools include construction sequencing and technical/commercial supplier-bid evaluation. The strategy is to capture more of each customer’s project spending, not just sell more engineering hours. Investor Day, pages 40–55.

Meanwhile, FY26 cost initiatives produced A$132m savings, partly supporting reinvestment. Capital returns remain material: a A$500m buyback completed in May and another A$300m program began. FY26 results. My test is whether internal investment improves cash earnings faster than buybacks improve EPS mechanically.

Durability and direction

Worley’s strongest advantages are accumulated engineering knowledge, customer familiarity and the ability to staff complex international programs. Management reported more than 175 portfolio contracts and predominantly reimbursable work exceeding 75% in its investor presentation. Reimbursable contracts reduce exposure to construction-cost escalation relative to fixed-price turnkey work; they do not remove project cancellations or staff-utilisation risk. Investor Day, pages 61–62.

Bookings rose 23% to A$15.5bn, with major US gas and power work supporting the pipeline. Yet Middle East disruption cost A$58m in FY26 EBITA. FY26 results. This is exposure to customers’ investment decisions, not a guaranteed beneficiary of every energy-security announcement.

My assessment: the opportunity is real, but conversion is the missing proof. Operating cash flow fell to A$401m from A$741m. Appendix 4E. I would want backlog converting into cash, fewer recurring restructuring charges and better statutory earnings before treating the low adjusted multiple as durable value.

Model decision: watch. The A$13.41 target assumes 7% underlying EPS growth, A$0.50 annual dividends and a 14× exit. That returns 13.9% in AUD, even on an earnings base before acquired-intangible amortisation.

Scenario book

All rows use the same convention: annual dividends, 0.15% transaction costs each way, and local-currency IRR before personal taxes. “Nothing improves” holds EPS and the multiple flat; it is not the bear case.

Momentum uses a 45% payout of each year’s modelled EPS; Absa uses 55%. Adobe has no dividend in the model. Sonic’s annual dividend assumptions are A$1.20 bull, A$1.08 base and flat, A$0.90 bear. Worley’s are A$0.55 bull, A$0.50 base and flat, A$0.35 bear.

Where the numbers can change the decision

Momentum sensitivity at the R40.59 reference price

At an unchanged multiple, the local return drops to 13.2% even if EPS grows as modelled. At the 9× multiple, changing the assumed rand decline from 3% to 5% is enough to miss the 15% USD test. The Monday decision is whether that margin is worth using a slot, not whether the stock looks cheap on P/E.

What to research next

PayPal is a useful contrast with Adobe: Q3 results showed revenue growing 2%, operating profit down 8% and free cash flow down 17%. A simple 5% EPS-growth, 12× exit case gives a US$70.43 target from US$59.28, roughly 6% annualised price return before dividends and fees. It did not displace Adobe.

Model and execution rules

• Sell target: starting EPS × (1 + annual EPS growth)³ × exit multiple. Momentum: R5.30 × 1.07³ × 9 = R58.43. The base dividends are approximately R2.55, R2.73 and R2.92 per share.
• Return and entry: the initial cash outflow includes the purchase fee. Sale proceeds net of the exit fee and three annual dividends determine IRR. For non-USD stocks, each cash flow is converted at the corresponding modelled FX rate. The ceiling discounts those cash flows at 15%; it is not the sale target.
• Monday only: after rechecking the market, lock any selected target and entry ceiling before the next full regular session. Fund US$10,000 of simulated capital per filled pick using that session’s close if within the ceiling. No retroactive buy at Friday’s price.
• Daily updates: report position and benchmark changes after verified sessions. If an eligible regular-session high goes above a locked target after entry, book the paper sale at the target for that session. Pre-entry and after-hours highs do not count.
• Benchmark: allocate the same timed USD contributions to VOO, include reinvested distributions once, and show invested capital, net dollar return, ROI and the percentage-point spread. Use time-weighted returns where staggered contributions make a simple ROI comparison misleading. Keep local total-return indices as a separate stock-selection check.

Appendix | Named candidates and decisions

The 46 reviews below range from valuation triage to full modelling. A name without a target is either deferred or screened out for the stated reason, not assigned a hypothetical target.

United States · 11 names

All prices below are 13 November 2026 regular-session closes, 16:00 EST, not after-hours quotes. Multiples are the provider's current trailing or consensus-forward figures and were used to triage, not substituted for company guidance.

These are decisions about this test, not claims that every nonselected business is overvalued. “Not captured” means the field was not used, not zero.

Australia · 14 names

Prices below are AUD at 13 November 2026 unless stated otherwise. Forward multiples are the data provider's estimates, used for triage rather than as independently verified forecasts.

Sources: company-specific statistics pages SHL, WOR, CSL, CBA, NAB, QBE, WDS, FMG, WTC, MND, SCG, REA, HLS. Prices also appear in the inventory. SHL and WOR underlying multiples are calculated from the primary EPS disclosures below.

South Africa · 21 names

The ratios below are provider snapshots retrieved on 14 November, not a perfectly simultaneous closing dataset. Finalist valuations are recalculated from dated closes and company filings.